Modeling of cardiovascular autonomic control in sickle cell disease

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content MODELING OF CARDIOVASCULAR AUTONOMIC CONTROL IN SICKLE CELL DISEASE by Suvimol Sangkatumvong A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (BIOMEDICAL ENGINEERING) May 2011 Copyright 2011 Suvimol Sangkatumvong ii Dedication To my beloved parents iii Acknowledgements Throughout my years in the PhD program, I owed my gratitude to great many people. Without whom this dissertation would not have been possible and my graduate school experience would not have been as memorable. My deepest gratitude is to my advisor, Dr. Michael C.K. Khoo. His advice has guided me through my PhD progress. I felt truly fortunate to have a very supportive advisor that allows me the freedom to explore different ideas while always be there to guide me to the right path. I would also like to express my great gratitude to my co-advisor, Dr. Thomas D. Coates. I have learned so much from your vast knowledge. His passion has inspired me to continue to do research. I am also very thankful for the tremendous amount of time his have spent on teaching me not only scientific and medical stuffs, but also hours and hours that he spent training me to give a talk and revising my writings. I would like to thank Drs. John Wood, Herbert Meiselman, and David D’Argenio for your guidance during my qualifying exam and beyond. I am grateful for all valuable comments and suggestions which Dr. Wood and Dr. Meiselman have shared during our weekly research group meetings. I am deeply grateful to both for many advices that helped me both on the technical and medical aspects of my work. The completion of my research work would not have been possible if not for the contributions from the entire sickle cell research team at CHLA and USC. Drs. Thomas Coates (my co-advisor), Roberta Kato, and Jon Detterich have spent numerous times iv setting up the experiments and working with subjects to collect those precious data. I would also like to thank Adam Bush for his devoted assistance in experiment set up and data analysis. This study would not be accomplished in this timely manner without his help. Also many thanks to Rose Wenby, Tamas Alexy, Miklos Rabai, and everyone who has assisted and given inputs to the project. I would like to acknowledge all the support personal in the research team at the hospital. The study organization from Tatiana Hernandez, Ani Dongylian, and Anne Nord has been instrumental for this project. I am also thankful to Sierra Betinis whose help allowed me to work efficiently and never had to worry about getting reimbursement. I also appreciate the financial support from the NIH that funded the study discussed in this dissertation. I am grateful to Dr. Sharon Myers, Dr. James Polk, and Mary Ann Murphy for reading and commenting on the writing aspects of my dissertation. I am grateful to the USC Biomedical Engineering staff for their various assistances during my study; special thank to Mischal Diasanta who has always been there to ensure I completed all the requirements for my PhD goal. I am also thankful to all my lab-mates at the Cardiorespiratory Sleep Lab whom friendships made my years of working at USC enjoyable. Their inputs and many valuable discussions helped me better understand my research work; my sincere thanks to Olga Ivanova, Jarree Chaicharn, Suradej Tretriluxana, Limei Cheng, Wenli Wang, Winston Tran, Flavia Oliveira, Patjanaporn ‘Sang’ Chalacheva, Nestor Cabrera Munoz, and Jasmine Thum. v I am thankful to many good friends who helped me through ups and downs during my years at USC. Their support gave me joy and helped me stay focused on my graduate study; my sincere gratitude to Peggy Hong, Kenny Kuwahara, Rosa Chan, Dan Song, Dalad Nattwongasem, Pom Vivitvorakit, Narawut Pakaprot, Piriya Prathuangwong, and my cousin Waiyawuth Eudchongpasit. I am also grateful to the Thai student group at USC for their support and friendship throughout the years. I am heartily grateful to my lovely husband, Nuntachai Art Poobuapheun. Throughout the years he has given me his love, trust, and the supports required for me to overcome setbacks on my study. I am very thankful to have him by my side and brighten up my days, and to have his shoulder to lean on through some of the tough times. Finally and most importantly, I would like to thank my dearest family for their ever-lasting, unconditional love and support. I am also appreciated the generosity and support from my extended family, especially my grandparents whose kindness and understanding have warmed up my heart. I am very thankful to my brothers, Apinant, Peeranat, and Supakit; their love and trust have always lifted up my spirit. Their updates of things at home helped me stay connected and less homesick. Greatest thank to my mom and dad, Julai and Chanin Sangkatumvong, whom this dissertation is dedicated to. Not only have they been a constant source of love and support which strengthen and guiding me forward through all steps in my life, they have also always believed in me and encouraged me to pursue my dream leading to the finishing of this dissertation. vi Table of Contents Dedication .......................................................................................................................... ii Acknowledgements .......................................................................................................... iii List of Tables .................................................................................................................. viii List of Figures ................................................................................................................... ix Abbreviations ................................................................................................................. xiv Abstract .......................................................................................................................... xvii Chapter 1. Introduction ................................................................................................... 1 1.1. Sickle cell disease (SCD) ................................................................................... 1 1.2. Cardiovascular autonomic control ...................................................................... 4 1.3. Sickle Cell Disease and Autonomic Dysfunction ............................................ 10 1.4. Background on Autonomic Control Stimuli used in this study ........................ 14 1.5. Computational Techniques to Characterize the Cardiovascular Autonomic Control .............................................................................................................. 19 1.6. Aims of the Study ............................................................................................. 22 1.7. Significance of the Study ................................................................................. 24 1.8. Organization of the Dissertation ....................................................................... 24 Chapter 2. Experiments and Data Processing .............................................................. 26 2.1. Subjects ............................................................................................................ 26 2.2. Measurements ................................................................................................... 32 2.3. Experimental Procedures .................................................................................. 35 2.4. Data Pre-processing .......................................................................................... 42 2.5. Statistical Analyses ........................................................................................... 47 Chapter 3. Time-varying Analysis of Heart Rate Variability .................................... 51 3.1. Introduction ...................................................................................................... 51 3.2. Recursive Autoregressive Technique for Analysis of HRV ............................ 52 3.3. Recursive Least-squares Parameter Estimations .............................................. 54 3.4. Time-varying Parameter Estimate Variability Reduction (TV-PEVR) Technique ......................................................................................................... 62 3.5. Adjustment for the Effect of Respiration on HRV ........................................... 64 Chapter 4. Minimal Model for Assessing Cardiovascular Control ............................ 71 4.1. Introduction ...................................................................................................... 71 4.2. Model Description ............................................................................................ 72 4.3. Physiological Information Extracted from the Model ...................................... 75 4.4. Orthogonal Expansion of Impulse Responses .................................................. 77 4.5. Linear Time-invariant Model Parameter Estimation ........................................ 89 vii 4.6. Linear Time-varying Model Parameter Estimation .......................................... 95 Chapter 5. Results and Discussion .............................................................................. 102 5.1. Baseline Physiological and Autonomic Measurements in SCD patients ....... 103 5.2. Autonomic Nervous System Response to Hypoxia in SCD Patients ............. 106 5.3. Autonomic Nervous System Response to Sighs in SCD Patients .................. 119 5.4. Autonomic Control Changes with Transfusion Protocol ............................... 130 5.5. Autonomic Nervous System Response to Cold Face Test ............................. 134 5.6. Evaluation of Minimal Model of HRV Control ............................................. 149 5.7. Relationship between HRV and Laboratory Blood Test ................................ 168 Chapter 6. Future Research Opportunities ................................................................ 177 6.1. Closed-loop Minimal Model of Heart Rate Control ...................................... 177 6.2. Minimal Model of Peripheral Perfusion Control ........................................... 178 6.3. Non-linear Model of Vascular Resistance ...................................................... 186 Chapter 7. Conclusion .................................................................................................. 191 References ...................................................................................................................... 197 viii List of Tables Table 2.1 Characteristic of the subjects in the controlled hypoxia trial ..................... 28 Table 2.2 Characteristic of the subjects in the pre- and post transfusion trial ............ 29 Table 2.3 Characteristics of the subjects in the CFT trial .......................................... 30 Table 2.4 Data used for a power analysis to determine the number of subjects for this study. .............................................................................................. 31 Table 2.5 List of measurements acquired during experiments and their sampling frequencies .................................................................................................. 34 Table 2.6 Example of 2x2 contingency table presenting a relationship between occurrences of sighs and PU drops in one subject. .................................... 49 Table 5.1 Physiological measurements during 5-minute baseline ........................... 105 Table 5.2 HRV measurements during 5-minute baseline ......................................... 106 Table 5.3 Safety data from 19 CTL experiments and 33 SCD experiment .............. 108 Table 5.4 Comparison of SaO 2 and PO 2 drops after hypoxia exposure. The values are shown in mean (SD) ................................................................ 111 Table 5.5 Differences in baseline parameters related to the autonomic control system (excluding HRV parameters) pre- and post-transfusion. ............. 132 Table 5.6 Differences in baseline HRV parameters between pre- and post-transfusion. ....................................................................................... 133 Table 5.7 Statistics of the descriptors extracted from impulse responses and transfer functions of ABR and RCC from sickle cell patients pre- and post-transfusion. ................................................................................ 154 Table 5.8 p-values from pairwise comparisons of the changes in the ABR and RCC descriptors in response to the CFT .................................................. 165 Table 5.9 Blood test data and HRV parameters in response to hypoxia that used to generate scatter plots ............................................................................ 169 Table 5.10 Summary of p-value from the correlation results 11 lab parameters vs. 6 HRV parameters in response to hypoxia. ........................................ 170 Table 5.11 Scatter plots between HRV responses to hypoxia and lab data ............... 171 Table 5.12 Summary of p-value from the partial correlation results RBC and WBC counts vs. 6 HRV parameters in response to hypoxia .................... 173 ix List of Figures Figure 1.1 Normal RBCs and sickle cells ...................................................................... 2 Figure 1.2 Representation of the autonomic control of the cardiovascular system ....... 4 Figure 1.3 Reflexes of the ANS ..................................................................................... 7 Figure 1.4 Comparison of percentage of abnormalities in test results among 3 subject groups: SCD patients, normal whites of Hispanic origin, and normal blacks. ............................................................................................ 12 Figure 1.5 Relationship between sickle cell disease severity and RSA Score............. 13 Figure 1.6 A classic example of spectral analysis of heart rate variability (HRV) ..... 21 Figure 2.1 Total sample size as a function of power computed from the preliminary values of differences of RRI, HFP ra , and LHR ra between SCD and CTL ............................................................................................. 32 Figure 2.2 Experimental setup ..................................................................................... 34 Figure 2.3 An example of measurements recorded during a hypoxia episode in an SCD patient ............................................................................................ 37 Figure 2.4 An example of data from the deep breath protocol. ................................... 39 Figure 2.5 An example of data from the FMD protocol. ............................................. 40 Figure 2.6 Experimental setup for the cold face test protocol. .................................... 41 Figure 2.7 An example of data from the deep breath protocol. ................................... 42 Figure 2.8 Oxygen dissociation curves ........................................................................ 45 Figure 2.9 Examples of data sections which have been manually removed and interpolated ................................................................................................. 47 Figure 3.1 Sample of PSD and its corresponding RRI from our hypoxia study .......... 54 Figure 3.2 Normalized mean squared error (NMSE) of RRI estimates from the AR model with various model orders and forgetting factors ..................... 58 Figure 3.3 Time-varying spectral analysis of a simulated signal ................................ 59 Figure 3.4 An example of spectral estimations of r-wave to r-wave interval (RRI) from 1-minute section before sigh and from 1-minute section after sigh. .................................................................................................... 61 Figure 3.5 Diagram showing the processing steps for the TV-PEVR method ............ 63 Figure 3.6 High-frequency power (HFP) and low-frequency to high-frequency ratio (LHR) of the simulated signal from Figure 3.3 ................................. 64 x Figure 3.7 Fluctuation in airflow with the corresponding fluctuation in RRI. ............ 65 Figure 3.8 Changes in minute ventilation from the baseline value as a response to hypoxia in 8 SCD patients and 8 normal controls .................................. 66 Figure 3.9 Flowchart representing the algorithm to adjust for the respiration effect to HRV. ............................................................................................ 66 Figure 4.1 Diagram of the minimum model describing the fluctuation of RRI .......... 73 Figure 4.2 An example of the parameters extracted from an impulse response (h ABR or h RCC ). ............................................................................................. 75 Figure 4.3 An example of the parameters extracted from a transfer function (ABR or RCC).. .......................................................................................... 76 Figure 4.4 Diagram representing the impulse response estimation method using weighted kernels. ........................................................................................ 78 Figure 4.5 The first four discrete-time Laguerre functions from the definition in (Marmarelis, 2004) for  = 0.4. System length, M = 50 ............................ 82 Figure 4.6 The first four discrete-time Laguerre functions from the definition in (Brinker, 1995) for  = 0.4 or p = = 0.6325. System length, M = 50 ..... 83 Figure 4.7 The cascade filter structure to generate Laguerre functions ....................... 84 Figure 4.8 The cascade filter structure to generate Meixner functions (Adopted from Brinker, 1995 with modification). ..................................................... 85 Figure 4.9 The first four discrete-time Meixner-like functions from the definition in (Brinker, 1995) for  = 0.4 or p = = 0.6325. “Order of generalization” n = 8. System length, M = 50 ............................................ 87 Figure 4.10 Procedure to get the estimation of h RCC and h ABR ....................................... 90 Figure 4.11 Respiratory-cardiac coupling impulse response (h RCC ) and arterial baroreflex impulse response (h ABR ) used for model testing ........................ 94 Figure 4.12 ABR and RCC impulse calculated from all 3 steps as detailed above. A simulated signal generated from real ∆Resp and ∆SBP data, and the impulse responses shown in Figure 4.11. ............................................. 95 Figure 4.13 Arterial baroreflex (ABR) impulse response and respiratory-cardiac coupling (RCC) impulse response used for testing of the time-varying minimal model. ........................................................................................... 97 Figure 4.14 Diagram describing test signal generating steps ........................................ 98 Figure 4.15 Time-varying arterial baroreflex (ABR) impulse response ....................... 99 Figure 4.16 An example of time-varying impulse responses estimation from a simulated dataset ...................................................................................... 100 xi Figure 4.17 An example of time-varying impulse responses estimation from a simulated dataset ...................................................................................... 101 Figure 5.1 Blood oxygen, perfusion, and heart rate responses to hypoxia stimulus in SCD and in CTL. ................................................................... 110 Figure 5.2 Time courses of heart rate variability (HRV) indices of SCD patients vs. CTL subjects. ...................................................................................... 112 Figure 5.3 Heart rate variability responses during hypoxia. ...................................... 113 Figure 5.4 An example of measurements recorded during a hypoxic episode in SCD patients ............................................................................................. 120 Figure 5.5 Examples of perfusion drops following spontaneous sighs ..................... 120 Figure 5.6 Comparison of sigh frequency and likelihood of PU drop for each sigh in SCD patients vs. in CTL subjects ................................................. 122 Figure 5.7 Changes from baselines of r-wave to r-wave interval (RRI) following spontaneous sighs in SCD and CTL groups. ............................................ 123 Figure 5.8 Time-courses of parasympathetic indices of heart rate variability (HRV) in SCD and CTL groups ............................................................... 124 Figure 5.9 Time-courses of sympathetic indices of heart rate variability (HRV) in SCD and CTL groups ........................................................................... 125 Figure 5.10 Heart rate variability responses during sighs ........................................... 126 Figure 5.11 Linear regression plots between chances of perfusion (PU) drop if a spontaneous sigh occurs (in percent) and the heart rate variability parameters.. .............................................................................................. 128 Figure 5.12 Baseline physiological parameters which show significant changes post-transfusion compared to the pre-transfusion .................................... 131 Figure 5.13 Changes in inspired tidal volume (ViVol) from baselines following cold face stimulus ..................................................................................... 135 Figure 5.14 Hand Oxygen Saturation (rSO 2 ) from baselines following cold face stimulus ................................................................................................... 137 Figure 5.15 Peripheral Perfusion (PU) from baselines following cold face stimulus .................................................................................................... 139 Figure 5.16 Changes in systemic vascular resistance (SVR) from baselines following cold face stimulus .................................................................... 140 Figure 5.17 Changes in systolic blood pressure (SBP) from baselines following cold face stimulus ..................................................................................... 141 Figure 5.18 Changes in r-wave to r-wave interval (RRI) from baselines following cold face stimulus ..................................................................................... 143 xii Figure 5.19 Time-courses of heart rate variability (HRV) during cold face stimulus. ................................................................................................... 144 Figure 5.20 Heart rate variability responses during cold face ..................................... 145 Figure 5.21 Respiratory Sinus Arrhythmia Gain (G rsa ) Changes from baseline during cold face ........................................................................................ 147 Figure 5.22 Sample of 5-minute input data from a deep-breath maneuver, as well as the impulse responses and transfer function gains calculated from the minimal model. ................................................................................... 152 Figure 5.23 Average values of descriptors extracted from the impulse responses and transfer functions of ABR and RCC; pre- and post- transfusion. ...... 153 Figure 5.24 Transient response of arterial baroreflex (ABR) during 5-minute baseline. .................................................................................................... 156 Figure 5.25 Frequency response of arterial baroreflex (ABR) during 5-minute baseline. .................................................................................................... 157 Figure 5.26 Transient response of respiratory-cardiac coupling (RCC) during 5-minute baseline. .................................................................................... 158 Figure 5.27 Frequency response of respiratory-cardiac coupling (RCC) during 5-minute baseline ..................................................................................... 159 Figure 5.28 Changes in descriptors extracted from time-varying impulse responses and transfer functions of the arterial baroreflex (ABR) gain from baselines following cold face stimulus. ................................... 161 Figure 5.29 Changes in descriptors extracted from time-varying impulse responses and transfer functions of the respiratory-cardiac coupling (RCC) gain from baselines following cold face stimulus ........................ 162 Figure 5.30 Scatter plot between RBC count and WBC count .................................... 173 Figure 5.31 Wiring of the inflammatory reflex (figure from Tracey, 2002) ............... 175 Figure 6.1 Control box diagram of the minimal close-loop model (from Khoo, 2008). ................................................................................... 177 Figure 6.2 Diagram of the minimum model describing fluctuation of peripheral perfusion ................................................................................................... 180 Figure 6.3 Time series of (A) systolic blood pressure (SBP), (B) peripheral perfusion (PU contra and PU ipsi ) during an FMD experiment. ..................... 184 Figure 6.4 Impulse responses which represent (A) the effect of sympathetic control (h Symp ) and, (B) the effect of endothelial function (h Endo ) to the change in peripheral resistance. .......................................................... 184 xiii Figure 6.5 Physiological Signal recorded from a healthy subject breathing spontaneously (Figure from Eckberg, 1995). ........................................... 187 Figure 6.6 Muscle sympathetic nerve activity and respiration at different diastolic pressures (Figure from Eckberg, 1995) ..................................... 188 Figure 6.7 Relationship between x 1 = ViVol, x 2 = DBP and y = 1/SVR. Each point represents values of x 1 , x 2 , and y from a breath .............................. 189 Figure 6.8 Relationship between ViVol, DBP and SVR. Each point represents values of ViVol, DBP and SVR from a breath ......................................... 190 xiv Abbreviations ABR arterial baroreflex AIC Akaike Information Criterion ANC absolute neutrophil count ANS autonomic nervous system AR autoregressive ARX autoregressive with exogenous input Br/Min breathing frequency CFT cold face test CHLA Children’s Hospital Los Angeles CMBC concentration of moving blood cells CNS central nervous system CTL normal control DBP diastolic blood pressure DLF discrete Laguerre function ECG electrocardiograph FIR finite impulse response FMD flow mediated dilation G rsa respiratory sinus arrhythmia gain HbA regular adult hemoglobin HbS sickle hemoglobin HF high-frequency HFP high frequency power of HRV HFP ra respiratory-adjusted HFP HRV heart rate variability hs-CRP high-sensitivity C-reactive protein  forgetting Factor xv LF low-frequency LFP low frequency power of HRV LHR low-to-high ratio of HRV LHR ra respiratory-adjusted LHR MCV mean corpuscular volume MSNA muscle sympathetic nerve activity MDL minimum description length N 2 nitrogen NIRS near infrared spectroscopy NMSE normalized mean squared error NO nitric oxide O 2 oxygen pO 2 partial pressure of oxygen PSD power Spectral Density PU perfusion unit RBC red blood cell RCC respiratory-cardiac coupling Resp respiration signal RLS recursive least square RRI r-wave to r-wave interval RSA respiratory sinus arrhythmia rSO 2 regional oxygen saturation SaO 2 oxygen saturation SBP systolic blood pressure SCA sickle cell anemia SCD sickle cell disease SCT sickle cell trait SVR systemic vascular resistance TV-AR time-varying autoregressive xvi TV-ARX time-varying autoregressive with exogenous input TV-PEVR time-varying Parameter Estimate Variability Reduction USC University of Southern California Vent minute ventilation ViVol inspired tidal volume VOC vaso-occlusive crises WBC white blood cell WGN white Gaussian noise xvii Abstract Sickle cell disease (SCD) is a genetic disorder that is characterized by recurrent episodes of vaso-occlusive crisis (VOC) from the sickling behavior of red blood cells. Currently, no technique can distinguish the cause or predict the occurrence of a crisis accurately and reliably. One area which has rarely been studied in SCD patients is their autonomic nervous system (ANS). Since the ANS is responsible for the moment-to- moment control of the vascular tone, we hypothesized that the ANS plays an important role in the initiation of their VOC. Computational techniques, including spectral analysis of HRV and a model which characterizes the dynamics of baroreflex and respiratory- cardiac coupling, were used to assess cardiovascular autonomic control in SCD patients and normal control (CTL) subjects. These analysis techniques were applied to responses elicited from the subjects during the application of non-invasive and easily reproducible physiological interventions, such as transient-controlled hypoxia and the cold face test. Our results demonstrate impairment in the ANS in SCD patients. In particular, hypoxic responses in SCD subjects showed a significantly stronger parasympathetic withdrawal compared to the CTLs. Furthermore, the autonomic responses to the cold face stimulus in SCD subjects showed an absence of the shift to parasympathetic dominance, as evidenced in the CTLs. In addition to the HRV analysis, model-based assessment also revealed the absence of both arterial baroreflex and respiratory-cardiac coupling augmentations in SCD patients during the cold face stimulus, while in CTL subjects both mechanisms showed tendencies to increase during the test. xviii During the data analysis period, we noticed that spontaneous sighs triggered marked periodic drops in peripheral microvascular perfusion. While the sigh frequency was the same in both groups, the probability of a sigh inducing a perfusion drop was significantly higher in SCD subjects in comparison to the CTLs. Evidence for sigh- induced sympathetic nervous system dominance was seen in SCD subjects, but was not significant in CTL. HRV analysis suggested that the cardiac ANS responses to sighs are not different between the two groups, after adjusting for the effect of post-sigh respiration. However, the peripheral sympathetic response in SCD subjects appeared to be enhanced in this group relative to the CTLs; and, furthermore, sighs may play a role in initiation of vaso-occlusive events in this group of patients. In brief, all assessments we performed in this study suggested that the ANS responses to perturbations in SCD patients are more biased toward parasympathetic withdrawal and sympathetic activation, compared to normal controls. The complete mechanism is still a topic of investigation. Thus far, we have shown a relationship between the degree of this abnormality and the degree of both the anemia and infection/inflammation. We suspect that a mechanism related to the inflammatory reflex might play an important role in the ANS impairment in this group of patients. In conclusion, this study draws attention to an enhanced ANS-mediated peripheral sympathetic driven vasoconstriction in SCD that could increase red cell retention in the microvasculature, promoting vaso-occlusion. This cascade of events could be the mechanism which triggers the VOC. 1 Chapter 1. Introduction 1.1. Sickle cell disease (SCD) 1.1.1. Pathology SCD is a genetic disorder characterized by acute vaso-occlusive episodes during which normally flexible red blood cells (RBC) are transformed from their regular round shape to a rigid sickle shape, occluding the microvasculature. The most common types of SCD include: sickle cell anemia (SCA, SS genotype), sickle-hemoglobin C disease (SC genotype), sickle beta-plus thalassemia (S  + genotype), and sickle beta-zero thalassemia (S  0 genotype). In patients with SCD, regular adult hemoglobin (HbA) mutates into sickle hemoglobin (HbS) when valine substitutes for the glutamine acid at the sixth position of the beta-globin chain of hemoglobin (Frenette, & Atweh, 2007; Ingram, 1959). This causes HbS to polymerize in a de-oxy state, resulting in a marked reduction in red cell flexibility as oxygen is released from the RBCs (Frenette, & Atweh, 2007; Itano, & Pauling, 1949; Pauling et al., 1949). This polymerization process is reversed when the RBCs return to the lung and are re-oxygenated. In cases where this process fails to reverse, a large scale vaso-occlusive event might take place, causing adverse events such as acute pain, acute chest syndrome, multi-organ dysfunction, stroke, or renal dysfunction. As a result, vaso-occlusive events are a leading cause of death among SCD patients (Platt et al., 1994; Quinn et al., 2007). 2 Figure 1.1 Normal RBCs and sickle cells. a) Normal RBCs flowing freely in a blood vessel; b) sickle RBCs blocking flow in a blood vessel. (Retrieved January 20th, 2010 from the National Heart Lung and Blood Institute website: http://www.nhlbi.nih.gov/health/dci/Diseases/Sca/SCA_WhatIs.html) 1.1.2. Vaso-occlusive crises (VOC) While SCD is often characterized by episodes of VOC between periods of normalcy, the sickling process actually occurs continually. However, what triggers the transition from low-rate sickling to a wide-spread crisis is not known. Many attempts 3 have been made to assess the probability and severity of these crises, including daytime pulse oximeter measurements (Uong et al., 2006), biomarker measurements (Platt et al., 1994; Miller et al., 2000), transcranial Doppler measurements (Valadi et al., 2006), and echocardiograms (Ahmed et al., 2004). Although these have shown some degree of causal association to sickle cell related complications, they cannot yet reliably identify and predict the crises. Although certain factors, including inflammation, dehydration, infection, stress, cold weather, alcohol use, high altitudes, sleep apnea, and hypoxia, seem to predict VOCs, the mechanism that initiates the crises remains unknown. Nonetheless, it has been shown that upon deoxygenation of flexible sickle RBCs, there is a delay time before they polymerize extensively and form the rigid sickle shape (Eaton, & Hofrichter, 1995). If they fail to traverse the microvasculature within this period, sickle cell transformation might occur, causing an occlusion in that vessel bed. Thus, any factors which increase RBC transit time through the capillaries would increase the likelihood of a sickling event, and thus increase the risk of a VOC. One factor which controls peripheral blood flow, which in turn controls transit time, is the peripheral branch of the autonomic nervous system (ANS). In particular, the nervous system signals the peripheral vessel to constrict in response to stimuli. Although abnormalities in ANS are known to be risk factors for cardiovascular adverse cardiovascular events in the general population (Curtis, & O'Keefe, 2002), this link is unclear in SCD patients and sickle cell trait carriers, even though they have some degree 4 of cardiovascular autonomic system dysfunction (Connes et al., 2006; Hedreville et al., 2008; Jaja et al., 2008; Persson, 1996; Romero Mestre et al., 1997) (more details on autonomic dysfunction in SCD can be found in Section 0). In fact, up to 23% of deaths of adults with SCD are so-called “sudden death” events with no detectable cause found at autopsy (Darbari et al., 2006; Perronne et al., 2002; Platt et al., 1994). Thus, the present study hypothesizes that abnormal autonomic control may be responsible for triggering VOCs. We, therefore, propose to assess the autonomic control of heart rate and peripheral perfusion of SCD patients in order to determine if there is a causal link between the ANS and VOCs. 1.2. Cardiovascular autonomic control Figure 1.2 Representation of the autonomic control of the cardiovascular system 5 The moment-to-moment fluctuation of regional blood flow and heart rate in humans is partly regulated by the intrinsic properties of the myocardium and the vascular smooth muscles, as well as the ANS (Berne, & Levy, 2001). Heart rate is continually controlled by both sympathetic and parasympathetic branches of the ANS. While in the periphery, the intrinsic properties of the peripheral vessel, together with the peripheral sympathetic branch of the ANS regulates the vascular tone. A diagram representing the ANS controls of the cardiovascular system is shown in Figure 1.2. 1.2.1. Control of heart rate The ANS controls heart rate, mainly by its cardiac sympathetic and parasympathetic divisions (Berne, & Levy, 2001; Dampney, 1994; Kandel et al., 2000). In general, these two divisions function in opposition to each other to maintain homeostasis. The sympathetic system activation increases heart rate and constricts peripheral vessels, preparing the body for ‘flight or fight’. Activation of the parasympathetic system, on the other hand, decreases heart rate, restoring the body to a basal condition (Berne, & Levy, 2001). The parasympathetic system operates by releasing acetylcholine at the vagal nerve terminals, innervating the Sinoatrial (SA) and Atrioventricular (AV) nodes (Kandel et al., 2000). The release of acetylcholine directly activates special potassium channels in the cardiac cells without a second messenger operation. Because of this direct activation, these channels open quickly, resulting in a rapid decrease of heart rate, i.e. heart rate increases within seconds after the activation (Warner, & Cox, 1962). 6 While the parasympathetic system functions by innervating the heart at the SA and AV nodes, the sympathetic nerves innervate both the nodal regions and penetrate the myocardium along the coronary vessels. Activations of these nerves stimulate the beta type adrenergic receptors, resulting in an increased heart rate. Responses to these activations are much slower compared to parasympathetic responses (Warner, & Cox, 1962; Berne, & Levy, 2001) because the sympathetic signaling requires an intracellular buildup of second messengers, mainly cyclic AMP (Kandel et al., 2000). The difference in the speed of the heart rate responses between the sympathetic and the parasympathetic divisions leads to fluctuations in heart rate in different frequency ranges. This fluctuation in heart rate have been studied and used widely by researchers to assess the ANS functions non-invasively (ESC/NASPE, 1996) (more details about the analysis of this fluctuation and its applications in this study are described in Section 1.5). There are many factors that affect the sympathetic and parasympathetic controls of heart rate (Dampney, 1994; Hainsworth, 1991). The two important ones are respiration and blood pressure, which control heart rate through the respiratory sinus arrhythmia and the arterial baroreceptor reflex, respectively (Figure 1.3). The functions of these reflexes are described here and were also evaluated in this study. 7 Figure 1.3 Reflexes of the ANS 1.2.1.1. Respiratory sinus arrhythmia (RSA) Heart rate increases during inspiration and decreases during expiration, due to the RSA mechanism (Berne, & Levy, 2001; Neff et al., 2003). RSA is generated both by reflex mechanisms and by a direct interaction between respiratory and the cardiac vagal centers in the medulla. These centers receive reflex information from 1) the stretch receptors in the lung, 2) stretch receptors in the right atrium (Bainbridge reflex), and 3) baroreceptors in the carotid sinuses and aortic arch. The combination of these reflex signals results in a fluctuation of heart rate or RSA, primarily through the parasympathetic branch of the ANS (Eckberg, 1983; Katona, & Jih, 1975; Grossman et al., 1991). This leads to a rapid increase of the amount of acetylcholine released at the Blood pressure (baroreflex) Respiration Cardiac Output Parasympathetic Nervous System Sympathetic Nervous System Humoral Factors Thermoregulation 8 vagal nerve endings during exhalation, causing a rapid decrease in heart rate. Conversely, the reverse mechanism occurs during inhalation, causing an increased heart rate. This clear relationship between respiration and of heart rate fluctuation is widely considered a diagnostic target for assessment of the parasympathetic nervous system. 1.2.1.2. Baroreflex Another factor that regulates the fluctuation of heart rate is the blood pressure. Changes in arterial pressure result in an acute change in heart rate through the baroreflex response. This reflex receives information through the baroreceptors, stretch-sensitive mechanoreceptors, located in the aortic arch and in the carotid sinuses. These receptors are activated when there is an increase in arterial pressure; they then send a signal to the brain stem, where the sympathetic nerve are inhibited (Berne, & Levy, 2001; Kandel et al., 2000). This results in a decrease of both peripheral resistance and heart rate, decreasing the blood pressure to a stable value. The opposite mechanism is in place when the blood pressure drops in order to increase the blood pressure to a normal level. Not only does this reflex respond to rapid changes in blood pressure, it also operates continuously to maintain baseline blood pressure at a stable level. Its activities are believed to be partially responsible for the low-frequency fluctuations of heart rate (Sleight et al., 1995). 1.2.2. Control of peripheral perfusion Unlike the regulation of heart rate, which is primarily achieved through neural mechanisms, the regulation of peripheral blood flow is achieved through both local 9 (intrinsic) and central (extrinsic) mechanisms (Berne, & Levy, 2001). The resistances of the peripheral circulations are adjusted through the smooth muscle fibers lining the walls of small arteries and arterioles. These muscle fibers contract to increase vascular resistance or relax to decrease the resistance, increasing or decreasing the blood pressure, respectively. The contraction and relaxation of the smooth muscles in skin and splanchnic organs are largely controlled by the CNS; while in other organs, such as the heart and the brain, they are more essentially controlled by the local or intrinsic regulation (Berne, & Levy, 2001). 1.2.2.1. Intrinsic control of peripheral perfusion The intrinsic control of peripheral perfusion adjusts blood flow in some tissues to suit the metabolic activities, shear stress, and changes in arterial blood pressure. One function of this control is the myogenic mechanism, which allows vascular smooth muscles to contract when arterial blood pressure increases, and relax when the arterial blood pressure decreases, thus maintaining blood flow within an acceptable range when blood pressure changes. In particular, an increased intravascular pressure causes an increase of shear stress on the endothelial wall, activating a release of nitric oxide (NO), a potent vasodilation molecule (Joannides et al., 1995; Berne, & Levy, 2001; Persson, 1996). In addition to the regulation of NO, metabolic mechanisms also play a very important role in the local regulation of blood flow. Any intervention that causes the O 2 supply to decrease to a level inadequate for tissues to function normally results in the vasodilation response. 10 1.2.2.2. Extrinsic control of peripheral perfusion In addition to intrinsic regulation of perfusion which originate in the smooth muscle cells themselves, smooth muscles in blood vessels are also innervated with sympathetic nerve fibers (Berne, & Levy, 2001). These nerve activities help maintain tonic level of peripheral resistance. When the nerves are innervated, blood pressure increases; while when the magnitude or frequency of the nerve firing increases, the blood pressure decrease (Persson, 1996). The parasympathetic nerves, on the other hand, innervate relatively few blood vessels and thus have little effect on the peripheral resistance. 1.3. Sickle Cell Disease and Autonomic Dysfunction Cardiovascular dysfunctions have also been correlated with ANS abnormalities in the multiple diseases, including diabetes, sleep apnea, and heart disease. In SCD patients, some degree of physiological cardiovascular adaptation has been reported, including elevations in heart rate and declines in cardiac output, resulting primarily from associated chronic hemolytic anemia (Lester et al., 1990; Batra et al., 2002). Transient cardiac dysfunction and myocardial ischemia have also been reported during their crises (Norris et al., 1991; Deymann, & Goertz, 2003). Although acute atherosclerosis-related myocardial infarctions are rare in SCD patients, diverse cardiovascular anomalies including heart murmurs, cardiomegaly, biventricular hypertrophy, and abnormalities in the cardiac conducting system have been described by multiple researchers (Assanasen et al., 2003; Pavlu et al., 2007; Batra et al., 2002; James et al., 1994; Norris et al., 1991; 11 Manci et al., 2003). These dysfunctions in SCD patients are often correlated with sudden death. While some degree of ANS abnormalities in SCD patients and sickle cell trait carriers has been reported (Connes et al., 2006; Romero Mestre et al., 1997; Pearson et al., 2005), its sources, as well as its possible consequences to initiation of VOC, are still unclear. Therefore, we review these works on ANS functions in SCD patients as well as sickle cell trait carriers to get a better understanding on the works in this area. In the first study, Romero Mestre et al. employed a standard battery test to assess 25 SCD patients (Romero Mestre et al., 1997). The measurement results were compared with published normal ranges. They found that in all tests SCD patients had higher percentages of abnormalities than normal white and normal black control subjects (Figure 1.4). 12 Figure 1.4 Comparison of percentage of abnormalities in test results among 3 subject groups: SCD patients, normal whites of Hispanic origin, and normal blacks. (From Romero Mestre et al., 1997) The relationship between ANS abnormalities and SCD was also supported by Romero-Vecchione et al., who reported that during a tilt-test (a test of baroreflex dependent sympathetic reactivity), the levels of following responses were lower in 18 SCD patients compared to controls: increases in systolic and diastolic blood pressure, heart rate, plasma cortisol and aldosterone concentration (Romero-Vecchione et al., 2002). Although the mechanism is still not well understood, their study suggested an attenuation in baroreflex related sympathetic response in this group of patients. These findings were later confirmed by Jaja et al., who also suggested that this responses in SCD patients could be improved with the administration of vitamin C supplements (Jaja et al., 2008). 13 In another study, Pearson et al. applied both psychological and physical stimuli to 19 children with homozygous SCD (SCA or SS genotype) and found that children with higher degree of disease severity had lower RSA scores (r = –0.45, p < 0.05, Figure 1.5), (Pearson et al., 2005). Consequently, they concluded that children with greater levels of parasympathetic withdrawal during challenges had significant higher severities of SCD. Figure 1.5 Relationship between sickle cell disease severity and RSA Score (From Pearson et al., 2005) Like SCD patients, sickle cell trait (SCT) carriers also showed signs of ANS abnormalities (Connes et al., 2006; Connes et al., 2008). Studies by Connes et al. of 23 SCT subjects and 17 normal controls showed that during sleep the total power of heart rate variability was lower in SCT than in normal control. Moreover, the high-frequency power was lower, and the ratio between low and high-frequency of heart rate variability was higher in SCT than normal controls. This suggested a lower vagal and a higher 14 sympathetic modulation in SCT compared to controls. The data from this group also showed a negative correlation between blood viscosity and an index related to parasympathetic activity, suggesting that the ANS in SCT may be a result of their abnormal blood rheology. Although cardiovascular autonomic dysfunctions in SCD patients were reported by several groups, a causal relationship between these dysfunctions and a known risk factor of crisis in SCD patients has not been reported, at least to our knowledge. Consequently, we decided to study this rarely explored area. In the present study physiological responses to a known vaso-occlusion factor, namely hypoxia, were measured. Other standard autonomic stimuli such as flow mediate dilation, the cold face test, and vital capacity sigh were also employed to assess the autonomic responses in SCD patients. Following an exposure of each stimulus, the cardiovascular autonomic control was assessed non-invasively. All the measurements and computational techniques for this assessment are described in subsequence sections. 1.4. Background on Autonomic Control Stimuli used in this study 1.4.1. Controlled hypoxia As mentioned earlier, hypoxia is one of the most common triggers of VOCs (Quinn, & Ahmad, 2005). Research by Hargrave et al. found that low nocturnal oxygen saturations were significantly associated with elevated rates of painful crises in children 15 with SCD (P < 0.0001, n = 95) (Hargrave et al., 2003). In another study, nocturnal hypoxemia also was a good predictor of CNS events such as stroke or seizure in SCD (Kirkham et al., 2001). In the latter study, about 40% of patients with mean overnight O 2 less than 96% exhibited some CNS event within 5 years following the study. In contrast, less than 10% of patients with mean overnight O 2 more than 96% were found to have experienced a CNS event within the same 5 year period. Although many effects of hypoxia on SCD anemia have been studied, the effects of hypoxia on the ANS have rarely been explored. In the present study of 11 SCD patients and 14 normal controls we employed a brief, controlled episode of hypoxia as a noninvasive way to simulate hypoxia which occurs naturally overnight. 1.4.2. Vital capacity sigh During our hypoxia study, we noticed that SCD patients had pronounced physiological responses to spontaneous sighs. Therefore, we further studied these responses and found that vital capacity sighs have been used widely as a measure of peripheral sympathetic autonomic controls (O'Brien, & Gozal, 2005; Baron et al., 1993). Compared to the vasoconstriction response to deep breath in healthy controls, the vasoconstriction response to deep breath is depressed in neuropathy patients (Baron et al., 1993), but is increased in obstructive sleep apnea patients (O'Brien, & Gozal, 2005). Vaso-constriction following deep inspiration was first described by Binet and Sollier in 1895 (Binet, & Sollier, 1895). They observed that deep breathing was associated with a decrease in finger volume as measured by a finger plethysmograph. It 16 was not until 40 years later that Bolton et al. (1936) discovered that this event is related to sympathetic activation. In the absence of any stimulus, decreases in finger and toe volume of all 4 limbs of a subject were recorded in their experiments (Bolton et al., 1936). They discovered later that the volume decreases occurred immediately following a sigh. The latency between a sigh and the subsequent volume decrease was also found to be similar in all 4 limbs (2 – 3 seconds). A similar decrease in finger volume recurred with voluntary deep breath. They found that, in subjects who had undergone a sympathectomy, the associated limb did not show any decrease in volume after deep breathing, but the other 3 limbs did. They also showed that the decrease in finger volume was independent of cardiac output; when occluding the limb with a pressure cuff above the finger, the decrease in finger volume after deep breathing was still observable. The use of this test in assessing the autonomic control functions has gained more interest in the past few decades. This vasoconstriction response to deep breath has also been referred to with various other names, including sigh-vasoconstriction responses, inspiratory-gasp vasoconstrictive response (IGVR) (Rauh et al., 2003; Allen et al., 2002), or the deep breath vaso constriction reflex (Inwald et al., 1996). Measurement techniques for assessment of this response include plethysmograph or photo plethysmograph (PPG) and laser Doppler flow (LDF). A study has shown that the correlation of these 2 techniques increase with an increase in the number of successive sighs (Rauh et al., 2003). 17 In the present study, we employed this vital capacity sigh technique to assess the control of peripheral perfusion in 11 SCD and 11 normal control subjects. These data indicate that some SCD patients experience frequent drops in perfusion which often coincide with spontaneous sighs. 1.4.3. Flow mediated dilation (FMD) Assessment of the sigh-vasoconstriction reflex gives us some insight into the extrinsic control of peripheral blood flow; intrinsic blood flow regulation can also be assessed, using the FMD technique (Rodgers et al., 1990; Kelm, 2002). Pressure was applied on subjects’ forearms through a blood pressure cuff for 3 minutes. Upon release of the cuff, the rush of blood flow through the artery results in increased shear stress on the vascular wall. The endothelial cells response to this event by releasing NO causing, the vessel to dilate (Kelm, 2002). Impairment of FMD mechanism has been associated with atherosclerosis and other endovascular issues (Kelm, 2002; Cox et al., 1989). Results of previous FMD studies on SCD patients have been controversial. Rodgers et al. and Tharaux et al. have reported increases in peak flow and time-to-peak of the response in SCD patients compared to the response in normal controls (Rodgers et al., 1990; Tharaux et al., 2002), while de Montalembert et al. reported impairment of vasodilation response in SCD patients (de Montalembert et al., 2007). To assess the FMD response, the widely used techniques are the B-mode ultrasound and laser Doppler. Researchers who have employed laser Doppler to measure peripheral perfusion in SCD patients also observed increased oscillation in 18 microcirculation blood flow following the forearm occlusion (Rodgers et al., 1990). This oscillation in microcirculation flow at period of approximately 10 seconds was also shown in sickle cell patients in earlier studies by Kennedy et al. and Lipowsky et al. (Kennedy et al., 1988 ; Lipowsky et al., 1987). Its origin is still unclear. Work by Rodgers et al. has suggested that the oscillation may be caused by vasomotor tone and may represent compensation for the altered rheology of sickle cell patients (Rodgers et al., 1990). In the present study, we hypothesize that alterations of the FMD response in SCD patients could be multi-factorial and autonomic nervous reactivity could be a contributing factor. We also think that the 10-second oscillation of blood flow seen in the experiments by Rodgers et al. (1990) could be due largely to the sympathetic control of vasomotor tone. One supporting evidence was published by Stauss et al. who demonstrated that the sympathetic modulation of vasomotor tone in humans is most efficient with stimuli frequency between 0.075 and 0.1 Hz (Stauss et al., 1998), corresponding to the 10 second oscillation reported in Rodgers et al.’s study. In addition, several groups have also described the effects of sympathetic activity that reduced the degree of flow-mediated vasodilation response (Hijmering et al., 2002; Dyson et al., 2006). 1.4.4. Cold face test (CFT) Another test we employed to assess the ANS in SCD patients is the application of a cold face stimulus to the forehead, which triggers the diving reflex. This reflex has been shown to induce both parasympathetically-mediated bradycardia and sympathetically- 19 mediated increases in peripheral resistance (Heistad et al., 1968; Kawakami et al., 1967; Khurana, & Wu, 2006). Its function is to optimize respiration, facilitating a long stay underwater. Compared to humans, this reflex is stronger in aquatic mammals such as seals and dolphins (Zapol et al., 1989). The cold face test has commonly been used as a non-invasive method of assessing cardiovascular autonomic control. This test is utilized by applying a cold gel pack to the forehead, resulting in the activation of cardiac parasympathetic and peripheral sympathetic nervous systems (Stemper et al., 2002; Heistad et al., 1968). Activating these systems results in a decrease of heart rate and vasoconstriction. Studies have suggested that this reflex is independent of respiration pattern (Khurana, & Wu, 2006; Stemper et al., 2002), and thus it can be used to assess the autonomic nervous system with minimal cooperation from the subjects. 1.5. Computational Techniques to Characterize the Cardiovascular Autonomic Control After data were collected, we post-hoc analyzed the data to obtain information regarding ANS response to the stimuli. Two analysis techniques we used are heart rate variability and minimal model analysis. Some background information regarding these techniques are reviewed here. 20 1.5.1. Heart rate variability (HRV) Interest in using fluctuation in heart rate or HRV as a noninvasive measure to assess autonomic nervous activity has greatly increased in recent decades. However, it was not until the ‘90s that a group of experts in the field known as the Joint Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology set up standards for measurements and interpretations of HRV (ESC/NASPE, 1996). Since then these standards have been widely referenced by cardiologists and biomedical researchers. According to this reference, vagal activity is commonly accepted a major contributor to high-frequency components of HRV (HF, 0.15 – 0.4 Hz). On the other hand, the low-frequency components of HRV (0.04 – 0.15 Hz) may be due to both vagal and sympathetic activity (Eckberg, 1997). Thus, the ratio of LF to HF spectral powers has been used widely by researchers as an index of “sympathovagal balance” (Cerutti et al., 2001; ESC/NASPE, 1996). The amount of high and low frequency content in ECGs can be quantified through power spectrum analysis. The content of a frequency range can be express as a number, or “power” which has been widely used to quantify levels of vagal activity. Note that the HRV parameters represent degrees of autonomic regulation rather than absolute levels of autonomic tone. An example of a HRV analysis result during an orthostatic test is shown in Figure 1.6. 21 Figure 1.6 A classic example of spectral analysis of heart rate variability (HRV). (a) In this orthostatic test, when the subject was at rest, the power in the low frequency (LF) range was similar to the power in the high frequency (HF) range, as shown by the power spectral density plot and the pie chart. (b) When the subject was tilted vertically, the sympathetic modulation increased while the vagal modulation decreased. This resulted in diminished HF power and the LF power become more dominant. The total power of HRV also decreased, as shown by the size of the pie chart. (From ESC/NASPE 1996, p. 359) 1.5.2. Minimal model characterizations of physiological systems In addition to assessing functions of the ANS through direct measurement of HRV power in different frequency ranges, mathematical modeling can also be employed to examine the mechanisms which regulate the measurements of vital signals while reducing the other effects that are subject to wide range variation. Our minimal model of HRV control intends to explore the dynamics of arterial baroreflex and respiratory- cardiac coupling, the two primary reflexes which regulate heart rate variability. Diagram showing the control of these reflexes are presented in Figure 1.3. a) b) 22 The results from this model allow us to gain insight on the contributions of respiration and blood pressure to HRV. Previous studies from our research group tested this model in obstructive sleep apnea patients and showed that this group of patients has a significant reduction of both respiratory-cardiac coupling and baroreflex gains (Jo et al., 2003; Khoo, 2008). The mathematical computation of this model has been revised in this study for a better accuracy and faster computational time. More details about both models and computational techniques are presented in Chapter 4. 1.6. Aims of the Study 1.6.1. To non-invasively assess cardiac autonomic control using an improved method for HRV analysis in subjects with SCD. Our first aim is to assess the autonomic reactivity of SCD patients using non- invasive, reproducible physiological interventions (such as hypoxia; cold face test; vital capacity sigh). The subsequent changes in HRV from the interventions were analyzed using a time-varying respiration-adjusted spectral analysis technique, which removes variations in heart rate due to respiration and permits quantification of HRV due to autonomic dysregulation alone. In addition, the relationships HRV parameters and blood markers were also analyzed to gain insight into the pathophysiology that underlies the autonomic dysfunctions. 23 1.6.2. To determine the relationship between changes in cardiac autonomic control as determined using heart rate variability and changes in peripheral vascular perfusion. As decreased perfusion could lead to a VOC, we monitored the changes in peripheral perfusion using laser-Doppler spectrometry, along with other physiological measurements. We aimed to assessed, from these data, whether there was a change in microvascular perfusion in response to autonomic stimuli; and if so, whether there was a relationship between changes in perfusion and possible accompanying changes in autonomic cardiovascular control, as deduced from HRV analysis. 1.6.3. To quantify the main determinants of heart rate variability, using a mathematical modeling approach A number of primary mechanisms are involved in the control of heart rate. We aimed to quantify the relative roles of these mechanisms using a mathematical model to explain physiological processes. This model characterizes the dynamics of arterial baroreflex and respiratory-cardiac coupling, the two primary reflexes which regulate heart rate variability. The ability to more specifically identify the extent of abnormality in each of these mechanisms may lead to an improved regimen of treatment for SCD patients. 24 1.7. Significance of the Study A large number of patients with SCD die from sudden death with no cause found on autopsy (Platt et al., 1994; Perronne et al., 2002). Autonomic dysregulation leading to characteristic changes in heart relate variability has been shown to be associated with sudden death due to cardiac arrhythmia in many diseases. Proving this relationship in SCD patients would enhance our understanding of the mechanism of crisis which could possibly be clinically useful. This study may also aid understanding of the mechanism of the disease by quantitatively characterize the factors which cause stasis in microvasculature leading to vaso-occlusive crisis. Ultimately the modeling approach applied here may possibly lead to detection of early signs of crises and improved quality of life for SCD patients. 1.8. Organization of the Dissertation The objectives and background pertinent to this research were presented in the previous sections of this chapter. Chapter 2 describes the subjects, experiment setups, and the autonomic intervention used for assessing the ANS responses in SCD patients. After each experiment raw data from all channels were synchronized and pre-processed, these processes are also discussed in Chapter 2. Chapter 3 presents the HRV analysis technique. A time-varying analysis of HRV was used to assess the function of the sympathetic and the parasympathetic divisions of the ANS in all subjects. Techniques which allow tracking of rapid changes in HRV 25 following intervention and adjust for the effect respiration pattern changes during intervention were proposed in this study and are presented in Chapter 3. In addition to assessing the ANS through HRV analysis, a minimal model which describes the control of heart rate is presented in Chapter 4. Physiological concepts, mathematical descriptions, and testing of the model using simulated data are presented in this chapter. Chapter 5 presents results and discussion of the results. Baseline measurements of the ANS functions and baseline values of model parameters were compared between SCD patients and normal controls. ANS responses to autonomic stimuli and spontaneous sighs are presented and compared among subject groups. Parameters extracted from computational models which describe the control of heart rate. In addition, we present the correlation results between HRV parameters and information from blood tests in this chapter. Future research opportunities in the area of this study are discussed in Chapter 6. Additional modeling concepts and some preliminary results for these possible future directions are presented here. Finally, Chapter 7 presents an overall discussion and conclusion of the study. 26 Chapter 2. Experiments and Data Processing To study the cardiovascular autonomic control in sickle cell patients, we employed several physiological measurement systems and non-invasive interventions. This chapter discusses the experimental setups for these interventions and the data pre- processing techniques used to prepare data for subsequent analyses (details in Chapter 3 and 4). All experiments included in this study were parts of multiple on-going clinical trials carried out at Children’s Hospital Los Angeles (CHLA). Each trial was designed by a collaborative research group consisted of researchers from CHLA and University of Southern California (USC), and carried out by physicians from CHLA. 2.1. Subjects Results included in this dissertation were obtained as a part of 3 clinical trials that looks at different aspects of the ANS in SCD patients. The main purpose of subject recruitment and experiments performed in each trial was different based on the requirements of each principle investigator. Since this was a collaborative project in which all experiments took place at CHLA, some of the subjects were recruited separately based on different criteria while some of the subjects participated in multiple experiments from different trials. For all subjects, informed consent was obtained prior to each experiment. The protocol for the study was approved by the institutional committee on human experimentation at CHLA. Characteristics of the subjects are described below. 27 2.1.1. Clinical trial 1: Controlled Hypoxia The main objective of this trial was to compare physiological responses between SCD patients and normal controls (CTL) to hypoxia. We measured physiological and autonomic responses to a transient controlled hypoxia stimulus in 11 sickle cell patients and 14 CTL subjects, all of which were African American. Characteristic of the subjects are shown in Table 2.1. Eligibility criteria were: 1) thirteen years or older; 2) homozygous sickle cell disease, hemoglobin SC disease, hemoglobin S β 0 -thalassemia; 3) no transfusions in the past three months; 4) no sickle cell-related symptoms in the past month; 5) no chronic anti-inflammatory therapy; 6) no infections or acute medical problems within four weeks; 7) no pregnancy at time of experiment; 8) signed informed consent. CTL subjects met the same criteria, except for not having SCD. During the study, five of the 11 patients were treated with hydroxyurea, while the other 6 patients were not treated with hydroxyurea. Two of the control subjects were sickle cell trait carriers, while the other 12 were not. 28 Table 2.1 Characteristic of the subjects in the controlled hypoxia trial SCD (n = 11) CTL (n = 14) female/male 5/6 8/6 Age (years) 21.5  4.4 30.9  7.3 Genotype 8 SS, 3 SC 12 AA, 2 AS Hemoglobin (g/dl) 10.0  1.5 13.6  1.5 Reticulocyte count (%) 8.2  5.4 1.1  0.3 2.1.2. Clinical trial 2: Pre- and post-transfusion As of February 2011, this clinical trial is ongoing; and participants are still actively recruited. Parts of the data obtained from this trial were used in this dissertation. The main objective of this trial was to assess the relationship between chronic transfusion and FMD responses in SCD patients. Subjects in this protocol were in chronic outpatient transfusion program. During the period of this study, patients continued their transfusions according to their standard clinical protocols. No patient was transfused specifically for the purposes of this study. For each chronically transfused SCD patients, the data were captured twice. 1) The initial visit (pre-transfusion): on the day of their transfusion, prior to their regular transfusion; and 2) The second visit (post-transfusion): on the day after transfusion or within 120 hours from the transfusion. 29 The wait period between transfusion and the second visit was required as transfusions might cause abrupt changes in cardiac output, blood pressure and other hemodynamic parameters, due to fluid changes. Thus we chose to make the measurements on the day after transfusion to allow stabilization of blood volume. For non-transfused SCD patients and normal control subjects, the data were captured at one visit. The pre- and post- transfusion data from 9 chronically transfused patients were analyzed as a part of this dissertation. We chose subjects who had not had any symptoms of sickle cell crisis within two weeks before the measurements were made. Characteristics of subjects who participated in this trial are shown in Table 2.2. Table 2.2 Characteristic of the subjects in the pre- and post transfusion trial Pre-transfusion Post-transfusion female/male 6/3 Age (years) 24.8  14.0 Years on Transfusion 9.1  4.3 Hemoglobin (g/dl) 9.4  0.9 12.0  1.2 HbS (%) 25.5  9.3 18.3  4.6 Reticulocyte count (%) 8.8  5.7 5.6  2.7 2.1.3. Clinical trial 3: Cold Face Test The objective of this trial was to assess the autonomic responses to hypoxia and the CFT, which is one of the widely use tests to assess the ANS (more details about this test see sections 1.4 and 2.3). The subjects from this trial include SCD patients (SS, SC, Sb 0 30 thalassemia, SB + thalassemia), both in the chronic transfusion program or not, as well as the CTL subjects. All subjects were 10 years of age and older at the time of their experiments. As of February 2011, this trial is ongoing in which participants are still actively recruited. Data from 36 subjects (14 CTL, 15 non-chronically transfused, and 7 chronically transfused subjects) were analyzed as a part of this dissertation. Characteristics of these subjects are shown in Table 2.3. Table 2.3 Characteristics of the subjects in the CFT trial Control Non-chronic-transfused Chronic-transfused female/male 7/7 9/6 3/4 Age (years) 29.3  13.2 22.7  12.9 23.6  11.9 Hemoglobin (g/dl) N/A 9.3  2.8 9.2  1.8 Hematocrit (%) N/A 26.8  8.0 27.5  4.2 Reticulocyte count (%) N/A 6.5  5.0 10.9  6.1 Note that for the chronically transfused patients, the experiments were conducted prior to their transfusion and on the day of transfusion. 2.1.4. Sample Size Based on our preliminary data from the controlled-hypoxia experiments (Sangkatumvong et al., 2008; Sangkatumvong et al., 2010), we have performed a power analysis and determined that 20 subjects per group were needed in order to obtain the statistical power of 0.8 for at least 2 main parameters (Table 2.4 and Figure 2.1). 31 However, during the period of this study, we have not analyzed the data to this number (20 subjects per group) yet. Therefore, we believe that the reason why some of the results showed no statistically significance could be due to the lack of sufficient number of subjects. A larger group of subjects will be required for a future study. We used the power analysis method to calculate the sample size. The effect size and standard variation for the power analysis (Cohen, 1988) were estimated from means and standard deviations of the 3 parameters shown in Table 2.4 and Figure 2.1, assuming a 2-tail t-test. Table 2.4 Data used for a power analysis to determine the number of subjects for this study. Each parameter is 30-second average value of heart rate variability responses to hypoxia. RRI = R-wave to R-wave Interval, HFP ra = respiratory-adjusted high-frequency power, and LHR ra = respiratory-adjusted low frequency to high frequency ratio Mean ± SD SCD (n = 8) CTL (n = 8) RRI -10.329 ± 3.160 -0.752 ± 5.421 HFP ra -35.984 ± 17.711 -23.742 ± 22.264 LHR ra 61.767 ± 64.576 12.567 ± 37.567 32 Figure 2.1 Total sample size as a function of power computed from the preliminary values of differences of RRI, HFP ra , and LHR ra between SCD and CTL 2.2. Measurements During all experiments, the subjects lay awake in a comfortable supine position for at least 10 minutes before each maneuver to allow their vital signs to return to the baseline levels. Five-minute baseline measurements were obtained during that time. The LifeShirt® physiological monitoring system (VovoMetrics, Ventura, California) was used to directly measure finger oxygen saturation (SaO 2 ), electrocardiograph (ECG), respiratory trace, and end-tidal carbon dioxide, and from that information extrapolate r- wave to r-wave interval (RRI), minute ventilation (Vent), breathing frequency (Br/Min), and Inspired Tidal Volume (ViVol). The PeriFlux laser Doppler flowmetry (PeriFlux System, Perimed, Sweden) was used to record the peripheral perfusion on a nail capillary bed of the index finger of each of the subject’s hands. The Nexfin continuous non- 0.2 0.4 0.6 0.8 1 10 20 30 40 50 60 70 80 Power Sample Size per Group from RRI from aHFP from aLHR 33 invasive blood pressure monitor [BMEYE, Amsterdam, The Netherlands] was used to record blood pressure. Near infrared spectroscopy (NIRS) transducers were placed on the back of subjects’ hands on both sides to monitor the regional oxygen saturation (rSO 2 ) [INVOS® System, Somanetics, Troy, MI]. All the measurements mentioned above are non-invasive. An example of an experimental setup is shown in Figure 2.2 shows. All signals were sampled at the instrument’s maximum sampling rate as shown in Table 2.5. Data from all recording devices was first imported into MATLAB numerical computing program (MathWorks Inc., Natick, Massachusetts) and then synchronized using a program developed with MATLAB. 34 Figure 2.2 Experimental setup. Photo taken from an experiment from the controlled hypoxia protocol. Table 2.5 List of measurements acquired during experiments and their sampling frequencies Measurements Instruments Sampling Frequencies (Hz) ECG LifeShirt® 200 Respiration LifeShirt® 50 Skin Perfusion Periflux 32 Blood Pressure Nexfin 200 Tissue oxygenation Somanetic 0.2 Respiratory + ECG sensors Cerebral Oxygenation Sensor End-tidal CO 2 Sensor Pulse Oximeter Sensor Micro-capillary blood flow sensor Breathing Valve 35 We also took blood samples prior to each experiment for measures of viscosity, markers for anemia (e.g. red cell and white cell counts, reticulocyte count, and plasma Hb), and a marker for inflammation (i.e. high sensitivity c-reactive protein) (see Section 5.7 for more detail). Most of these measurements are routinely collected as part of our standard chronic transfusion protocol. 2.3. Experimental Procedures In all experiments, we measured most (if not all) parameters mentioned above, as well as collected blood samples only in SCD subjects. Subject characteristic information, such as age, sex, weight, height, were also recorded prior to each experiment. This section explains the protocol for all 4 maneuvers used in this study, which include controlled hypoxia, sighs, the CFT, and the FMD test. Scientific background about these maneuvers is shown in section 1.4. 2.3.1. Controlled Hypoxia As hypoxia is a known trigger to vaso-occlusive crises in SCD, a brief, controlled episode of hypoxia was used to noninvasively simulate natural nocturnal hypoxia episodes. During the experiments, the subjects spontaneously breathed through a face mask connected to a one-way valve which allows switching from room air to 100% nitrogen (N 2 ). Subjects were awake in a supine position and breathed room air for at least 10 minutes before the valve was switched to 100% N 2 for 5 breaths, then switched back 36 to room air. The subject underwent the second 5 breath exposure to N 2 approximately 15 minutes later. The duration and magnitude of the hypoxic stimulus were designed to mimic the episodes of hypoxia that occur naturally during sleep, while at the same time taking into account considerations for subject’s safety. The whole process was controlled so that the subjects were not aware of the change from room air to N 2 . Each N 2 breathing session produced a transient bout of hypoxia, which could be observed by measuring the SaO 2 drop (see Figure 2.3 for an example). 37 Figure 2.3 An example of measurements recorded during a hypoxia episode in an SCD patient. (A) SaO 2 measured by a pulse oximeter. (B) Microvascular perfusion as measured by Laser Doppler; perfusion unit (PU) is represented as arbitrary unit (au). (C) Respiratory trace (resp) (C) r-wave to r-wave interval (RRI). Time t = 0 indicates the nadir of oxygen saturation after the N 2 breathing. The session which caused a maximum SaO 2 drop and absence of movement artifact was selected for analysis. The time at onset of the SaO 2 drop to its minimum value following the hypoxia stimulus was noted as t = 0 for each set of data. SaO 2 (%) 60 70 80 90 100 PU (au) 0 5 10 15 Resp (L) 0 1 2 3 time (seconds) -100 0 100 200 300 RRI (sec) 0.8 A) B) C) D) 38 2.3.2. Vital Capacity and Spontaneous Sighs During our hypoxia experiments, we observed marked peripheral perfusion drops which occurred immediately following spontaneous sighs (see Figure 2.3 B) and C) for an example). We further explored this incidence and found that it is a vaso-constriction reflex which is mediated by the sympathetic nervous system. Thus we decided to explore the response to sighs in SCD patients in detail and to incorporate it into our protocol. More background detail on this sigh-vasoconstricion reflex is presented in 1.4.2. We also employed this reflex as a technique to assess peripheral perfusion control in SCD patients, by employing a deep breath maneuvers in our experiments. We analyzed data from vital capacity sighs (deep breath maneuvers) as well as from spontaneous sighs detected throughout an experiment. A vital capacity sigh maneuver was conducted after the subject had rested in a supine position and the baseline measurements were collected for at least 10 minutes. The subject was instructed to take a deep breath in and fully out for 1 breath, then relaxed and return to spontaneous breathing. This procedure was repeated twice, each with 1 minute separation. Some measurements from this maneuver are shown in Figure 2.4. 39 Figure 2.4 An example of data from the deep breath protocol. The subject was instructed to perform 3 vital capacity sighs at 1 minute intervals. A) respiratory trace or tidal volume (V t ), B) R-wave to R-wave interval, C) systolic blood pressure (SBP), and D) peripheral perfusion (PU). 2.3.3. Flow Mediated Dilation The FMD technique has been widely used by researchers to assess vascular health, with standard techniques to assess FMD response include both B-mode ultrasound and laser Doppler perfusion measurements. Lack of an FMD response has been associated with atherosclerosis and other endovascular issues (Kelm, 2002; Cox et al., 1989). More background on this technique is presented in Section 1.4.3. 50 100 150 200 250 300 0 500 1000 V t (liters) 50 100 150 200 250 300 0.6 0.8 1 1.2 RRI (seconds) 50 100 150 200 250 300 100 110 120 SBP (mmHg) 50 100 150 200 250 300 50 100 150 PU (au) time (seconds) A) B) C) D) 40 In this study, we conducted FMD experiments in order to assess both the standard and the oscillating component of the FMD responses. The protocol of the experiments was as follow. After the patients had rested and the baseline measurements had been collected for 10 minutes, a pressure cuff was inflated around their upper arms to the level above subjects’ systolic blood pressure, remained for 3 minutes, and then released. All signals were monitored during the experiment and up to 10 minutes after the release of the cuff. An example of data from this protocol is shown in Figure 2.5. Figure 2.5 An example of data from the FMD protocol. The brachial artery of the subject was occluded for about 3 minutes before releasing. A) tidal volume (Vt), B) R-wave to R-wave interval, C) systolic blood pressure (SBP), and D) peripheral perfusion (PU). A) B) C) D) 0 100 200 300 400 500 600 0 200 400 V t (liters) 0 100 200 300 400 500 600 0.7 0.8 0.9 RRI (seconds) 0 100 200 300 400 500 600 60 80 100 SBP (mmHg) 0 100 200 300 400 500 600 0 100 200 PU (au) time (seconds) 41 2.3.4. Cold Face Test The CFT is among the most widely used tests for the ANS. We have summarized some background on this test in Section 1.4.4. To test this reflex in SCD patients and normal controls, we placed a wet ice pack (0˚ C) on the subject’s forehead for 1 minute. All signals were monitored during the experiment and up to 10 minutes after the removal of the ice pack. An example of the test and a data recorded from a test is shown in Figure 2.6 and Figure 2.7.This protocol has previously been used by our group for assessing cardiac autonomic response in obstructive sleep apnea patients (Chaicharn et al., 2006). Figure 2.6 Experimental setup for the cold face test protocol. 42 Figure 2.7 An example of data from the deep breath protocol. The ice pack was placed on the subject’s forehead for 60 seconds before being removed. Time t = 0 indicates on-set of the 60- second application of the cold pack. A) inspired tidal volume ( ∆ViVol), B) R-wave to R-wave interval ( ∆RRI), C) regional oxygen saturation ( ∆rSO 2 ), and D) peripheral perfusion ( ∆PU). 2.4. Data Pre-processing During each experiment signals were collected from multiple devices at their maximum sampling rates, ranging between 0.2 – 200 Hz as detailed in Table 2.5. These signals were first re-sampled to 30 Hz using a linear interpolation technique for pre- processing, in order to have a uniform sampling rate across all the devices. ViVol (mL) 0 100 200 RRI (sec) -20 0 20 40 60 rSO 2 (%) -30 -20 -10 0 10 time (sec) 0 100 200 300 PU (au) -100 -50 0 50 A) B) C) D) 43 2.4.1. Synchronization of the devices Before each experiment, we tried to manually synchronize all the device clocks to be within 1 second from the experimental computer clock. The implementation, however, was less than ideal. Therefore in order to adjust for this asynchronization issue among data recorded from different devices, we developed a program on Matlab® to adjust for the delay between devices. This program compared the RRI derived from blood pressure (Nexfin device) and ECG (Lifeshirt device). The location of the peak of the cross- correlation between RRI measured by these 2 devices was used to find the time delay between them. We found that the time-shift between devices was up to 1 minute and had corrected for it accordingly. We also observed some timing asynchronization between a Lifeshirt® signal and the computer clock. To adjust for this discrepancy, we used the perfusion unit (PU) signal recorded by the Perimed device based on the computer clock. The location of the peak of the cross-correlation between the PU and the r-wave marker derived from the ECG signal was used for synchronizing the data from the Lifeshirt® device. 44 2.4.2. Computation of partial pressure of oxygen One of the important parameters measured in this study is the SaO 2 . While SaO 2 reflexes the level of arterial oxygenation, the partial pressure of O 2 (pO 2 ) is what the chemoreceptor detected and adjusted for by increasing the ventilation through the chemoreflex (Marshall, 1994). Because SCD patients are known to have a right-shifted oxygen dissociation curve ((Becklake et al., 1955), Figure 2.8), the measured SaO 2 may look lower in SCD patients than normal controls while the corresponding pO 2 may have a similar value. 45 Figure 2.8 Oxygen dissociation curves showing relationships between partial pressure of O 2 (pO 2 ) and Oxygen saturation (SaO 2 ) for normal controls, sickle cell traits carriers, and sickle cell patients (Figure fromBecklake et al., 1955) Since we utilized hypoxia as a stimulus of the cardiovascular autonomic system in this group of patients, it was crucial to compare the degree of stimulus between the control and patient groups. In order to do so, we chose the signal which was sensed by the receptor in response to hypoxia, which is pO 2 . Therefore to compare the degree of stimulus applied to each group, we converted SaO 2 to pO 2 using the subject’s own oxygen saturation curve. 46 The oxyhemoglobin dissociation curves for HbS and HbA were used to estimate the pO 2 form the SaO 2 data in these experiments. To accomplish this, the O 2 dissociation data collected from our Hemox-Analyzer (TCS Scientific Corporation, New Hope, Pennsylvania) were fitted to a sigmoidal curve using the least-squares method for each patient. The relationship between SaO 2 and pO 2 is shown in Equation 2.1. 0 2 2 2 1 2 * 1 ln x dx A SaO A A pO                    Equation 2.1 where A 1 , A 2 , x 0 , and dx are parameters that characterize the shape of each sigmoidal O 2 dissociation curve. We then used the parameters for each subject to convert SaO 2 to pO 2 . The dissociation data fit this equation for each patient with an R 2 > 0.99. Thus, using the set of parameters, A 1 , A 2 , x 0 , and dx for each patient, we calculated the pO 2 for each time series and removed the differences between patients and controls due to different oxygen binding properties of their respective hemoglobin types. 2.4.3. Manually data clean up All raw 30 Hertz data analyzed as a part of this study were manually inspected prior to any analysis. Data sections with either measurement noise or physiological anomaly were removed and linearly interpolated to fill in the missing part. If the abnormal sections in a recording lasted for a significant portion of the experiment or occurred during the section of interest (e.g. during or immediately after an autonomic stimulus), the dataset would be excluded from the analysis. Examples of data sections 47 that were manually cleaned up are shown in Figure 2.9. Only RRI traces are shown in these examples, nonetheless, we applied the same maneuver to all signals analyzed in this study. Figure 2.9 Examples of data sections which have been manually removed and interpolated. (a) raw RRI with measurement noise; and (b) raw RRI with ectopic beats. The gray traces present the ECG signal; the light blue traces present the original RRI values; and the dark blue traces present the corrected RRI values. 2.5. Statistical Analyses To compare autonomic responses among subject groups (controls, transfused- and non-transfused- SCD subjects), statistical analyses are needed. We performed these analyses and verified the results using multiple statistical packages, including SigmaPlot ® (Systat Software, Inc., San Jose, California), JMP ® (SAS, Cary, North Carolina), and time (sec) 0 1020 3040 ECG, RRI 0.0 0.2 0.4 0.6 0.8 1.0 1.2 time (sec) 0 1020 3040 ECG, RRI 0.0 0.4 0.8 1.2 1.6 A) B) 48 Matlab ® (MathWorks, Natick, Massachusetts). The statistical analyses we used include rank-sum test, chi-square test, two-way repeated measures analysis of variance, and regression analysis. This section describes the usages of this tests in our study. 2.5.1. Rank-sum Test To statistically assess differences between the SCD and CTL subjects, we used the rank-sum test to compare the following measurements between the two groups: baseline physiological and HRV parameters, probabilities of perfusion drops for each sigh, and frequencies of sigh breaths. 2.5.2. Chi-square Test The association between sighs and perfusion drops was analyzed using the chi- square test. Each breath was treated as an event, and was categorized as either a sigh or a non-sigh breath. If a breath was immediately followed by a PU drop, that breath was marked as a PU drop event. The number of occurrences of each event (breath with sigh/no sigh, PU drop/no PU drop) were used to construct a 2x2 contingency table for each subject, reflecting the relation from sighs to perfusion drops, an example of which is shown in Table 2.6. This information was subsequently used in a chi-square test for the possible association between sighs and PU drops for each subject. 49 Table 2.6 Example of 2x2 contingency table presenting a relationship between occurrences of sighs and PU drops in one subject. Number of Breaths PU drop no PU drop sigh 33 1 no sigh 6 355 2.5.3. Two-way Repeated Measures Analysis of Variance Two-way repeated measures analysis of variance (2W RMANOVA) was performed on all HRV indices to compare the SCD and CTL groups. One set of measurements from a baseline period and another during a stimulus period were obtained from each HRV time-course. Post-hoc pairwise comparisons using the Holm-Sidak method were conducted to determine whether: a) each HRV index following a stimulus (sigh, hypoxia, or cold face) differed from its baseline; b) the values of the HRV index in response to a stimulus from the two groups of subjects differed from each other. 50 2.5.4. Regression Analysis The last statistical test we used in this study was the regression analysis. Main measures from blood tests regularly used for diagnostic of SCD were regressed with HRV changes in response to the hypoxia stimulus. We assumed a linear model for all pairs of parameters in this study. This allows us identify the relationship between the underlying pathology and the ANS responses. 51 Chapter 3. Time-varying Analysis of Heart Rate Variability 3.1. Introduction In all studies described in this dissertation, time-varying spectral estimation of HRV was used to non-invasively assess the ANS responses to the autonomic stimuli. Spectral powers of HRV in different frequency ranges have been shown to characterize the sympathetic and the parasympathetic divisions of the ANS (ESC/NASPE, 1996). In particular, high-frequency power (HFP, 0.15 to 0.4 Hz) of HRV has been widely used to quantify the levels of the parasympathetic modulations. The ratio of low-frequency power (LFP, 0.04 to 0.15 Hz) to HFP (denoted as low-to-high ratio, LHR) has been used as a broad index of sympathovagal balance. More background information about the autonomic nervous system and heart rate variability analysis has been described in Chapter 1. Measurement data, including ECG, blood pressure, respiration, and oxygen saturation, used for our analysis were acquired from multiple devices, at various sampling frequencies as detailed in Chapter 2. These data were then pre-processed to improve the data quality and then re-sampled at 2 Hz for further processing. We consider this frequency to be adequate since the fluctuation of heart rate or HRV contains frequency components mainly up to 0.5 Hz (ESC/NASPE, 1996). A sampling rate of 2 Hz would 52 allow us to study the parameters of interest up to 1 Hz, thus adequate for studying the cardiovascular autonomic control through heart rate and for obtaining a good resolution for frequency domain analysis. 3.2. Recursive Autoregressive Technique for Analysis of HRV In order to track rapid ANS responses to stimuli such as hypoxia or sighs, we used a time-varying autoregressive (TV-AR) model with a recursive least square (RLS) estimation (Bianchi et al., 1993; Cerutti et al., 2001; Blasi et al., 2003) to compute the power spectral density (PSD) of the RRI. This method allows a new estimate of the RRI spectrum to be calculated with each successive time-step (0.5 second in our case). For each point in time the RRI can be described with an autoregressive (AR) model as follows: ) ( ) ( ) ( ) ( 1 n e j n y n a n y p j j       Equation 3.1 where y(n) represent RRI at time n; p represents the orders of the model; a j are the model parameters which were estimated using the RLS minimization algorithm; and e(n) represents the residual error of the model estimation at time-point n. Once the model parameters a j at each time step n are estimated, PSD at time n can then be calculated as: 53 2 1 2 2 ) ( 1 ) , (      p j T j f i j e y e n a T f n S   Equation 3.2 where S y (n, f ) represents the spectrum of the RRI time-series at time n; 1 i   , and T is the sampling period which is 0.5 second in our case. A sample of PSD during a transient hypoxia episode is shown in Figure 3.1. From this estimate of the running RRI spectrum, we calculated the HRV parameters, including HFP, and LHR, on a time-varying basis, using the same technique described in (Blasi et al., 2003). 54 Figure 3.1 Sample of PSD and its corresponding RRI from our hypoxia study. Note: time t = 0 in this case, indicates the onset of an oxygen saturation (SaO 2 ) drop following a hypoxia stimulus 3.3. Recursive Least-squares Parameter Estimations A weighted recursive least-squares algorithm was used to estimate the parameters a j from Equation 3.1. For each step the changes in RRI are modeled using an AR model with a model order between 4 to 10. Model order selection is presented in section 3.3.1. An important model parameter for this time-varying technique is the forgetting factor ( ), 10 1 10 2 10 3 10 4 10 5 10 6 -10 0 10 20 30 40 50 60 0.1 0.2 0.3 0.4 PSD (ms 2 /Hz) time (seconds) Frequency (Hz) time (seconds) 020 40 60 RRI (ms) 600 700 800 900 55 ranging from 0 to 1, which reflects the memory of the adaptive filter. When =1, all data before the present time are used in computing a PSD estimate. Small  implies that the most recent data points are weighted much more heavily. Method for selection of  is discussed in section 3.3.2. For each time step, t, the solution for the model parameter is calculated from minimizing the following cost function in Equation 3.3 (Ljung, 1999), 2 1 Equation 3.3 where e(t) is the prediction error at time t. Thus the importance of a data point k sample prior to the present point is discounted exponentially when calculating the model parameters at the present point. Let = [a 1 a 2 a 3 … a p ] be the parameter estimate at time t. can be estimated from 1 and estimation error from the previous step as 1 . Equation 3.4 y(t) is the observed RRI at time t and (t) is the prediction of y(t) based on observation up to time t-1. The gain, K(t), determines how much the prediction error affects the current parameter estimations. The RRI prediction, (t), can be calculated as 1 Equation 3.5 56 where is the regression vector. For an autoregressive model, this regression vector is 12 … Equation 3.6 The gain, K(t) can be calculated as 1 1 Equation 3.7 where 1 1111 . Equation 3.8 The derivation of this solution can be found in Section 11.2 of (Ljung, 1999). The parameter, , calculated from the above steps is designed to minimize the cost function in Equation 3.3 at time t. To improve the conversion time of the estimation, a two-step estimation process was implemented. First the parameter, , was estimated from a short section of data; typically RRI data from the first minute of recording was used for this step. Initial values were set to default values; 0 is zero, θ(0) is all zeros, and P(0) is 10 4 times the identity matrix. The second step uses the final values of , θ, and P from the previous step as initial values, 0 , θ(0) and P(0), for an estimation of the model parameter, , for the whole range of recording. 57 3.3.1. Model Order Selection In the basic TV-AR model for HRV estimation, our algorithm first selected the model order using the Akaike Information Criterion (AIC) (Ljung, 1999; Kay, 1988; Shiavi, 2007). Values of cost function for AIC were computed for a model order (p in Equation 3.1) between 4 and 10. A model order which minimized this cost function (Equation 3.9) was selected for the TV-AR model. AICp N . ln  p 2 .p Equation 3.9 where N is the length of the signal in numbers of samples,  p is the variance of the estimation residuals when the model order = p is used, and AIC is the cost function needed to be minimized for the AIC criterion. The first term on the right of Equation 3.9 is a penalty for higher estimation error while the second term is a penalty for higher model order. Residual from the estimation is considered here as a function of changes in the model order and the forgetting factor. Figure 3.2 shows the normalized mean squared error (NMSE) of the RRI estimates from the TV-AR model with various model orders and forgetting factors from 17 subjects. 58 Figure 3.2 Normalized mean squared error (NMSE) of RRI estimates from the AR model with various model orders and forgetting factors. The graph shows means and standard errors of NMSE from 17 subjects at each value of the forgetting factor and the model order. ff in this figure represents a forgetting factor. It appeared that variation in NMSE increases with the model order in a low forgetting factor case and decreases with model order in a high forgetting factor case. In addition, variation in NMSE with change in model order is minor compared to those with changes in the forgetting factor, which is discussed in the next section. 5 6 7 8 9 10 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34 0.36 model order (p) NMSE ff = 0.86 ff = 0.88 ff = 0.90 ff = 0.92 ff = 0.94 ff = 0.96 ff = 0.98 59 3.3.2. Forgetting Factor Selection One important parameter for the TV-AR model is the forgetting factor, , which ranges from 0 to 1. A high  allows a long past data section for parameter estimation, producing a more robust estimation, but with a tradeoff for a long time-constant. In the present study, a simulated signal (shown in Figure 3.3) was used to determine a  suitable for the data in our study. Figure 3.3 Time-varying spectral analysis of a simulated signal consisting of 2 sinusoidal components and an impulse at t = 0: (A) the simulated signal; (B) time-varying power spectral density (TV PSD) estimation using recursive least square (RLS) algorithm with  = 0.98; (C) TV PSD estimation using RLS algorithm with  = 0.90. Note that the effect of an impulse at t = 0 second lasted over 1 minute in the case of  = 0.98, while the effect lasted for a much shorter period in the case of  = 0.90. time (seconds) -100 0 100 200 X 0 5 10 15 20 1 10 100 1000 10000 -100 0 100 200 0.0 0.1 0.2 0.3 PSDx Time (sec) Frequency (Hz) ff = 0.98 1 10 100 1000 10000 -100 0 100 200 0.0 0.1 0.2 0.3 PSDx Time (sec) Frequency (Hz) ff = 0.90   A BC 60 The simulated signal (Figure 3.3A) composed of 2 sinusoidal components and an impulse at time t = 0. At time t  0, the simulated signal contains a weak 0.1 Hz signal together with a strong 0.3 Hz signal, intended to mimic the frequency contents of RRI variability. A large half-sinusoidal impulsive disturbance, representing a sigh, at time t = 0 was added to the simulated signal. Random noise was also added to the simulated signal. We then applied the TV-AR model to estimate the time-varying PSD of the simulated signal. Different forgetting factors, ranging from  = 0.90 to  = 0.98, were used to test the model. Figure 3.3B and Figure 3.3C show that the model was able to track the changes in frequency components using  at the two ends of the range tested. The estimation appeared to be smoother using  = 0.98 than it was using  = 0.90. Nonetheless, the spillover effect of an impulse at time t = 0 appeared to last over a minute in the case of  = 0.98. The effect of the impulsive disturbance decayed much faster in the case of  = 0.90. However there was a high degree of variability in the estimated time-varying spectra. To determine the appropriate  for the HRV data analysis, we looked at the real RRI responses to sighs from experimental data. One-minute time-invariant PSD of RRI was computed both before and after a spontaneous sigh. We denoted time at peak sigh inspiration to be t = 0. The pre-sigh PSD of RRI was computed using data from t = -80 to 61 -20 seconds; and the post-sigh PSD of RRI was computed using data from t = 20 to 80 seconds. An example of the analysis is shown in Figure 3.4. Figure 3.4 An example of spectral estimations of r-wave to r-wave interval (RRI) from 1-minute section before sigh and from 1-minute section after sigh: (A) time-course of respiratory trace during a sigh; (B) the corresponding time-course of RRI; (C) stationary estimation of power spectral density (PSD) of RRI for 1-minute data section before sigh, time = -80 to -20; (D) stationary estimation of PSD of RRI for 1-minute data section after sigh, time = 20 to 80. Note that the effect of a sigh which spread the power to all frequency ranges does not seem to appear between time = 20 to 80 as shown in D. respiratory (mL) 0 1000 2000 time (seconds) -80 -60 -40 -20 0 20406080 RRI (ms) 800 1000 1200 frequency (Hz) 0.0 0.1 0.2 0.3 0.4 0.5 PSD 1 10 100 1000 10000 RRI,pre sigh frequency (Hz) 0.00.1 0.20.3 0.40.5 PSD 1 10 100 1000 10000 RRI,post sigh A B C D 62 The time-invariant PSD of RRI before and after a sigh appeared to be relatively similar, Figure 3.4C and Figure 3.4D. The change in frequency components of RRI in response to the sigh did not seem to last beyond 20 seconds after the sigh. Therefore, we decided that the TV-AR model with  = 0.98 would respond too slowly to the sigh, and thus  = 0.90 was selected for all our subsequent analyses. 3.4. Time-varying Parameter Estimate Variability Reduction (TV-PEVR) Technique To overcome the problem of high parameter estimate variability with lower values of , we employed the following technique: first we compute a RRI spectral estimate from a conventional TV-AR technique. Then we added a white Gaussian noise (WGN) to the original signal to perturb the system. The power of the additional WGN is equal to the standard deviation of the residuals [e(n) in Equation 3.1]. This new data sequence was then used again as y(n) in Equation 3.1. The parameter estimation procedure was repeated until successive median time-courses of the HRV indices (HFP, LHR) differed by less than 5%. These median time-courses were then used to represent each subject. A diagram showing processing steps for this TV-PEVR method is shown in Figure 3.5. 63 Figure 3.5 Diagram showing the processing steps for the TV-PEVR method A time-course of HFP and LHR calculated from the simulated signal in Figure 3.3 using our TV-PEVR method is shown in Figure 3.6. 64 Figure 3.6 High-frequency power (HFP) and low-frequency to high-frequency ratio (LHR) of the simulated signal from Figure 3.3 calculated with the time-varying parameter estimate variability reduction (TV-PEVR) technique. Plots show LHR estimates from the conventional TV-AR method and our proposed TV-PEVR method at  = 0.98 and at  = 0.90. By using  = 0.90 with our TV-PEVR technique, we were able to achieve both a fast response time and robust estimates of the time-varying HRV spectral indices. 3.5. Adjustment for the Effect of Respiration on HRV Respiration is known to affect HRV. Fluctuation of heart rate increases with increased fluctuation in respiration as shown in Figure 3.7, which shows different patterns of airflow and RRI. The left column was recorded during wakefulness, the right column was recorded during a normal sleep, and the center column was during an arousal HFP (%baseline) 1000 2000 3000 4000 5000 time (seconds) -100 0 100 200 300 LHR (%baseline) 500 1000 1500 2000 ff = 0.98, TV-AR ff = 0.98, TV-PEVR ff = 0.90, TV-AR ff = 0.90, TV-PEVR λ λ λ λ 65 from sleep from an apnea. There is a clear correlation between the degree of fluctuation in airflow and fluctuation in RRI. Figure 3.7 Fluctuation in airflow with the corresponding fluctuation in RRI. (Figure from (Khoo et al., 1999)) During an intervention, the respiration pattern may change drastically, especially if the intervention involves stimulating hypoxia which will activate the chemoreflex leading to an increase in ventilation. The controlled hypoxia maneuver is a clear example of such intervention. We have noticed that SCD patients have higher changes in ventilation in response to hypoxia than in the normal controls (Figure 3.8). 66 Figure 3.8 Changes in minute ventilation from the baseline value as a response to hypoxia in 8 SCD patients (SCD) and 8 normal controls (CTL). Time = 0 indicates the onset of oxygen saturation drop. This could mask the direct effect of hypoxia to changes in HRV. Therefore, we have developed an algorithm that uses simultaneously measured respiration to adjust the heart rate and un-mask the effect of transient hypoxia on HRV. The flow chart of the algorithm is shown in Figure 3.9. Figure 3.9 Flowchart representing the algorithm to adjust for the respiration effect to HRV. time (seconds) 0 50 100 150 Vent (%) -40 -20 0 20 40 60 SCD CTL Respiratory adjusted RRI. y ra (n) Measured RRI, y(n) Corresponding Respiration, x(n) Stable Respiration, x 0 (n) ARX Model, a j , b k Respiratory-uncorrelated Component of RRI. y ru (n) Respiratory-correlated Component of RRI. y rc (n) y' rc (n) Measured data Model computation Result 67 To compute respiratory-adjusted HRV, we first estimated the time-varying transfer function between respiration and RRI. A time-varying autoregressive with exogenous input (TV-ARX) model (Ljung, 1999) was used to describe the relationship between respiration trace (denoted x) and RRI (denoted y): ) ( ) ( ) ( ) ( ) ( ) ( 0 1 n e k n x n b j n y n a n y r k k p j j           Equation 3.10 where p and r represent the orders of the model for the autoregressive and exogenous input parts respectively. a j and b k are the model parameters which were estimated using an RLS algorithm. e(n) represents the residual error of the model estimation at time-point n. Next, we partitioned y into respiration correlated and respiration uncorrelated components, denoted by y rc and y ru respectively. ) ( ) ( ) ( ru rc n y n y n y   Equation 3.11 where ) ( q 1 q ) ( 1 0 rc n x a b n y p j j j r k k k         Equation 3.12 and ) ( q 1 1 ) ( 1 ru n e a n y p j j j      Equation 3.13 68 q -m represents a backward shift of m time-steps. Thus, ) ( q 1 1 ) ( q 1 q ) ( 1 1 0 n e a n x a b n y p j j j p j j j r k k k              Equation 3.14 From Equation 3.14, the following expression for the RRI power spectrum ( ) ( S f y ) was derived: ) ( S ) H( ) ( S ) ( S 2 ru f f f f x y y   Equation 3.15 where H( f ) is the transfer function from respiration to RRI at frequency f, and can be computed from:         p j j f j r k f k k e a e b f 1 T π 2 i 0 T π 2 i 1 ) H( Equation 3.16 In Equation 3.15, S f is the spectrum of the respiration time-series, and S f is the spectrum of the uncorrelated component of HRV. We defined the Respiratory Sinus Arrhythmia Gain (G rsa ) as the average of the magnitude of H(f) in the frequency range from 0.15 to 0.4 Hz.    4 . 0 15 . 0 rsa d ) H( 15 . 0 4 . 0 1 f f G Equation 3.17 69 Equation 3.15 shows quite clearly that the HRV spectrum is influenced directly by the respiration spectrum, S x ( f ), which is determined by the ventilation level and the ventilatory pattern. Thus, changes in the ventilatory pattern alone can lead to changes in HFP and LFP, thereby complicating the interpretation of these indices of HRV as measures of autonomic activity even within the same individual. To overcome this problem, we introduced the notion of “respiration-adjusted” indices in the following way: for each subject, we manually selected a 10-breath segment of data, in which ventilation was relatively uniform and breathing was occurring spontaneously, from the section prior to the start of the hypoxic stimulus and used this segment as the baseline for comparison of subsequent changes. Subsequently, we calculated the power spectral density, S x0 ( f ), of this segment of the respiratory signal, x o (n), and used it to replace S x ( f ), in Equation 3.15, in the subsequent sections of data following the start of the hypoxic stimulus. Therefore, we defined the power spectrum of the adjusted RRI as ) ( S ) H( ) ( S ) ( S 0 2 ru ra f f f f x y y   Equation 3.18 S yra ( f ) was then used to compute the respiration-adjusted high-frequency power, HFP ra , and low-frequency to high-frequency ratio, LHR ra , in the same way the corresponding conventional HRV indices were calculated. 70 These time-varying HRV parameters were then used to compare the ANS response to various stimuli between SCD and control groups. The analyses results are presented in Chapter 5. 71 Chapter 4. Minimal Model for Assessing Cardiovascular Control 4.1. Introduction While the HRV analysis method presented in the last chapter allowed understanding the balance between the sympathetic and the parasympathetic modulations of heart rate, what causes the changes in this balance is yet to be determined. Background about the factors which influence HRV is presented in section 0. In brief, two of the main factors are the blood pressure and the respiratory influence on heart rate. This chapter presents a proposed mathematical model to explore the dynamics of arterial baroreflex (ABR) and respiratory cardiac coupling (RCC), the two primary reflexes which regulate heart rate variability. Mathematical models of physiological systems have been used widely to examine the mechanisms which regulate the measurements of vital signals while reducing the extrinsic effects that are subjected to wide range variation (Duffin et al., 2000; Jo et al., 2003; Chaicharn et al., 2006; Khoo, 2008; Mukkamala et al., 2006). Thus, the results from this model allow us to understand the relationships between respiration and blood pressure to HRV, while reducing the effects of other factors. Our group had previously tested a preliminary version of this model in obstructive sleep apnea patients. We were 72 able to show that this group of patients has a significant reduction in both ABR and RCC gains compared to the normal population (Jo et al., 2003; Khoo, 2008). We have revised this model to allow a better assessment of the ABR and RCC mechanisms mentioned above. In particular, this modification allowed for the removal of the confounding effect of respiration on blood pressure so that a comparison of this mechanism across subjects could be more accurate. We have applied this time-invariant assessment of ABR and RCC mechanisms to compare 5-minute baseline measurements between the SCD patients and normal controls. Moreover, we have extended this model to include a time-varying assessment of the changes in ABR and RCC mechanisms in response to autonomic stimuli. This chapter explains the background and mathematical computation of the model, as well as presents some validations of the model with simulated data. The empirical model results and statistical analysis are presented in the next chapter. 4.2. Model Description The fluctuation in RRI ( ΔRRI) is known to be influenced by fluctuations in blood pressure ( ΔSBP) through the arterial baroreflex, changes in respiration pattern ( ΔResp) through the respiratory-cardiac coupling, and various other inputs (Berne, & Levy, 2001; Katona, & Jih, 1975; Neff et al., 2003; Persson, 1996; Sleight et al., 1995). For modeling purposes we denoted the impulse response that relates the ΔSBP to ΔRRI as h ABR (ABR impulse response), and denoted the impulse response that relates the ΔResp to ΔRRI as 73 h RCC (RCC impulse response). A diagram explaining the structure of this model is shown in Figure 4.1, where ω RR represents the stochastic and other influences on RRI fluctuation which are not explained by the model. Figure 4.1 Diagram of the minimum model describing the fluctuation of RRI During a stable resting condition, the model was assumed to be linear and time- invariant, as characterized by the following equation: ∆ ∆ 1 0 ∑ ∆1 0 Equation 4.1 where T RCC and T ABR are latencies associated with RCC and ABR and M is the length of the impulse responses, i.e. the number of samples used to represent each impulse response. T RCC was assumed a negative number, as studies have shown that the measured change in respiration always follows the measured change in RRI (Angelone, 74 & Coulter, 1964; Saul et al., 1989). This is due to the latency between the vagal firing and the actual onset of respiratory activity. T ABR was assumed a positive number, as the measured change in RRI always follows the measured change in SBP. This is because the variation in blood pressure directly affects the vagal firing rate with some delay (Katona et al., 1970). Note that Figure 4.1 takes a linear time-invariant form, which is suitable when there is no disturbance to the system. In other words, this linear time-invariant computation is suitable for assessing physiological functions during a quiet resting baseline. However, when an autonomic stimulus is applied to a subject, the transfer function that characterizes the behavior of the system would become a function of time, as the stimulus might have altered the physiological condition of the system. Thus, in this situation, we apply a linear, time-varying model to describe the system and track the change of the model characteristics following the stimulus, using the equation: ∆ , ∆1 0 ∑ , ∆1 0 . Equation 4.2 Note that in Equation 4.2 h RCC and h ABR are now functions of time, t, indicating that we allowed the functions to change with time following an autonomic stimulus. 75 4.3. Physiological Information Extracted from the Model While we can assess the dynamics of the relationships between ∆Resp and ∆RRI, and between ∆SBP and ∆RRI through h RCC and h ABR , descriptors from these impulse responses are needed for comparing these dynamics among subject groups, as well as among different physiological conditions. We consider the peak points of the impulse responses, the negative peak for h RCC and the positive peak for h ABR , as the gains. Figure 4.2 shows an example of the parameters extracted from an impulse response (ABR or RCC). Figure 4.2 An example of the parameters extracted from an impulse response (h ABR or h RCC ). Positive peak and negative peak of each impulse response were recorded from each subject for group comparisons. To understand the frequency responses of the ABR and the RCC, we converted both impulse responses to their frequency domain function, i.e. impulse responses were converted to their transfer functions using an AR method. We extracted the average gains time (seconds) 05 10 15 ABR or RCC impulse response -200 -150 -100 -50 0 50 100 Positive peak Negative peak 76 in low frequency (0.04 – 0.15 Hz) and high frequency (0.15 – 0.40 Hz) ranges from the RCC and ABR transfer functions. Since the RCC mechanism is mainly modulated by the parasympathetic input and the ABR mechanism is mainly modulated by the sympathetic pathways (ESC/NASPE, 1996), we expected the average gains in their corresponding frequency ranges to be more sensitive to perturbation. From each transfer function, 3 parameters - LF gain, HF gain and overall (LF + HF) gain - were extracted for statistical comparisons. Figure 4.3 An example of the parameters extracted from a transfer function (ABR or RCC). Low-frequency (LF) gain, high-frequency (HF) gain, and overall gain were recorded for each impulse response for group comparisons. For a time-invariant analysis of a baseline data section, there are only one set of the parameters mentioned above (positive peak, negative peak, LF gain, HF gain, and overall gain) for each dataset. We generally selected a 5-minute quiet section from an experiment to represent each subject. For the time-varying analysis, we tracked the changes in all the aforementioned parameters as a response to an autonomic stimulus. frequency (Hz) 0.0 0.1 0.2 0.3 0.4 0.5 ABR or RCC transfer function 0 5 10 15 20 LF HF Overall 77 Thus, for each time point (corresponding to every 0.5 seconds for 2 Hz sampled data) we computed an updated set of parameters. 4.4. Orthogonal Expansion of Impulse Responses To directly compute the impulse responses, h RCC and h ABR , all points represented in the impulse responses need to be calculated. Thus, an important attribute of an impulse response is a discrete-time impulse response length or the “memory” of the kernel, N. This number, N, is determined from the ratio of the effective kernel memory to the sampling interval, T. We have noticed from our previous study (Chaicharn et al., 2008) that both impulse responses, h RCC and h ABR , have a significant value only up to approximately 25 seconds. With a sampling rate of 2 Hz or sampling period, T = 0.5 seconds, the ratio of the effective kernel memory to the sampling interval is 25/0.5 = 50 samples. Therefore, we selected N = 50 for all impulse response estimations in this study. Consequently, 50 values would need to be estimated for each impulse response (100 values for the combination of both ABR and RCC impulse responses). The number of values that need to be estimated is very large compared to the input data points. In particular, a 5-minute baseline section of ∆RRI, ∆SBP, ∆Resp at 2 Hz sampling rate consists of 600 data point for each parameter, or 1800 data point total. Using these only 1800 empirical data points to estimate the 100 values for the 2 impulse responses could result in a noisy estimation. 78 To overcome this problem, Wiener has proposed a method to truncate the modeling problem by using an expansion of the kernel through orthogonalization of the Volterra series for a WGN input (Wiener, 1958). This technique, which significantly reduces the number of parameter to be estimated to practically less than 20 parameters, has been adopt widely and modified by various groups (Jo et al., 2003; Marmarelis, 1993; Brinker, 1995; Ogura, 1985). The impulse responses were estimated as a sum of several weighted kernels (Marmarelis, 1997). Therefore for each impulse response (ABR or RCC), the number of parameters required to represent the impulse response was reduced to only the number of orthogonal basis functions. Figure 4.4 shows a diagram representing the structure of this kernel-based estimation method. Figure 4.4 Diagram representing the impulse response estimation method using weighted kernels. a n and b n are the parameters needed to be estimated. * ΔResp * * * Basis Func. . . . ΔRRI · a 1 + · b 4 · b 3 · b 2 · b 1 · a 4 · a 3 · a 2 . . . . . . * ΔSBP * * * Basis Func. . . . kernels RLS algorithm for estimation of the coefficients 79 Selecting the number of basis functions used in an impulse response is essential; a too low number of basis functions could lead to a low accuracy in the estimation, while a too a high number could cause the estimation to be subjected to noise. The suitable number depends on the property of the system to be estimated (e.g. frequency response, signal to noise ratio, and prior knowledge of the shape of the impulse response). In the present study, we decided to use between 4 to 10 basis functions to represent an impulse response. This allowed us a decent accuracy and an acceptable level of estimation noise. The expansion-of-kernels technique used in this study is based on the Volterra- Weiner theory of nonlinear systems (Marmarelis, 2004). The discrete-time Volterra models explain the relationship between one-input and one-output signals as: 0 1 2 2 1 , 2 1 2 2 1 Equation 4.3 where y denotes the output signal, x denotes the input signal, n represents the discrete-time index (n = t/T), T is the sampling interval, and m, m 1 , and m 2 denote the discrete-time lag (Marmarelis, 2004). A previous study by our group showed that the system depicted in Figure 4.1 can be explained with a reasonable goodness-of-fit by a linear equation (Jo et al., 2007). Moreover, for a relatively noisy short section of input, adding higher-order components reduces the degree of freedom, leading to a noisy model description (Marmarelis, 2004; 80 Jo et al., 2007), especially when the model is extended to a time-varying version. Therefore, we decided to describe our system with a linear mathematic model. In another words, we assume that the systems can be explained by their 1 st -order impulse responses. Another benefit of using only the 1 st -order part is that the physiological interpretation is more apparent. For a linear system, only the first 2 components on the right side of Equation 4.3 are considered. Consequently, for our 2-input, 1-output system, the output y can be expressed as a function of inputs x and u as: ∑ 1 0 ∑1 0 Equation 4.4 where N is the “memory” of the kernel, D x and D u are the lags between inputs x and u to output y respectively, and (n) is the estimation residual. Note that we assumed the constant term (k 0 in Equation 4.3) to be zero. This is because in our study we have subtracted from every signal its average value. Therefore both inputs and output are zero means; thus, there is no offset between the inputs and the output. Although we have employed different orthogonal basis functions which are more suitable for modeling of systems in our studies than Wiener’s; namely Laguerre’s and Meixner’s basis functions, both have been developed under considerable influence from Wiener’s ideas. Thus, in the present study, the kernels were generated from convolutions of the inputs and Laguerre or Meixner orthonormal basis functions through a Least 81 Square estimator (Lin et al., 2004; Chaicharn et al., 2006; Asyali, & Juusola, 2005). The detail of the generation of the Laguerre’s and Meixner’s function is presented in the next section. 4.4.1. Laguerre and Meixner expansion of kernels To date, the Laguerre expansion technique is one of the most widely adopted techniques for estimating Volterra kernel (Marmarelis, 2004). Applications of the “discrete Laguerre functions” (DLFs) in physiological system modeling have been discussed in various publications (Marmarelis, 1993; Jo et al., 2007; Blasi et al., 2006; Parker et al., 1999). The orthonormal set of DLFs was described by (Ogura, 1985) which was later described in detail in (Marmarelis, 2004). /2 1 1/2 ∑ 1 0 Equation 4.5 where b j (m) denotes the j th -order orthonomal DLF, the integer m ranges from 0 to M-1 (M = length of the system), and the real positive number  (0 <  < 1) is a critical DLF parameter that determines the rate of exponential (asymptotic) decline of these functions. The most common way to implement these functions is to use a recursive method (Jo et al., 2007). To initialize the recursive computation, we need to compute the 0 th -order DLF, b 0 (m), as: 1 . Equation 4.6 82 Then for each of the higher-order DLF we first compute the value of the function at time zero: 0 √ 1 0,0. Equation 4.7 For all j > 0 and m > 0, can be calculated recursively as: √1 √ 1 1 1 . Equation 4.8 Note that Equation 4.7 is actually a special case of Equation 4.8 when the value of b j (m) for all m < 0 is zero. The parameter  was selected in a way that all Laguerre function decline to a value sufficiently closes to zero when m is equaled to system length, M. An example set of Laguerre basis functions is shown in Figure 4.5. The higher the value of , the longer the time period needed for the Laguerre basis functions to converge to zero. Figure 4.5 The first four discrete-time Laguerre functions from the definition in (Marmarelis, 2004) for  = 0.4. System length, M = 50 0 10 20 30 40 50 -0.5 0 0.5 1 time index, m Laguerre functions j = 0 j = 1 j = 2 j = 3 83 Besides using the recursive method explained above, the Laguerre function can be generated using an IIR filter (Brinker, 1995). Note that the definition of the Laguerre function given by Brinker (1995) is slightly different from that described in (Marmarelis, 2004; Jo et al., 2007). The main difference is that the Laguerre basis functions with an odd order, i.e. j = 1, 3, 5, …, are multiplied with -1. The discrete Laguerre functions are given in (Brinker, 1995) as: 1 , Equation 4.9 where p = √ was referred to by Brinker (1995) as the Laguerre parameter and 1 / . An example set of Laguerre basis functions where  = 0.4 or p = √ = 0.6325 is shown in Figure 4.6. Note the differences between these basis functions and those in Figure 4.6. Figure 4.6 The first four discrete-time Laguerre functions from the definition in (Brinker, 1995) for  = 0.4 or p = √ = 0.6325. System length, M = 50 0 10 20 30 40 50 -0.5 0 0.5 1 time index, m Laguerre functions j = 0 j = 1 j = 2 j = 3 84 Note that for the all odd numbers of j, the functions in Figure 4.6 are flipped versions of the similar functions in Figure 4.6. Nonetheless, the Laguerre functions in both figures are orthogonal basis functions and can be used for expansions of kernels. The z- transform of the Laguerre functions in Equation 4.9 was giving by Brinker (1995) as: 1 2 1 2 1 Equation 4.10 From this equation, Laguerre functions can be constructed from an IIR filter as shown below. Figure 4.7 The cascade filter structure to generate Laguerre functions (Adopted from Brinker (1995). Laguerre functions with an order from 0 to j are generated from this structure. We also used Meixner-like basis functions for impulse response estimation. The advantage of these functions over the Laguerre’s function is the slower initial rising of the basis functions, which make them more suitable to estimate slow physiological system responses compared to the Laguerre’s function. From the Laguerre’s functions, a Meixner-like function can be generated simply by an orthogonal transformation (Brinker, 1 1 1 1p δm impulse 85 1995). A set of Laguerre functions [b 0 (m) b 1 (m) b 2 (m)… b j (m)], can be transformed to a set of orthogonal Meixner-like functions, M, by multiplying with an orthogonal transformation matrix, A (n) . Equation 4.11 Figure 4.8 shows the structure to generate the Meixner-like orthogonal functions. Figure 4.8 The cascade filter structure to generate Meixner functions (Adopted from (Brinker, 1995) with modification). Meixner-like functions with an order from 0 to k and “Order of generalization of n” are generated from this structure. A (n) is the orthogonal matrix that transforms j Laguerre basis functions to k Meixner-like basis functions. Properties of the transformation matrix A (n) were described in detail by Brinker (1995). Here we have summarized the steps to generate Meixner-like functions from Laguerre functions. Let, , Equation 4.12 and U is an upper-band matrix with a j × j dimension: 1 1 1 1p δm impulse 86 1 0 01 00 1 …0 …0 …0 00 0 …1 , Equation 4.13 where j is the number of Laguerre basis functions needed for a generation of k Meixner basis functions with an “order of generalization” equaled to n. Since one of the properties of the transformation matrix A (n) is that it generates the z-transform of the Meixner-like function, G (n) (z) with an order of n+k+1 for both poles and zeros. Therefore we need j = n+k+1 Laguerre functions to generate that Meixner-like function. And for the matrix to be squarable, we first generated U to have a dimension of j × j where j = n+k+1. Since A (n) is an orthogonal matrix, we have T T T . Equation 4.14 From the above equation, we can calculate L (n) from inverting the Cholesky decomposition of U n {U n } T . L chol T 1 , Equation 4.15 where chol(Y) denotes a Cholesky factorization of matrix Y, giving an upper triangular matrix R which satisfies the equation {R T }R = Y. 87 Then A (n) can be calculated from Equation 4.12. Only the first k rows of A (n) are used for the transformation of j Laguerre functions of k Meixner-like functions in Equation 4.11. An example set of Meixner-like basis functions where  = 0.4 or p = √ = 0.6325, and n = 4 is shown in Figure 4.9. Figure 4.9 The first four discrete-time Meixner-like functions from the definition in (Brinker, 1995) for  = 0.4 or p = √ = 0.6325. “Order of generalization” n = 8. System length, M = 50 The advantage of these Meixner-like functions in representing an impulse response of a physiological system is that these functions start rising from a value close to zero and reach their maximum values at time index > 0. Since these basis functions appear more similar to a physiological impulse response which normally has some delay time before reaching its maximum value, typically a lower number of Meixner-like functions than the number of Laguerre functions is needed to describe a physiological impulse response with the same degree of accuracy (Brinker, 1995). 0 10 20 30 40 50 -0.4 -0.2 0 0.2 0.4 time index, m Meixner-like functions j = 0 j = 1 j = 2 j = 3 88 The latency between the beginning of the basis function (time index, m = 0) to the peak of the basis functions can be adjusted with the “order of generalization” n. When n = 0, Meixner functions become Laguerre functions which have the exponential decay component only in the basis functions. The latency is increased with an increasing value of n. In other words, the larger the value of n, the longer it takes for the basis functions to reach their maximum values. 4.4.2. Model Estimation and Optimization To apply the Laguerre and Meixner-like functions to estimate the impulse responses h x and h u from Equation 4.4, we first convolved a set of Laguerre or Meixner- like functions ( and ) with the inputs x and u (Jo et al., 2007). ∑ 1 0 Equation 4.16 ∑ 1 0 Equation 4.17 The output y(n) can then be written as a function of v j (n) and w j (n) as: ∑ 1 0 ∑ 2 0 Equation 4.18 where S1 and S2 are the numbers of Laguerre or Meixner-like functions used for inputs x and u respectively. The coefficients · and · can be estimated by a least-square fitting where the time series and are inputs and the time series is the output. This therefore reduces the estimation problem to an estimation of S1+S2+2 89 parameters, instead of estimating every point on the 2 impulse responses. This reduction in the degree of freedom or in the number of parameters to be estimated leads to smoother and more robust estimated impulse responses. To select the number of basis functions {S1, S2}, the delays {D x , D u }, and the orders of generalizations {n x , n u } that best fit the data without adding too much redundancy to the computation, we computed an estimation of y(n) from Equation 4.18 for all combinations of {S1, S2}ranging from 1 to 7, D x (RSA delay) ranging from -2 to 1 seconds, Du (ABR delay) ranging from 0.5 to 2 seconds, and {n x , n u } ranging from 0 (Laguerre functions) to 6. The “optimal model” was then selected for the combination which resulted in the minimum value of the minimum description length (MDL) (Barron et al., 1998). 4.5. Linear Time-invariant Model Parameter Estimation To estimate the linear time-invariant transfer functions, h ABR and h RCC , from a 5- minute baseline data section, instead of using Equation 4.1directly as a one step process, we introduced a 3-step model parameter estimation procedure to improve the accuracy of the model prediction. This technique allowed an adjustment of the confounding effect of respiration on blood pressure when calculating the ABR impulse response, as well ass the confounding effect of blood pressure on respiration when calculating the RCC impulse response. The details of this estimation process, shown in Figure 4.10, are described in the next section. 90 Figure 4.10 Procedure to get the estimation of h RCC and h ABR 4.5.1. 3-step estimation of the time-invariant model parameters The 3-step computation technique with one pre-processing step aforementioned is explained below. Pre-processing: We first low-pass filtered the RRI, Systolic blood pressure (SBP), and respiration signal (Resp) signals to remove spikes or high-frequency noise. An finite impulse response (FIR) filter with 0.55 Hz cut-off frequency was used to remove the high-frequency noise, as the frequency of interest for a study of the ANS is lower than 0.5 Hz. We also removed low-frequency trends from all the data by using polynomial fitting as described in (Lin et al., 2004). Each signal would then have a mean of zero. Thus we denoted each resulting signal as ΔRRI, ΔSBP, and ΔResp respectively. Step 1: Removal of the respiration correlated part of blood pressure from the original blood pressure, followed by the use of a 2-input-1-output model to obtain the 91 intermediate impulse responses, h ABR-temp and h RCC-temp . As the two inputs mentioned above (Resp and SBP) were highly coupled with each other, we first de-correlated them using an autoregressive with exogenous input (ARX) algorithm. This step is described by the “ ΔSBP and ΔResp decorrelation” box in the diagram. Equation 4.19 shows an ARX model of ΔSBP which describes ΔSBP at current time, n, as a function of past ΔSBP data as well as past (and current) values of ΔResp data. ∆ ∆ ∆0 1 Equation 4.19 The residual error term in Equation 4.19, e(n), represents that part of ΔSBP(n) that cannot be accounted for by the ARX model. The unknown ARX model parameters, a i (i = 1,…, p) and b k (k = 1,…, q), were estimated using a least-square minimization. Therefore, ΔSBP could be decomposed into two additive components: a respiratory-correlated component ( ΔSBP rc ) and a respiratory-uncorrelated component ( ΔSBP ru ); Equation 4.20 where ∆ ∑ ∆0 Equation 4.21 is the component of ΔSBP which can be explained by respiration. By rearranging Equation 4.19 through Equation 4.21, the respiratory uncorrelated part of SBP was calculated using the following equation: 92 ∆ ∆ ∆0 Equation 4.22 ΔSBP ru was then used for our first step of 2-input 1-output model parameter estimations. The relationship between the inputs and output for step 1 are as follows: ∆ ∆ 1 0 ∆ 1 0 Equation 4.23. The impulse responses, h RCC-temp and h ABR-temp in this case, were then estimated as the sum of several weighted kernels as described in section 4.4. Step 2: Removal of the blood pressure-induced part of RRI from the original RRI, followed by the use of a 1-input-1-output model to obtain h RCC . To get an accurate estimation of h RCC , we removed the contribution of ΔSBP to ΔRRI using the ABR impulse response calculated from the 1 st step (h ABR-temp ). The blood pressure uncorrelated part of ΔRRI, denoted as ΔRRI bu , could be calculated from Equation 4.24. ∆ ∆ ∆1 0 Equation 4.24 We then calculated the final solution of the RCC impulse response (h RCC ) from ΔResp and ΔRRI bu as follows: 93 ∆ ∆ 1 0 Equation 4.25 From Equation 4.25, h RCC was estimated as the sum of several weighted kernels which were generated from convolutions of the input ΔResp and Laguerre or Meixner-like orthonormal basis functions through a Least Square estimator. This was done in a fashion similar to that described in step 1, except that only 1 input was used in this step. Step 3: Removal of the respiration-induced part of RRI from the original RRI, followed by the use of the 1-input-1-output model to get h ABR . This step is similar to the previous ones except that the parameter to be calculated is now for h ABR . The same techniques were applied in this step but instead of using Equation 4.25, Equation 4.26 was used for the calculation of the final ABR impulse response (h ABR ). The respiratory non-correlated part of ΔRRI, denoted as ΔRRI bu , was calculated from Equation 4.26 and later used for model parameter estimation in Equation 4.27. ∆ t ∆RRI h RCC i ∆Resp t i T RCC M1 i0 Equation 4.26 ∆ ∆ 1 0 Equation 4.27 4.5.2. Time-invariant model testing To test the model, we generated a set of ABR and RCC impulse responses to be similar to those in Figure 4.11 and convolved the impulse responses with a 5-minute 94 section of ΔSBP and ΔResp respectively. The summation of the 2 products was denoted as ΔRRI for model testing. The derived impulse responses are shown in Figure 4.12 for all steps 1, 2 and 3. Figure 4.11 Respiratory-cardiac coupling impulse response (h RCC ) and arterial baroreflex impulse response (h ABR ) used for model testing 0 5 10 15 20 25 -150 -100 -50 0 50 h RCC time (seconds) 0 5 10 15 20 25 0 2 4 6 8 h ABR time (seconds) 95 Figure 4.12 ABR and RCC impulse calculated from all 3 steps as detailed above. A simulated signal generated from real ∆Resp and ∆SBP data, and the impulse responses shown in Figure 4.11. From these recovered impulse responses, we concluded that our linear time- invariant model could be used for estimation of the ABR and RCC impulse responses accurately. 4.6. Linear Time-varying Model Parameter Estimation The model estimation steps described in the previous sections were used only when we assumed both h ABR and h RCC to be linear, time-invariant, such as while the subject was at rest. However when a stimulus, such as hypoxia, cold face, or deep breath, was applied, we assumed that these impulse responses can change their characteristic 0 5 10 15 20 25 -50 0 50 RCCtemp (step1) RCC 0 5 10 15 20 25 0 2 4 6 8 ABRtemp (step1) ABR 0 5 10 15 20 25 -150 -100 -50 0 50 RCC (step2) time (seconds) 0 5 10 15 20 25 0 2 4 6 8 10 ABR (step3) time (seconds) 96 over. Therefore, we used a time-varying version of this model with the data acquired during a stimulus. To compute the time-varying model descriptors, we first used a linear, time-invariant technique described in the previous sections to identify the model structure as well as to estimate the latencies between each input and the output and the model coefficients. 4.6.1. Computation method for a time-varying model parameters estimation To estimate the time-varying impulse responses, we applied both the model structure and the input-output latencies acquired from the time-invariant step. In the time- varying model, however, the model coefficients are updated at every time point of the data (every 0.5 seconds). One important parameter for the time-varying system modeling is the forgetting factor, λ, which can range between 0 to 1 (Ljung, 1999; Blasi et al., 2003). When λ = 0, only the current data is used to describe the estimation error. On the other hand, when λ = 1, both current data and all the past data is used. When λ is small the system can adapt quickly but has more tendencies to be unstable (Ljung, 1999). For the purpose of our project, we use a least-square algorithm to select λ between 0.8 to 0.99 that results in the smallest model prediction error. 4.6.2. Time-varying model testing To test the time-varying minimal model mentioned above, we first generate a set of test impulse responses as shown in Figure 4.13. 97 Figure 4.13 Arterial baroreflex (ABR) impulse response and respiratory-rardiac roupling (RCC) impulse response used for testing of the time-varying minimal model. Solid lines represent true impulse responses for time t < 0; dashed lines represent true impulse responses for time t > 0. At time t = 0, both ABR and RCC impulse responses change their shapes abruptly from the solid lines to the dashed line in Figure 4.13. Figure 4.14C and Figure 4.14D show 3- dimentional plots of these impulse responses. Then we generated 2 test inputs (SBP and Resp) from 2 WGN sequences (Figure 4.14A and Figure 4.14B). These impulse responses and test inputs were constructed in a way that their magnitudes and powers are in the same range as experimental results from previous publication from our group (Jo et al., 2003; Khoo, 2008). These two input signals were convolved with time-varying ABR and RCC impulse responses as the filters. Their outputs were summed and added with a WGN with power equaled to 10% of the signal power (Figure 4.14E). lag (seconds) -5 0 5 10 15 20 25 ABR impuse response -4 -2 0 2 4 6 8 10 lag (seconds) -5 0 5 10 15 20 25 RCC impuse response -200 -150 -100 -50 0 50 100 (A) (B) 98 Figure 4.14 Diagram describing test signal generating steps. (see text for a description) time (seconds) -60 -40 -20 0 20 40 60 simulated SBP -10 -5 0 5 10 time (seconds) -60 -40 -20 0 20 40 60 simulated Resp -0.4 -0.2 0.0 0.2 0.4 * -2 0 2 4 6 8 10 0 5 10 15 20 2 -60 -40 -20 0 20 40 ABR Impulse Response lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 0 5 10 15 20 2 -60 -40 -20 0 20 40 RCC Impulse Response lag (seconds) time (seconds) + time (seconds) -60 -40 -20 0 20 40 60 simulated RRI -150 -100 -50 0 50 100 150 10% WGN 0 0 A) B) C) D) * E) 99 Time-varying system identification was performed using Matlab® and tested with the simulated inputs and output described above. The estimations of time-varying ABR and RCC impulse responses are shown in Figure 4.15. Figure 4.15 Time-varying arterial baroreflex (ABR) impulse response (A) and respiratory- cardiac coupling (RCC) impulse response (B) computed from the simulated data. Figure 4.15 shows that our algorithm was able to recover both impulse responses reasonably. Some degrees of delay in the changing of impulse responses was observed at time t = 0 when the impulse responses change abruptly. This delay was due to the forgetting factor, . In this particular example  = 0.95 was selected, as this value of  resulted in the smallest estimation error. A smaller  would lead to a faster response time, however, with a less robust estimation of impulse responses and, in this case, less estimation accuracy. -2 0 2 4 6 8 10 -5 0 5 10 15 20 25 -60 -40 -20 0 20 40 ABR Impulse Response (computed) lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 -5 0 5 10 15 20 25 -60 -40 -20 0 20 40 RCC Impulse Response (computed) lag (seconds) time (seconds) A) B) 100 We also performed more tests on various simulated impulse response and found the recovery impulse responses to track the ideal impulse responses (Figure 4.16 and Figure 4.17). Figure 4.16 An example of time-varying impulse responses estimation from a simulated dataset. (A) ideal Arterial Baroreflex (ABR) impulse response; (B) ideal Respiratory-Cardiac Coupling (RCC) impulse response; (C) computed ABR impulse response; (D) computed RCC impulse responses. -2 0 2 4 6 8 10 -5 0 5 10 15 20 25 30 -60 -40 -20 0 20 40 ABR Impulse Response (computed) lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 -5 0 5 10 15 20 25 -60 -40 -20 0 20 40 RCC Impulse Response (computed) lag (seconds) time (seconds) -2 0 2 4 6 8 10 0 5 10 15 20 25 -60 -40 -20 0 20 40 ABR Impulse Response (ideal) lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 0 5 10 15 20 2 -60 -40 -20 0 20 40 RCC Impulse Response (ideal) lag (seconds) time (seconds) A) B) C) D) 101 Figure 4.17 An example of time-varying impulse responses estimation from a simulated dataset. (A) ideal Arterial Baroreflex (ABR) impulse response; (B) ideal Respiratory-Cardiac Coupling (RCC) impulse response; (C) computed ABR impulse response; (D) computed RCC impulse responses. In summary, this chapter presents a concept and computational method of a minimal model for assessing the control of heart rate, through the ABR and RCC mechanisms. We tested the program designed to compute the model in both time- invariant and time-varying modes with simulated data mimicking real impulse responses were used for this test. The results verified that our program work well and were able to recover the simulated impulse responses with a good accuracy. Empirical model results using the method presented in this chapter is presented in Chapter 5. -2 0 2 4 6 8 10 -5 0 5 10 15 20 25 30 -60 -40 -20 0 20 40 ABR Impulse Response (computed) lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 -5 0 5 10 15 20 25 -60 -40 -20 0 20 40 RCC Impulse Response (computed) lag (seconds) time (seconds) -2 0 2 4 6 8 10 0 5 10 15 20 25 -60 -40 -20 0 20 40 ABR Impulse Response (ideal) lag (seconds) time (seconds) -200 -150 -100 -50 0 50 100 0 5 10 15 20 2 -60 -40 -20 0 20 40 RCC Impulse Response (ideal) lag (seconds) time (seconds) A) B) C) D) 102 Chapter 5. Results and Discussion This chapter presents the results of an assessment of the microvascular perfusion and the ANS in SCD patients. In this study, we tested the autonomic responses to 3 autonomic stimuli, which were hypoxia, spontaneous sighs, and cold face stimulus. To assess the ANS responses to these stimuli, the ECG data from all experiments were analyzed using an HRV technique (for more detail see Chapter 3). Our data shows some degrees of abnormality in ANS in SCD patients that may be related to the decrease of microvascular perfusion and subsequently their sickle cell crises. In addition to monitoring the effect of autonomic stimuli to ANS responses in SCD and CTL subjects, the effects of transfusion on physiological and HRV measures are also evaluated and presented in this chapter. Although no significant improvement on HRV measures was detected post-transfusion from the 5-minute recordings, we did find an improvement in RRI and oxygen delivery parameters, suggesting that transfusion reduces the stress on the patients’ cardiovascular system. In addition to the perfusion data and the results from the ANS assessment using the HRV technique, we also present results from a model-based technique to assess reflex mechanisms involved in cardiovascular autonomic control (see Chapter 4 for more detail on the model computation). These model parameters measure the state of two major reflexes that control fluctuations in heart rate: the ABR and the RCC mechanisms. Here, 103 we present the results from the evaluation of the model parameters at baseline as well as the changes of these parameters in response to the cold face stimulus. This chapter concludes with a presentation of relationships between the ANS response and some data from the laboratory blood tests. Measurements from blood related to SCD diagnosis were regressed with HRV changes in response to hypoxia. The existence of these relationships gives us an insight on how the abnormalities in ANS system in SCD patients might evolve, and how these abnormalities might be related to the severity of SCD. 5.1. Baseline Physiological and Autonomic Measurements in SCD patients We first compared the physiological and the autonomic measurements between the SCD and the CTL subjects at their stable baselines, while the subjects rested in a supine position. Five-minute baseline measurements of heart rate and respiration parameters (i.e. minute ventilation, respiration rate, and inspired tidal volume) were not significantly different between the SCD group and the CTL group (P > 0.05, rank-sum test, Table 5.1). Baseline values of SaO 2 were significantly lower in SCD patients than in CTL subjects (median = 95.9 vs. 97.9 in SCD vs. CTL, P = 0.017, rank-sum test, Table 5.1). The levels of SaO 2 are usually lower in SCD patients relative to the levels of SaO 2 in normal individuals. This difference is due in part to the significant left shift in the oxyhemoglobin saturation curve (Becklake et al., 1955). Note that the pO 2 derived from 104 the SaO 2 using the oxyhemoglobin saturation curves were not significantly different between the 2 groups of subjects, indicating that the differences in SaO 2 were primarily the result of the differences in the oxygen binding properties of HbS and HbA. Another parameter which showed a statistical significant difference (p<0.05) between the 2 groups is regional oxygen saturation, rSO 2 , which measures local oxygen extraction in the area where transducer was located. We found that the rSO 2 are lower in SCD group compared to the normal control when the transducers was placed on subjects’ left forehead and left tibia (Table 5.1). Five-minute baseline HRV indices did not differ between the SCD and the CTL groups (Table 5.2). Note that, due to the limited duration of each experiment, the baseline HRV indices for each subject in this study were computed from a short baseline segment (5-minute). This is a relatively short data segment compared to what was published in the 1996 ESC/NASPE Taskforce which is considered the standard of stationary baseline HRV computation (ESC/NASPE, 1996). The 1996 ESC/NASPE Taskforce suggested using an average of measurements taken overnight for such computations in order to compare baseline HRV indices across subjects. Thus, our HRV measurements might be easily confounded. 105 Table 5.1 Physiological measurements during 5-minute baseline. Data shows median (25 percentile – 75 percentile) for each parameter. p-values were calculated from rank-sum test. SCD CTL p-value Vent (L/min) 10.8 (8.0 – 13.9) 7.8 (5.1 – 11.3) 0.203 Br/M 16.5 (14.6 – 17.6) 16.9 (13.7 – 18.9) 0.969 ViVol (mL) 602.7 (494.2 – 902.1) 503.5 (332.2 – 623.4) 0.298 RRI 0.83 (0.77 – 0.91) 0.83 (0.79 – 0.91) 0.681 SaO 2 95.9 (94.5 – 97.2) 97.9 (97.8 – 98.0) 0.017 pO 2 81.6 (72.2 – 88.3) 88.4 (85.6 – 95.2) 0.222 PU 20.1 (18.0 – 27.9) 22.5 (19.9 – 30.9) 0.431 Vel 41.0 (33.7 – 58.7) 59.9 (41.9 – 95.2) 0.076 CMBC 51.8 (33.3 – 73.1) 41.4 (25.1 – 70.5) 0.555 TB 4.419 (4.212 – 5.812) 5.286 (4.079 – 6.728) 0.555 rSO 2 (CH1, right forehead) 47.7 (44.6 – 67.4) 65.2 (57.3 – 70.3) 0.295 rSO 2 (CH2, left forehead) 54.0 (45.0 – 58.7) 65.1 (59.2 – 68.4) 0.035 rSO 2 (CH3, left arm) 48.3 (36.6 – 55.3) 53.6 (49.2 – 61.3) 0.336 rSO 2 (CH4, left tibia) 52.0 (43.6 – 59.8) 65.8 (63.9 – 71.9) 0.021 106 Table 5.2 HRV measurements during 5-minute baseline. Data shows median (25 percentile – 75 percentile) for each parameter. p-values were calculated from rank-sum test. SCD CTL p-value VLFP (ms 2 ) 963.4 (252.3 – 1127.4) 495.0 (340.1 – 807.9) 0.763 LFP (ms 2 ) 971.0 (304.9 – 1421.5) 828.2 (591.8 – 949.1) 0.681 HFP (ms 2 ) 1914.5 (1038.7 – 3777.3) 1498.2 (226.5 – 2325.0) 0.311 LHR 0.306 (0.168 – 0.738) 0.812 (0.342 – 1.702) 0.095 Total power (ms 2 ) 3884.5 (2683.2 – 7233.9) 3227.9 (1654.0 – 4812.1) 0.529 nLFP 0.22 (0.14 – 0.40) 0.43 (0.25 – 0.60) 0.095 nHFP 0.72 (0.54 – 0.84) 0.53 (0.56 – 0.72) 0.106 5.2. Autonomic Nervous System Response to Hypoxia in SCD Patients It is well known that hypoxia induces polymerization of HbS in SCD patients. Furthermore, the degree of nighttime hypoxia predicts stroke and the frequency of vaso- occlusive crises in SCD patients (Kirkham et al., 2001; Hargrave et al., 2003). In order to study these cardiovascular responses, we devised a protocol to induce transient hypoxia, similar to that which occurs naturally during sleep. We measured the oxygen saturation, electrocardiogram signal, tidal volume, and peripheral perfusion throughout the experiments. Experimental details can be found in Chapter 2. In brief, after the subject 107 has stabilized breathing room air for at least 5 minutes, we switched the subject’s breathing air content from breathing room air to 100 percent N 2 for 5 breaths and then switched back to room air. This entire process was completed without the subject being aware of the change. 5.2.1. Results 5.2.1.1. Safety of the Hypoxia Protocol Approximately 16% of all subjects were able to tell when they were hypoxic through symptoms such as headache, lightheadedness, or shortness of breath (Table 5.3). Subjects reported that these symptoms lasted no longer than 30 to 60 seconds and resolved when the hypoxia resolved. Four subjects reported pain, which they identified as mild vaso-occlusive pain within 24 hours after the experiment; none of the subjects visited the clinic or the ER for the pain. No serious adverse event was reported. Given this data, we concluded that this 5-breath N 2 protocol was safe. 108 Table 5.3 Safety data from 19 CTL experiments and 33 SCD experiments. Each experiment included 2 exposures to 5 breaths of N 2 . VOC = vaso-occlusive crisis; HA = headache; SOB = shortness of breath; ACS = acute chest syndrome; TIA = transient ischemic attack. Number of subjects who experienced the symptom time elapsed following experiment 1 hr 12 hr 24 hr 1 wk subject groups CTL SCD CTL SCD CTL SCD CTL SCD adverse event VOC Mild 0 0 0 0 0 4 0 4 Moderate 0 0 0 0 0 0 0 1 HA Mild 3 0 0 1 1 0 0 5 Lightheaded Mild 3 6 0 3 0 1 0 2 SOB Mild 1 5 1 0 0 0 0 1 ACS 0 0 0 0 0 0 0 0 TIA 0 0 0 0 0 0 0 0 Stroke 0 0 0 0 0 0 0 0 Cough Mild 0 0 0 0 0 0 1 3 Fever Mild 0 0 0 0 0 0 1 0 Moderate 0 0 0 0 0 0 0 1 Nausea and Vomiting Mild 0 0 0 0 0 0 0 1 Diarrhea 0 0 0 0 0 0 0 0 109 5.2.1.2. Physiological responses to transient hypoxia Time-courses of SaO 2 responses to hypoxia were computed for 11 SCD subjects and 13 CTL subjects. Hypoxemia was observed in both groups of subjects through decreases in SaO 2 . One of the 14 CTL subjects was excluded from this part of the study since a decrease in SaO 2 to a level lower than 95% was not detected in this subject. The decreases in SaO 2 were more pronounced in the SCD group than in the CTL group, as shown in Figure 5.1A. Nonetheless, after we converted SaO 2 to pO 2 , the degree of hypoxemia appeared to be more pronounced in the CTL group. The differences in maximum decreases in SaO 2 (in raw units) between subject groups were significant (t-test, Table 5.4). However, when the decreases were compared in terms of per cent changes from baseline or in terms of pO 2 , these subject group differences were no longer significant. The inset in Figure 5.1A shows the pO 2 responses to hypoxia stimulus in a subgroup of subjects whose blood oxygen dissociation curves were measured. Figure 5.1B shows perfusion responses to transient hypoxia. Contrary to our initial hypothesis, there seemed to be no change in microvascular perfusion in response to hypoxia in either subject group. 110 Figure 5.1 Blood oxygen, perfusion, and heart rate responses to hypoxia stimulus in SCD and in CTL. (A) Changes in oxygen saturation (SaO 2 ) from baselines following hypoxia stimulus from 11 SCD and 13 CTL subjects. (A inset) changes in partial pressure of oxygen (pO 2 ) from baselines following hypoxia stimulus from 5 SCD and 5 CTL subjects. (B) Changes in peripheral perfusion (PU) from baselines following hypoxia stimulus from 11 SCD and 10 CTL subjects. (C) Changes in r-wave to r-wave interval (RRI, inverse of heart rate) from baselines following hypoxia stimulus from 11 SCD and 13 CTL subjects. Time t = 0 indicates peak of hypoxia as observed by SaO 2 drop in each subject. Numbers of subjects in each graph vary due to availability of data. Each plot shows group mean and standard error. PU -40 -20 0 20 40 time (seconds) 0 100 200 300 RRI -10 -5 0 5 SaO -25 -20 -15 -10 -5 0 5 SCD CTL time (seconds) 0 100 200 300 pO (%baseline) 40 60 80 100 120 A) B) 2 2 C) (% change from baseline) (% change from baseline) (% change from baseline) 111 Table 5.4 Comparison of SaO 2 and PO 2 drops after hypoxia exposure. The values are shown in mean (SD). Data from subjects whose blood oxygen dissociation curves were measured. Parameter SCD (n=5) CTL (n=5) P-value SaO 2 value 71.61 (9.07) 82.69 (4.11) 0.038 SaO 2 (% baseline) -25.31 (9.43) -15.98 (4.61) 0.082 pO 2 value 42.21 (7.33) 46.84 (6.63) 0.325 pO 2 (% baseline) -42.66 (9.73) -51.10 (8.32) 0.370 While we did not see perfusion changes, increases in heart rate, corresponding to decreases in RRI, were observed in both SCD patients and CTL subjects following the transient hypoxia stimulus. This tachycardia response to hypoxia was more pronounced and lasted longer in SCD patients than the tachycardia response in CTL subjects (Figure 5.1C). 112 5.2.1.3. Autonomic nervous system responses to transient hypoxia Because changes in heart rate and peripheral perfusion are controlled by the modulations of sympathetic and parasympathetic divisions of the ANS, we computed surrogate measurements of these modulations through analysis of HRV as shown in Figure 5.2. A rank-sum test was also performed to compare the area under the curve of each parameter between t = -15 and 15 seconds, when t = 0 indicates nadir of SaO 2 following the hypoxia stimulus as measured by pulse-oxymeter, as shown in Figure 5.1. Figure 5.2 Time courses of heart rate variability (HRV) indices of SCD patients vs. CTL subjects: (A) high frequency power (HFP); (B) low-frequency to high-frequency ratio (LHR); (C) respiratory-adjusted HFP (HFP ra ); and (D) respiratory-adjusted LHR (LHR ra ). Time t = 0 indicates the peak of hypoxia as observed by SaO2 drop in each subject. All parameters are displayed in percentage (%) change of the parameters from their baseline values. time (seconds) 0 100 200 300 LHR (%baseline) 50 100 150 200 250 300 time (seconds) 0 100 200 300 HFP (%baseline) 50 100 150 200 SCA (n = 8) CTL (n = 11) time (seconds) 0 100 200 300 HFP (%baseline) 50 100 150 time (seconds) 0 100 200 300 LHR (%baseline) 50 100 150 200 ra ra A) B) C) D) 113 Figure 5.3 Heart rate variability responses during hypoxia. (A) Time-course of high frequency power (HFP) in response to hypoxia stimulus, when t = 0 indicate peak of hypoxia as measured by pulse-oxymeter. The average values between t = -15 and 30 second (shaded area) were used for group comparison. (B) Comparisons of HRV responses between the SCD group and the CTL group. Each bar shows the mean with standard error of area under the curve of each parameter between t = -15 and 30 seconds, when t = 0 indicates peak of hypoxia as measured by pulse- oxymeter. HFP = high frequency power, LHR = low-frequency to high-frequency ratio, HFP ra = respiratory-adjusted HFP, LHR ra = respiratory-adjusted LHR, and G rsa = respiratory sinus arrhythmia gain. * indicates p<0.05 from baseline. + indicates p<0.05 between groups. The symbol above a bar indicates a significant difference of the parameter from its own baseline. The symbol above a bracket indicates a significant difference of the parameter between the two subject groups. Analysis of HRV showed marked reductions in HFP following the hypoxia stimulus in the SCD patients, but not in the CTL subjects (Figure 5.2A). Reduction in HFP indicated a loss of parasympathetic modulation, which could result in tachycardia. HFP LHR HFPra LHRra Grsa % Change -50 0 50 100 Baseline Level SCD (11) CTL (13) time (seconds) 0 100 200 300 HFP (%Change) -50 0 50 SCD (11) CTL (13) ra ra rsa Baseline Level A) B) t = 0 at peak of hypoxia * * * * + + + 114 To compensate for the effect of respiration changes during hypoxia, we computed another set of the HRV variables using an algorithm to adjust for the effect of respiration to HRV. After adjusting for the effects of respiration patterns on HFP, the decreases in HFP ra following hypoxia remained present in the SCD patients (Figure 5.2C). G rsa , which represents the strength of the effect of respiratory pattern on heart rate, also decreased in SCD patients in response to hypoxia (Figure 5.3). Decreases in both HFP ra and G rsa suggested that the loss in parasympathetic modulations in the SCD patients following the hypoxia stimuli may be a direct effect of hypoxia on HRV regardless of changes in respiration patterns. On the other hand, sympathetic modulation, represented by LHR, appeared to be unaffected by hypoxia, or at least no effect was observable immediately after the stimulus (Figure 5.2B). However, after the adjustment for the effect of respiration patterns on LHR, increases in LHR ra were apparent in both the SCD and the CTL groups (Figure 5.2D); nonetheless, the levels of increase in LHR ra were not statistically significant (Figure 5.3). Just as a decrease in parasympathetic modulation contributes to tachecardia, an increase in sympathetic modulation also contributes to tachycardia. These data indicate a significant abnormality in autonomic nervous system regulation in patients with sickle cell anemia in response to transient hypoxia. 115 5.2.2. Discussion Our results demonstrated a clear causal relationship between transient hypoxia and marked reductions in the parasympathetic modulation and possible increases in the sympathovagal balance in the SCD patients, whereas no significant changes were noted in the normal controls. Furthermore, our technique to adjust for the effect of the ventilation pattern to HRV made the changes in both parasympathetic and sympathetic indices following the hypoxia stimuli in the SCD subject much more apparent, making study of non-respiratory evoked autonomic changes possible. These findings suggest that vagal tone is easily reduced in subjects with SCD following transient exposure to hypoxia. These responses could be responsible for the substantial increases in heart rate observed in this cohort after the hypoxic stimuli. This finding is also consistent with a report from Pearson et al. (2005) that children with greater parasympathetic withdrawal during challenges showed more clinical severity of sickle cell disease. This hypersensitivity of HRV response to hypoxia in SCD patients could be the result of a compensatory mechanism designed to increase oxygen delivery to tissues to counterbalance for their chronic hypoxemia. As shown in Table 5.1, SaO 2 baselines were lower in SCD patients and normal controls, suggesting that mild chronic hypoxemia occurs in this group of patients. Researchers have shown some degree of physiological cardiovascular adaptation in SCD patients, including elevations in heart rate and declines in cardiac output, primarily resulted from the associated chronic hemolytic anemia 116 (Lester et al., 1990; Batra et al., 2002). Furthermore, Weiskopf et al. reported linear increases in heart rates in healthy unmedicated human subjects as a result of acute isovolemic anemia (Weiskopf et al., 2003). Thus, in SCD patients, hypoxia-induced sickling of erythrocytes, which leads to a higher degree of anemia, could have caused the further increases in response sensitivity. In normal subjects, peripheral chemoreceptor stimulation by acute hypoxia activates both ventilatory and cardiovascular responses (Shoemaker, 2004). Increases in sympathetic activity was reported in rats submitted to chronic intermittent hypoxia (Zoccal et al., 2007) and in sleep apnea patients (Jo et al., 2003; Somers et al., 1995), who also suffer from chronic intermittent hypoxia. The sickling process occurs continually in subjects with SCD, resulting in sub-clinical vaso-occlusion. This occlusion results in chronic regional hypoxia. We speculate that the chronic sub-clinical vaso-occlusion may affect the sensitivity of chemoreceptors in SCD patients in a manner analogous to the effects of chronic intermittent hypoxia observed in animal models and sleep apnea patients. Acute hypoxia has been shown to enhance the sympathetic drive while decreasing parasympathetic activity, leading to an increase in LHR, the ratio representing sympathovagal balance (Buchheit et al., 2004; Iwasaki et al., 2006). Although this view has been widely accepted, the time-course of the changes remains ambiguous. Most studies in this area have been focused on the steady-state effect of the hypoxia response 117 during a longer duration of hypoxia (> 1 minute). Much shorter periods of hypoxia (5 breaths) were used in our experiments, out of safety concerns for the SCD subjects. This duration of exposure was clearly insufficient to induce significant changes in HRV in our normal control subjects but appeared sufficiently strong to evoke autonomic responses in SCD patients. The signal processing technique for HRV employed herein allowed us to directly study the non-respiratory-derived components of autonomic function that, we believe, may be important in the fundamental pathology of SCD. The premise that HRV can provide useful indices of cardiac autonomic control is based largely on the original study of (Katona, & Jih, 1975), which demonstrated in an animal preparation a linear relationship between respiratory-related fluctuations in RRI and vagal firing rates. Their latter result was obtained under conditions in which respiration was relatively well- controlled. Subsequent validation studies in humans employing pharmacological interventions to alter autonomic tone were also conducted under conditions in which respiration was controlled (Berntson et al., 1997). Although some studies have underscored the important effect of respiratory rate and tidal volume on the HRV spectrum (Grossman et al., 1991; Brown et al., 1993), this confounding influence of respiration has been largely ignored in the HRV literature. Brown et al. (1993) showed that, depending on breathing frequency, changes in respiration within a given individual can substantially alter estimates of both high-frequency and low-frequency power of the RRI spectrum. Our present study highlights the importance of correcting the autonomic 118 indices derived from HRV for the respiratory changes that accompany the brief exposure to hypoxia. The approach we have introduced of “computationally correcting” for these respiratory-related distortions of the HRV spectrum is analogous to the well-accepted statistical technique of “adjusting” for confounding variables. While this study was limited by the relatively small number of subjects, the preliminary findings we had reported indicate quite clearly that there were significant differences in autonomic reactivity to transient hypoxia in SCD subjects relative to normals. In particular, SCD subjects displayed a significant reduction in RRI and the parasympathetic indices of HRV following exposure to 5 breaths of N 2 , in contrast with normal controls who demonstrated little or no response. Furthermore, the signal processing technique employed here allows us to study the non-respiratory derived components of autonomic dysfunction, which we suspected may be important in the fundamental pathology of SCD. Given that otherwise unexplained sudden death accounts for up to 40% of the mortality in SCD and that autonomic dysregulation has clearly been associated with sudden death in other settings (Hathaway et al., 1998; Sathyaprabha et al., 2006; Piepoli, & Capucci, 2007), application of the experimental and analytical methodology introduced in the current study may yield important insights into the pathophysiology and perhaps clinical management of sickle cell disease. 119 5.3. Autonomic Nervous System Response to Sighs in SCD Patients Whereas the protocol developed for this study was originally aimed to assess the between hypoxia and the development of vaso-occlusive crises, we did not find an apparent change in PU directly associated with the drop in oxygen saturation as described in the previous section. Nonetheless, we did note periodic episodes of hypo-perfusion that were much more prominent in the SCD subjects than in the normal controls. These frequent sharp perfusion drops seemed to be independent of the occurrences of hypoxia and seemed to be associated with spontaneous sighs. Figure 5.4 shows an example of measurements recorded during a hypoxia stimulus, comprising multiple perfusion drops following the sighs. The perfusion drops following sighs appeared in some normal controls as well. However, these sharp perfusion drops seemed to occur more frequently in the SCD group than in the CTL group (see Figure 5.5 for an example). 120 Figure 5.4 An example of measurements recorded during a hypoxic episode in SCD patients. (A) Oxygen saturation as measured by a pulse oxymeter (SaO 2 ). (B) Peripheral perfusion as measured by Laser Doppler; perfusion unit (PU) is represented as an arbitrary unit (au). (C) Respiratory trace (Resp). (C) r-wave to r-wave interval (RRI). Figure 5.5 Examples of perfusion drops following spontaneous sighs (A) in SCD patients and (B) in normal controls. SaO2 (%) 60 70 80 90 100 PU (au) 0 5 10 15 Resp (mL) 0 1000 2000 3000 time (seconds) -100 0 100 200 300 RRI (sec) 0.8 time (seconds) 0 100 200 300 400 500 600 Respiration Perfusion A time (seconds) 0 100 200 300 400 500 600 Respiration Perfusion B 121 5.3.1. Results 5.3.1.1. Relationship between sighs and decreases in PU To statistically identify the association between sighs and perfusion drops, we analyzed the occurrences of spontaneous sigh-induced perfusion drops in 11 SCD and 11 CTL subjects. A significant relationship (p<.001 for 18 subjects and <.03 for one subject) between perfusion drops and sighs was observed in 19 out of 22 subjects (11/11 SCD and 8/11 CTL). No sighs were detected in 2 control subjects; therefore, the chi- square statistical analysis could not be performed in these 2 subjects. The frequencies of sighs (percentage of breaths quantified as sighs from all breaths) were not statistically different between the two groups (p=0.694, median=2.10% vs. 2.05% in SCD vs. CTL, Figure 5.6A). However, the median probability of a sigh being immediately followed by a perfusion drop was much higher in the SCD group than in the CTL group (77.8% vs. 16.7%, p<.001, Figure 5.6B). Thus, a spontaneous sigh is much more likely to result in a transient hypoperfusion in SCD patients than in CTL subjects. 122 Figure 5.6 Comparison of sigh frequency and likelihood of PU drop for each sigh in SCD patients vs. in CTL subjects. (A) Frequency of breaths classified as sighs; (B) probability of PU drop for each spontaneous sigh. Bar graphs show the mean + SD. P-values were calculated from a rank-sum test. 5.3.1.2. Time-course of HRV responses to sighs Decreases in RRI, or rises in heart rate, were observed during sighs. The levels of RRI drops did not differ between the SCD group and the CTL group. Compared to the rates of decrease in RRI following transient hypoxia stimuli (Figure 5.1), the rates of decrease in RRI following sighs (Figure 5.7) appeared to be significantly higher for both groups of subjects. 123 Figure 5.7 Changes from baselines of r-wave to r-wave interval (RRI) following spontaneous sighs in SCD and CTL groups. Time t = 0 indicates peak of sigh inspiration for each subject. Each plot shows group medians of RRI and either 75th or 25th percentiles. In both groups of subjects, there was no difference in the levels of changes in RRI between sighs that caused perfusion drops and sighs that did not. Consequently, we consolidated all spontaneous sighs from each subject, regardless of the PU drop responses, in all subsequent computations of RRI. Both CTL and SCD groups showed a post-sigh increase in HFP and HFP ra (Figure 5.8A and Figure 5.8C), but there was no significant difference in responses between the groups. Respiratory sinus arrhythmia gain (G rsa ), which is the average gain from respiration to RRI in the high frequency range, was not affected by sighs (Figure 5.8B), suggesting that the HFP increase was not simply the result of an increase in respiration but a real increase in parasympathetic activity. time (seconds) -20 0 20 40 60 RRI (%baseline) 85 90 95 100 105 110 SCD CTL 124 Figure 5.8 Time-courses of parasympathetic indices of heart rate variability (HRV) in SCD and CTL groups: (A) high frequency power (HFP); (B) respiratory sinus arrhythmia gain (G rsa ); (C) respiratory-adjusted HFP (HFP ra ). Time t = 0 indicates the beginning of a sigh inspiration. Each plot shows group medians of RRI with either 75 th or 25 th percentiles. Sympathovagal balance, represented by LHR, in response to sighs appeared to be higher in the SCD group than in the CTL group (Figure 5.9A). The differences between groups disappeared after adjusting for the effect of respiration, as shown by LHR ra (Figure 5.9B). However, in both groups, LHR ra showed small post-sigh decreases. HFP (%baseline) 0 100 200 300 400 500 SCD CTL Grsa (%baseline) 0 40 80 120 time (seconds) -20 0 20 40 60 HFPra (%baseline) 0 100 200 300 A B C 125 Figure 5.9 Time-courses of sympathetic indices of heart rate variability (HRV) in SCD and CTL groups: (A) low-frequency to high-frequency ratio (LHR); (B) respiratory-adjusted LHR (LHRra). Time t = 0 indicates the beginning of a sigh inspiration. Each plot shows group medians of RRI with either 75th or 25th percentiles. In order to compare the HRV responses to sighs between the two subject groups, we employed an area under the curve of each parameter between t = 5 and 20 seconds, when t = 0 indicated the beginning of sigh inspiration, as a surrogate measurement of each HRV parameter. Statistical analysis showed an increase in parasympathetic modulations (HFP in Figure 5.10). This increase remained present in both groups of subjects after adjustment for the effects of respiration pattern, as shown by HFP ra . No A B LHR (%baseline) 0 200 400 600 SCD CTL time (seconds) -20 0 20 40 60 LHRra (%baseline) 0 50 100 150 200 126 significant difference between SCD and CTL groups in any parasympathetic modulation parameters was detected. The increase in LHR was more pronounced in SCD patients than in CTL subjects. However this increase vanished in both groups after adjustment for the effects of respiration pattern (as shown by LHR ra in Figure 5.10), suggesting that the increase in sympathetic modulation is mainly due to the change in respiration pattern from the sigh. Figure 5.10 Heart rate variability responses during sighs. Each bar shows the mean with standard error of area under the curve of each parameter between t = 5 and 20 seconds, when t = 0 indicate the beginning of sigh inspiration. HFP = high frequency power, LHR = low-frequency to high- frequency ratio, HFP ra = respiratory-adjusted HFP, LHR ra = respiratory-adjusted LHR, and G rsa = respiratory sinus arrhythmia gain. Each bar shows the mean with standard error. * and + indicates P<0.05. The symbol above a bar indicates a significant difference of the parameter from its own baseline. The symbol above a bracket indicates a significant difference of the parameter between the two subject groups. 127 5.3.1.3. Relationship between occurrences of vasoconstriction following sighs and HRV For each subject, the chance of a perfusion drop after a sigh was regressed with HRV variables: both the baseline HRV values and the HRV responses to spontaneous sighs (Figure 5.11). The RRI response to sighs showed a statistically significant correlation with the chance of post-sigh perfusion drop. This suggested that autonomic modulation, as measured by variation in heart rate, might play a role in the transmission of sighs to vaso-constriction. 128 Figure 5.11 Linear regression plots between chances of perfusion (PU) drop if a spontaneous sigh occurs (in percent) and the heart rate variability parameters. (A) r-wave to r-wave interval (RRI); (B) high-frequency power (HFP); (C) low-frequency to high-frequency ratio (LHR); (D) respiratory sinus arrhythmia gain (Grsa). Each response to sighs represents the average value of the parameters between time t = 5 to 20 seconds, while t = 0 indicates the beginning of a sigh inspiration. Blue circles indicate control subjects (CTL) and red triangles indicate sickle cell disease patients (SCD). Data obtained from 9 CTL and 8 SCD subjects. % chance of PU drop if sigh 0 20406080 100120 RRI response (% baseline) 95 100 105 110 115 % chance of PU drop if sigh 0 20406080 100120 RRI baseline (seconds) 0.6 0.7 0.8 0.9 1.0 1.1 CTL SCD y = 98.82 + 0.12 x R2 = 0.41, p = 0.064 y = 93.69 + 0.11 x R2 = 0.65, p = 0.016 y = 0.88 - 0.0004 x R2 = 0.016, p = 0.742 y = 0.84 - 0.0002 x R2 = 0.0026, p = 0.904 % chance of PU drop if sigh 0 204060 80 100120 HFP baseline (second^2) 0.000 0.002 0.004 0.006 % chance of PU drop if sigh 0 20406080 100120 HFP response (% baseline) 0 100 200 300 400 500 600 % chance of PU drop if sigh 0 20406080 100120 LHR baseline 0.0 0.4 0.8 1.2 1.6 % chance of PU drop if sigh 0 20406080 100120 LHR response (% baseline) 0 100 200 300 400 500 600 % chance of PU drop if sigh 0 20406080 100120 Grsa baseline (second/liter) 0 5e-4 1e-3 2e-3 2e-3 % chance of PU drop if sigh 0 20406080 100120 Grsa response (% baseline) 40 60 80 100 120 140 y = 0.0018 - 0.00 x R2 = 0.025, p = 0.687 y = 0.0024 - 0.00 x R2 = 2.91e-4, p = 0.968 y = 203.78 + 2.79 x R2 = 0.363, p = 0.086 y = 172.02 + 0.32 x R2 = 0.013, p = 0.788 y = 0.397 + 0.0037 x R2 = 0.0837, p = 0.45 y = 0.233 + 0.0009 x R2 = 0.007, p = 0.8439 y = 222.95 + 1.127 x R2 = 0.025, p = 0.708 y = 134.08 + 0.491 x R2 = 0.021, p = 0.708 y = (-0.417 + 0.058 x) e-4 R2 = 0.1903, p = 0.28 y = (0.828 - 0.009 x) e-3 R2 = 0.164, p = 0.28 y = 77.33 - 0.2003 x R2 = 0.0767, p = 0.5067 y = 80.40+ 0.573 x R2 = 0.3178, p = 0.1139 129 5.3.2. Discussion Decreases in perfusion after sighs were observed in both groups of subjects. Because not all sighs were associated with peripheral perfusion drops, we speculated that the mechanism of the sigh-initiated PU drop responses could be nonlinear and involve thresholding. Nonetheless, co-occurrence of sighs with perfusion drops was statistically more frequent in the SCD subjects than in the controls. We think that this frequent reduction in perfusion with sighs in SCD patients, under some conditions, may be sufficient to initiate vaso-occlusive events. Peripheral perfusion is controlled by the vascular smooth muscles, which are innervated by peripheral sympathetic nerves. Neural activity in these nerves has been shown to be partly influenced by respiration (Eckberg et al., 1985), through multiple mechanisms. One of which is the hypoxic stress, which have been shown to increase the sensitivity to hyperventilation- sympathetic nerve response (Somers et al., 1989). It is widely known that patients with SCD are constantly in some state of hypoxemia (Hargrave et al., 2003). Consequently, we speculate that these factors may account for the hypersensitivity of the post-sigh peripheral vasoconstriction reflex in SCD patients. Analysis of HRV without adjusting for the effect of respiration suggested that SCD subjects have stronger sympathetic responses and weaker parasympathetic responses to sighs than the CTL subjects. However, after adjusting for respiration patterns, the differences in both HFP ra and LHR ra between the two groups subsided, 130 suggesting that the differences in ANS responses to sighs from the normals in the SCD group were due to their post-sigh respiratory responses. These differences in respiration patterns between groups is likely to be due to multiple factors, which still needed to be determined. 5.4. Autonomic Control Changes with Transfusion Protocol 5.4.1. Results Blood transfusion is one of the major treatments for SCD. To better understand the effect of chronic transfusions on the ANS, we calculated the 5-minute average baseline values of all physiological measurements and HRV parameters. Three baseline physiological parameters, which are RRI, concentration of moving blood cells (CMBC), and rSO 2 on both hands showed a statistically significant changes (p<0.05, paired t-test) post-transfusion compared to pre-transfusion (Figure 5.12). Other physiological measurements and all the HRV parameters showed no statistically significant improvement after the transfusion (Table 5.5 and Table 5.6), which could be because the duration of our baseline recordings were relatively short (5 minutes). Note that this is an on-going study; so far only data from 9 subjects have been analyzed. 131 Figure 5.12 Baseline physiological parameters which show significant changes post-transfusion compared to the pre-transfusion. RRI increased 0.0610 seconds (P = 0.022). Concentration of moving blood cells (CMBC) increased 20.161 au (P = 0.040). Regional O 2 saturation rSO 2 on the hand with blood pressure cuff (rSO 2 a) increased 3.686 % (P = 0.042). rSO 2 on the hand without blood pressure cuff (rSO 2 b) increased 5.302 % (P = 0.023). (paired t-test, N = 9) 132 Table 5.5 Differences in baseline parameters related to the autonomic control system (excluding HRV parameters) pre- and post-transfusion. Parameter Difference (after – before perfusion) Mean ±SD p-value R-wave to R-wave interval 0.0610±0.0646 0.022* Systolic blood pressure -10.886±32.593 0.346 Mean arterial pressure -9.706±21.928 0.221 Diastolic blood pressure -10.357±18.361 0.129 Cardiac output 0.125±1.482 0.807 Stroke volume 8.143±19.448 0.245 Systemic vascular resistance -375.235±545.243 0.073 † Left ventricular ejection time 0.00784± 0.0275 0.417 Peripheral perfusion 52.692±227.259 0.506 Blood velocity -7.448±34.236 0.532 Concentration of moving blood cells 20.161±24.749 0.040* Total backscatter 0.177±1.768 0.771 O 2 saturation 0.390±1.492 0.456 Regional O 2 saturation ipsi-latteral side 3.686±4.578 0.042* Regional O 2 saturation contra-latteral side 5.302±5.681 0.023* * indicates a significant difference pre- and post-transfusion (paired t-test, P < 0.05). † indicates a marginally significant difference pre- and post-transfusion (paired t-test, 0.05 < P < 0.10). 133 Table 5.6 Differences in baseline HRV parameters between pre- and post-transfusion. Parameter Different (after – before transfusion) Mean ±SD p-value LFP (ms 2 ) -170.3 ± 788.2 0.535 HFP (ms 2 ) 371.5 ± 731.7 0.166 LHR -1.377±3.016 0.208 log 10 (HFP) 0.484 ± 0.468 0.015* log 10 (LHR) -0.204 ± 0.428 0.190 From this preliminary data, only log 10 of baseline HFP increased after the transfusions, indicating that transfusion increased the parasympathetic modulation in this group of subject. Nonetheless other baseline heart rate variability measurements did not differ pre- and post-transfusion. The autonomic control stimuli may be needed in order to see an improvement from the transfusion. 5.4.2. Discussion Physiological recordings from SCD patients before and after their chronic transfusions suggest an improvement in the patients’ hemodynamic condition (through lower heart rate, increased concentration of moving blood cells, increased local oxygen saturation), which is to be expected after transfusion. This improved hemodynamic condition might have led to an improved parasympathetic function in SDC subjects shown by HFP. LHR did not change after the transfusion suggesting that the sympathetic 134 system did not improve with transfusion. More accurate HRV analyses of sleep studies are still warranted. 5.5. Autonomic Nervous System Response to Cold Face Test To further evaluate the ANS in SCD patients, we employed the cold face test, CFT, which is one of the most widely used stimuli for assessing the ANS. The protocol for the CFT experiment was explained in detail in Chapter 2. In brief, we placed a cold pack (a bag of ice at zero degrees Celsius) on each subject’s forehead for 1 minute after he/she had rested, and all vital signs were stable. The subjects who participated in this study included a wide range of phenotypes from normal controls, sickle cell traits, SCD patients who were on the chronic transfusion program, and SCD patients who were not. At the time of this publication, this study was still on-going; we were actively recruiting subjects and conducting experiments at CHLA. So far, we had analyzed datasets from a total of 36 subjects, including 14 African- American control subjects, 15 SCD patients who were not in CHLA’s chronic transfusion program, and 7 SCD patients who were. Note that the SCD patients in the non- chronically transfused group might have been transfused, but none of them had had any blood transfusion within a 3-month prior to the experiment. 135 5.5.1. Results During the application of the CFT, the control subjects did not change their ventilation (Figure 5.13), agreeing with the results from (Stemper et al., 2002). However, the non-chronically transfused SCD patients hyperventilated during the application of a cold pack, i.e. their ViVol increased significantly from their baseline levels. The chronically transfused patients had an increased ViVol as well, but it did not reach a statistically significant level, likely due to their higher within group variability. Figure 5.13 Changes in inspired tidal volume (ViVol) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time- course of ViVol changes from baseline. (B) Median of ViVol changes from baseline during CFT and recovery periods. Each plot shows group medians of ViVol with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05. time (seconds) -100 -50 0 50 100 150 200 ViVol (%baseline) -50 0 50 100 150 CTL (n=14) nonTransfused (n=15) Transfused (n=7) Control Transfused nonTransfused ViVol (%baseline) -20 0 20 40 60 80 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * A) B) 136 During the CFT, we also observed that all subject groups demonstrated decreased regional O 2 saturation, as shown by the level of rSO 2 recorded from their arms, to approximately 10% from baselines (Figure 5.14A). This decreased rSO 2 is likely the effect of the diving reflex, which reduces oxygen consumption in the periphery by decreasing oxygen delivery through vasoconstriction (Heath, & Downey, 1990; Heistad et al., 1968; Stemper et al., 2002). This reduced blood flow may result in decreased oxygen levels. Figure 5.14B shows that the decrease in rSO 2 in response to the CFT was statistically significant only in the CTL group. In the 2 SCD groups, there was a high variability in this rSO 2 response; no rSO 2 change even occurs in some patients. We suspected that this absence of response in a large number of SCD subjects may be due to their already low baseline rSO 2 levels from continual hypoxemia. Therefore, the same level of minor vasoconstriction might not affect the already obstructed blood vessels in SCD patients as strongly as the normal controls. 137 Figure 5.14 Hand Oxygen Saturation (rSO 2 ) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time-course of rSO 2 changes from baseline. (B) Median of rSO 2 changes from baseline during CFT and recovery periods. Each plot shows group medians of rSO 2 with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05 All 3 groups of subjects showed decreases in peripheral perfusion, to approximately 50% of the baseline level within 10 seconds after the cold pack was placed (Figure 5.15A). After that first 10 seconds, peripheral perfusions started to return to the baseline levels even though the ice pack remained on. In addition to the perfusion drop mentioned above, the two SCD groups showed another marked perfusion drop right after the cold pack was removed (at time = 60 seconds). A) B) time (seconds) -100 -50 0 50 100 150 200 rSO 2 (%baseline) -20 -15 -10 -5 0 5 10 CTL (n=9) non Transfused (n=14) Transfused (n=5) Control Transfused nonTransfused rSO 2 (%baseline) -10 -5 0 5 10 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * 138 Figure 5.15B shows the decreases in peripheral perfusions in all 3 groups, even though a statistically significant difference from the baseline was observed only in the control group. No significant difference from baseline was observed in the 2 SCD groups, likely due to their high within group variability. In all subject groups, the peripheral perfusion returned to the baseline level within 1 minute after the removal of the ice pack (t = 120 seconds). Therefore, the levels of perfusion observed during t = 180 to 240 seconds from placement of the ice pack showed no difference from the baseline. We suspect that the second drop of perfusion, occurring right after the ice pack was removed (Figure 5.15A), might be a response to hyper-ventilation in the two SCD groups (Figure 5.13). This hyper-ventilation in the SCD subjects may have activated the deep-breath-vasoconstriction reflex, which we have shown to be more sensitive in SCD subjects than normal controls (Figure 5.6). 139 Figure 5.15 Peripheral Perfusion (PU) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time-course of PU changes from baseline. (B) Median of PU changes from baseline during CFT and recovery periods. Each plot shows group medians of PU with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05 Other than peripheral perfusion, which relates closely to peripheral resistance (increased peripheral resistance causes decreased peripheral perfusion), we also continuously recorded the systemic vascular resistance (SVR) in a sub-group of the subjects, using Nexfin® blood pressure monitoring system. In all subject groups, the SVR increased to approximately 20% from the baseline levels within about 50 seconds time (seconds) -100 -50 0 50 100 150 200 PU (%baseline) -80 -40 0 40 80 CTL (n=14) nonTransfused (n=15) Transfused (n=7) A) B) Control Transfused nonTransfused PU (%baseline) -60 -40 -20 0 20 40 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * 140 into the cold pack application, then returned to the baselines levels within about 20 seconds after the cold pack was removed (Figure 5.16A). The time-courses of SVR did not differ among the subject groups. Unlike the second peripheral perfusion drop in the SCD subjects (Figure 5.15), we did not observe a second spike of SVR. This indicates that right after the cold pack was removed, there was a marked increase in vascular resistance in the peripheral circulation but not in the systemic ones. Figure 5.16 Changes in systemic vascular resistance (SVR) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time- course of SVR changes from baseline. (B) Median of SVR changes from baseline during CFT and recovery periods. Each plot shows group medians of SVR with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05 A) B) time (seconds) -100 -50 0 50 100 150 200 SVR (%baseline) -20 0 20 40 60 CTL (n=9) nonTransfused (n=9) Transfused (n=6) Control Transfused nonTransfused SVR (%baseline) -10 0 10 20 30 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * 141 Systolic blood pressure, SBP, increased slightly (to about 3% above baseline) but significantly (p<0.05) in the control subjects during the CFT, but not in the SCD groups (Figure 5.17). The increase in SBP agreed with data from (Stemper et al., 2002) who suggested that it results from the vasoconstriction response to the CFT, increasing vascular resistances and, thus, increasing blood pressure. In both SCD groups, we observed high variations within the groups, in accordance with the high variations we also observed in the SVR (Figure 5.16). Figure 5.17 Changes in systolic blood pressure (SBP) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time- course of SBP changes from baseline. (B) Median of SBP changes from baseline during CFT and recovery periods. Each plot shows group medians of SBP with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05. time (seconds) -100 -50 0 50 100 150 200 SBP (%baseline) -15 -10 -5 0 5 10 15 CTL (n=9) nonTransfused (n=9) Transfused (n=6) A) B) Control Transfused nonTransfused SBP (%baseline) -10 0 10 20 30 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * 142 Upon an application of the cold pack, the most considered ANS response to CFT is its bradycardia. In all 3 subject groups, their RRI decreased (tachycardia) briefly prior to the onset of the cold face stimulus (Figure 5.18A). Then the bradycardia response to the cold face stimulus started as soon as the cold pack was applied. The maximum RRI response to the cold face stimulus was reached around 20-30 seconds after the onset of the stimulus. The levels of changes in RRI following cold face stimulus did not differ among the 3 subject groups (Figure 5.18B). Note that we used only non-parametric statistical analyses (rank-sum test and ANOVA on ranks) and presented plots of medians for all cold face responses. This was because we found that raw data recorded from cold face sessions were subject to various confounding effects, which could have differ among subjects. For example, some subjects might have had tachycardia responses to anticipation to the test, or some subjects might have had autonomic responses to pain from the cold. We also noticed that some subjects attempted to talk or were moving during the relatively long stimulus period. Thus we decided to use the group medians to represent the subject groups, in order to remove the various confounding effects mentioned above. 143 Figure 5.18 Changes in r-wave to r-wave interval (RRI) from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. Time t = 0 indicates the starting time of the 60-second cold face stimulus. (A) Time- course of RRI changes from baseline. (B) Median of RRI changes from baseline during CFT and recovery periods. Each plot shows group medians of RRI with either 75 th or 25 th percentiles. * indicates a significant difference from baseline, P<0.05. In addition to basic physiological measures, HRV measures were continuously computed to assess the ANS response to cold face stimulus. The plots of time-courses of HRV variables during the cold face stimulus showed an increase in HFP (Figure 5.19A), which became more pronounced when we adjusted for the effect of respiration ( ∆HFP ra , Figure 5.19B). This concurs with literature, which shows the activation of the time (seconds) -100 -50 0 50 100 150 200 RRI (%baseline) -10 -5 0 5 10 15 CTL (n=14) nonTransfused (n =15) Transfused (n=7) Control Transfused nonTransfused RRI (%baseline) -10 -5 0 5 10 15 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) * A) B) 144 parasympathetic system in response to cold face stimulus (Khurana, & Wu, 2006), independent of the change in respiration patterns. During the CFT, the time-course of ∆LHR showed a decrease ( ∆LHR, Figure 5.19C). The level of change in the LHR was not as pronounced as in the ∆HFP. Similar to the HFP response, the LHR response to the CFT was also unaffected by respiration; ∆LHR ra still showed the same trend in response to the cold face stimulus as ∆LHR (Figure 5.19D). Figure 5.19 Time-courses of heart rate variability (HRV) during cold face stimulus in CTL, non- transfused SCD, and transfused SCD groups: (A) high frequency power (HFP); (B) respiratory- adjusted HFP (HFP ra ); (C) low-frequency to high-frequency ratio (LHR) ; (D) respiratory- adjusted LHR (LHR ra ). Time t = 0 indicates the beginning of a 60-second cold face stimulus. Each plot shows group medians of RRI with either 75 th or 25 th percentiles. time (seconds) -100 -50 0 50 100 150 200 HFP (%baseline) -100 0 100 200 300 400 CTL (n=14) nonTransfused (n =15) Transfused (n=7) time (seconds) -100 -50 0 50 100 150 200 LHR (%baseline) -100 -50 0 50 100 150 200 time (seconds) -100 -50 0 50 100 150 200 HFPra (%baseline) -100 0 100 200 300 400 time (seconds) -100 -50 0 50 100 150 200 LHRra (%baseline) -100 0 100 200 300 A) B) C) D) 145 Figure 5.20 Heart rate variability responses during cold face. Each bar shows the median with the 25 th or 75 th percentile of area under the curve of each parameter between t = 0 and 30 seconds, t = 30 to 60 seconds, or t = 60 to 120 seconds (post cold face), when t = 0 indicates the beginning of the 60-second cold face stimulus. HFP = high frequency power, LHR = low-frequency to high- frequency ratio, HFP ra = respiratory-adjusted HFP, and LHR ra = respiratory-adjusted LHR. * indicates P<0.05. † indicates 0.05<P<0.1. The symbol above a bar indicates a significant difference of the parameter from its own baseline. The symbol above a bracket indicates a significant difference of the parameter between the 2 subject groups. Statistical analysis of the HRV responses to cold face stimulus showed a significant increase in HFP from baseline in the CTL subject (p<0.05, Figure 5.20A). Although this response was reduced to a non-significant level after the adjustment for the effect of respiration ( ∆HFP ra , Figure 5.20A). This is to be expected as the diving reflex activation from CFT has been shown to increase parasympathetic nervous activity (Heath, & Downey, 1990; Heistad et al., 1968; Stemper et al., 2002). Decreases in LHR ra was also observed in the controls during the CFT, suggesting a decrease in the central Control Transfused nonTransfused HFP (%baseline) -50 0 50 100 150 200 250 cold (0 - 60 sec) recovery (180 - 240 sec) * Control Transfused nonTransfused LHR (%baseline) -100 -50 0 50 100 150 Baseline Level Control Transfused nonTransfused HFP ra (%baseline) -50 0 50 100 150 200 250 Control Transfused nonTransfused LHR ra (%baseline) -100 -50 0 50 100 150 Baseline Level * * * A) B) C) D) 146 sympathetic nervous system activity in the general population. Neither the parasympathetic nor the sympathetic nervous system response was observed in the SCD groups. This is due to their high variations of the response. We speculate that the variations in their ANS sensitivity to cold as well as other perturbations might be related to their frequency of VOC. Furthermore we computed the G rsa , which indicates the changes in heart rate in response to changes in respiration. Figure 5.21 shows that on average there was no significant change in G rsa , suggesting that the ANS responses to the CFT are not mediated by respiration. 147 Figure 5.21 Respiratory Sinus Arrhythmia Gain (G rsa ) Changes from baseline during cold face. (A) Time-course of G rsa change from baseline during cold face stimulus; (B) Bar graph shows the median with the 25 th or 75 th percentile of area under the curve of each parameter between t = 0 and 30 seconds, t = 30 to 60 seconds, or t = 60 to 120 seconds (post cold face), when t = 0 indicate the beginning of 60-second cold face stimulus. 5.5.2. Discussion The CFT has been used widely as at a non-baroreflex related test of cardio-vagal function (Khurana, & Wu, 2006). In our study, the heart rate responses appeared only in the normal controls but not in the two SCD groups, suggesting that their cardio-vagal function was impaired. The sympathovagal balance index, LHR ra , in our CTL subjects decreased significantly (Figure 5.20), suggesting that in normal controls the ANS balance time (seconds) -100 -50 0 50 100 150 200 G rsa (%baseline) -100 -50 0 50 100 150 200 CTL (n=14) nonTransfused (n =15) Transfused (n=7) Control Transfused nonTransfused G rsa (%baseline) -50 0 50 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) A) B) 148 in response to the CFT shifts from sympathetic to parasympathetic dominance. The absence of this response in both of the SCD groups also suggests impairment in the ANS response to a vagal stimulus, which supports our earlier hypothesis that in SCD patients the parasympathetic system may be impaired. In addition to the cardiac ANS responses to the CFT, we also observed a strong cold face response in the peripheral system through reduced peripheral perfusion (Figure 5.15B). This reduction of blood flow in the periphery is a well known mechanism of the diving reflex, which preserves energy while maintain oxygen delivery through the heart and the brain (Heath, & Downey, 1990). While this decrease in peripheral perfusion was consistent in the normal controls, the degree of this response varied in both SCD groups. This variation might be related to the discrepancy in the severity and the frequency of their VOCs. Moreover, only SCD subjects demonstrated a second peripheral perfusion drop immediately after the cold pack was removed (Figure 5.15A). We suspected that this second pronounce perfusion drop resulted from hyperventilation (Figure 5.13) and its subsequent decreased peripheral perfusion responses (see section 5.3). Since cold weather is widely thought to cause VOCs in SCD patients (Baum et al., 1987; Serjeant et al., 1994), we speculate that cold could trigger VOCs through hypoperfusion- hyperventilation events mediated by neural signaling. 149 5.6. Evaluation of Minimal Model of HRV Control While the HRV analysis gives us insights on the balance between the sympathetic and parasympathetic divisions of the ANS, it does not explain the contributions of respiration and blood pressure in controlling of the cardiovascular system. Among many other factors which affect the HRV control, changes in respiration pattern and blood pressure level have a very pronounced, continual influence on HRV. The mechanisms of the ANS which adjust heart rates to changes in blood pressure and changes in respiration pattern are called the arterial baroreflex, ABR, and the respiratory cardiac coupling reflex, RCC, respectively. Properties of these reflexes (e.g. their sensitivities, latencies between changes in respiration or blood pressure and subsequent change in heart rate, and frequency responses) can be measured through a minimal model described in detail in Chapter 4. 5.6.1. Changes in ABR and RCC gains after chronic transfusion To assess the baseline levels of the ABR and RCC sensitivities (i.e. the gains of these reflexes in time-domain and in frequency domain at different frequency bands) before and after chronic transfusions in SCD patients, we applied the stationary version of the model with 8 chronically transfused patients. The pre-transfusion data were obtained on the day of transfusion before the transfusion took place. The post-transfusion data were obtained between 12 to120 hours after the transfusion to allow the patients’ hemodynamics to be stable. For each experimental session (pre- or post-transfusion), we 150 processed data from two 5-minute segments: one during a stable baseline recording and another during a deep-breath maneuver. The average values of the gains extracted from the 2 segments were used to represent each experiment. The reason that two data segments were used for each subject instead on one was to get a better representation of the ABR and RCC gains for the subject. Moreover, the perturbation of the cardiovascular system from the deep-breath maneuver could improve the signal-to-noise ratios of the ABR and RCC gain estimations. Nonetheless, the baseline data segments allowed us to assess the non-perturbed condition of each subject; therefore, we decided to use the baseline data segments as well. A sample of RRI, respiration, and SBP time-courses, as well as the corresponding ABR and RCC impulse responses and transfer functions from a patient during a deep- breath maneuver is shown in Figure 5.22. Heart rate increase (i.e. RRI decreases) during an inspiration phase and reverse during expiration. This coupling is usually very strong and thus lead to a strong respiration frequency component of HRV. This also caused the impulse response of the RCC to have a clear negative peak (Figure 5.22D). The ABR, on the other hand, usually has a slower impulse responses and a positive peak as shown in Figure 5.22E, indicating that and increase in blood pressure results in an increase in RRI. This is because when the blood pressure decreases, the body needs to increase the cardiac output to supply sufficient oxygen to the periphery and thus responses to the blood pressure drop through the arterial baroreflex, resulting in an increase of heart rate (i.e. decrease in RRI) and subsequent a increase in cardiac output. These impulse responses 151 were then converted to the frequency domain as shown by their corresponding transfer functions in Figure 5.22F and Figure 5.22G. From these impulse responses and transfer functions of RCC and ABR, the peaks of the impulse responses, the average low-frequency, high-frequency, and the average total gains were extracted from both pre- and post- transfusion data. The values were plotted as shown in Figure 5.23. Statistical analysis using a paired t-test on each descriptor also showed that after the transfusion there were no significant change in any of these descriptors pre- and post-transfusion (Table 5.7). 152 Figure 5.22 Sample of 5-minute input data from a deep-breath maneuver, as well as the impulse responses and transfer function gains calculated from the minimal model. A) r-wave to r-wave interval (RRI); B) respiratory waveform (wave); C) systolic blood pressure (sbp); D) RCC impulse response (h RCC ); E) ABR impulse response (h ABR ); F) Magnitude of the RCC transfer function (|H rcc |); G) Magnitude of the ABR transfer function (|H abr |). 50 100 150 200 250 300 -200 -100 0 100 rri time (seconds) g 50 100 150 200 250 300 0 0.5 1 1.5 resp time (seconds) 50 100 150 200 250 300 -30 -20 -10 0 10 sbp time (seconds) 0 10 20 -100 -50 0 50 h RCC time (seconds) 0 10 20 -0.2 0 0.2 0.4 0.6 0.8 h ABR time (seconds) 0 0.5 1 2 4 6 8 10 Frequency (Hz) |Hrcc(f)| 0 0.5 1 0.02 0.04 0.06 0.08 0.1 0.12 Frequency (Hz) |Habr(f)| B) A) C) D) E) F) G) 153 Figure 5.23 Average values of descriptors extracted from the impulse responses and transfer functions of ABR and RCC; pre- and post- transfusion. pre post LF gain of H RCC 0 10 20 30 pre post HF gain of H RCC 10 20 30 pre post Overall gain H RCC 10 20 30 pre post Peak of h RCC -800 -600 -400 -200 0 pre post LF gain of H ABR 0.00 0.05 0.10 0.15 0.20 0.25 0.30 pre post HF gain of H ABR 0.00 0.05 0.10 0.15 0.20 0.25 0.30 pre post Overall gain of H ABR 0.00 0.05 0.10 0.15 0.20 0.25 0.30 pre post Peak of h ABR -6 -5 -4 -3 -2 -1 0 1 B) A) C) D) E) F) G) H) 154 Table 5.7 Statistics of the descriptors extracted from impulse responses and transfer functions of ABR and RCC from sickle cell patients pre- and post-transfusion. Descriptor Pre-transfusion (mean ± sd) Post-transfusion (mean ± sd) 95% CI for difference of means p-value LF gain of RCC 10.981 ± 6.410 10.366 8.679 -5.770 to 7.000 0.830 HF gain of RCC 13.981 ± 10.424 13.799 11.395 -4.439 to 4.805 0.930 Overall gain of RCC 13.163 ± 9.135 12.862 10.522 -4.477 to 5.078 0.888 Peak of RCC IR -201.202 ± 193.245 -232.121 ± 247.194 -48.135 to 109.973 0.393 LF gain of ABR 0.143 ± 0.0851 0.143 ± 0.0693 -0.0760 to 0.0756 0.995 HF gain of ABR 0.159 ± 0.118 0.132 ± 0.0806 -0.0586 to 0.113 0.486 Overall gain of ABR 0.154 ± 0.104 0.135 ± 0.0763 -0.0594 to 0.0988 0.581 Peak of ABR IR -0.873 ± 2.050 -1.483 ± 1.240 -1.238 to 2.458 0.468 We noticed that, for some subjects, the ABR impulse response showed negative peaks, in contradiction to what we had expected in normal controls (Lin et al., 2004). This could be due to low sensitivity in our model selection algorithm or underlying SCD pathology. 155 5.6.2. Differences in ABR and RCC gains between normal controls and SCD patients To compare the ABR and RCC gains between the CTL and SCD subject, we used data from the study with the cold face experimental protocol, which consisted of three subject groups, i.e. normal controls (n = 9), non-chronically transfused SCD patients (n = 9), and chronically transfused SCD patients (n = 6). Five-minute baseline data segments during which the subjects rested in a supine position was extracted from the experiments. Median ABR and RCC impulse responses from all subject groups are shown in Figure 5.24A and Figure 5.26, respectively. These impulse responses were converted to transfer functions in the frequency domain. Median ABR and RCC are shown in Figure 5.25A and Figure 5.27A. We also extracted descriptors from the impulse responses and the transfer functions; details about these descriptors were detailed in section 4.3 (see Figure 4.2 and Figure 4.3). Descriptors extracted from each impulse responses are the maximum and minimum values; and descriptors extracted from each transfer function are the low-frequency (0.04 to 0.15 Hz), high-frequency (0.14 to 0.40 Hz), and overall (frequency = 0.04 to 0.40 Hz) gains. 156 Figure 5.24 Transient response of arterial baroreflex (ABR) during 5-minute baseline. (A) ABR impulse responses from 3 subject groups (each group shows median and 25 th or 75 th percentile); (B) maximum values of ABR impulse response; (C) minimum values of ABR impulse response (each box plot shows median 25 th and 75 th percentiles, error bars show 10 th and 90 th percentiles). time (seconds) 0 5 10 15 20 ABR impulse resp -0.5 0.0 0.5 1.0 1.5 CTL (n = 9) non Transfused (n = 9) Transfused (n = 6) B) A) C) CTL nonTransfused Transfused max of ABR impulse response 0 5 10 15 CTL nonTransfused Transfused min of ABR impulse response -6 -4 -2 0 157 Figure 5.25 Frequency response of arterial baroreflex (ABR) during 5-minute baseline. (A) ABR transfer functions from 3 subject groups (each group shows median and 25 th or 75 th percentile); (B) low-frequency gain (LF gain, freq = 0.04 – 0.15 Hz) of ABR transfer function; (C) high- frequency gain (HF gain, freq = 0.15 – 0.40 Hz) of ABR transfer function; (D) overall gain (freq = 0.04 – 0.40 Hz) of ABR transfer function (each box plot shows median 25 th and 75 th percentiles, error bars show 10 th and 90 th percentiles). CTL nonTransfused Transfused LF gain of ABR Transfer Fn 0.0 0.1 0.2 0.3 0.4 CTL nonTransfused Transfused HF gain of ABR Transfer Fn 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 CTL nonTransfused Transfused overall gain of ABR Transfer Fn 0.0 0.1 0.2 0.3 0.4 0.5 0.6 time (seconds) 0.0 0.2 0.4 0.6 0.8 1.0 ABR transfer function 0.0 0.1 0.2 0.3 0.4 0.5 CTL (n = 9) non Transfused (n = 9) Transfused (n = 6) B) A) C) D) 158 Figure 5.26 Transient response of respiratory-cardiac coupling (RCC) during 5-minute baseline. (A) RCC impulse responses from 3 subject groups (each group shows median and 25th or 75th percentile); (B) maximum values of RCC impulse response; (C) minimum values of RCC impulse response (each box plot shows median 25th and 75th percentiles, error bars show 10th and 90th percentiles). CTL nonTransfused Transfused max of RCC impulse response 0 100 200 300 400 CTL nonTransfused Transfused min of RCC impulse response -1000 -800 -600 -400 -200 0 B) A) C) time (seconds) 0 5 10 15 20 RCC impulse resp -200 -150 -100 -50 0 50 100 CTL (n = 9) non Transfused (n = 9) Transfused (n = 6) 159 Figure 5.27 Frequency response of respiratory-cardiac coupling (RCC) during 5-minute baseline. (A) RCC transfer functions from 3 subject groups (each group shows median and 25th or 75th percentile); (B) low-frequency gain (LF gain, freq = 0.04 – 0.15 Hz) of RCC transfer function; (C) high-frequency gain (HF gain, freq = 0.15 – 0.40 Hz) of RCC transfer function; (D) overall gain (freq = 0.04 – 0.40 Hz) of RCC transfer function (each box plot shows median 25 th and 75 th percentiles, error bars show 10 th and 90 th percentiles). Comparisons of the descriptors among all subject groups are shown in Figure 5.24 to Figure 5.27. Nonetheless, none of these descriptors showed any significant difference among the three groups. The results from this and the previous section suggest that the baseline ABR and RCC mechanisms did not differ between SCD subjects (both CTL nonTransfused Transfused LF gain of RCC Transfer Fn 0 10 20 30 40 50 CTL nonTransfused Transfused HF gain of RCC Transfer Fn 0 10 20 30 40 50 CTL nonTransfused Transfused overall gain of RCC Transfer Fn 0 10 20 30 40 time (seconds) 0.0 0.2 0.4 0.6 0.8 1.0 RCC transfer function 0 5 10 15 20 25 CTL (n = 9) non Transfused (n = 9) Transfused (n = 6) B) A) C) D) 160 chronically transfused, and non-chronically transfused ones), and normal subjects, nor did they improve in SCD patients after their chronic transfusion. Since we could not distinguish subject groups using this time-invariant model, we decided to compute the time-varying version of the model in response to the cold face stimulus. Results from this time-varying analysis are presented in the next section. 5.6.3. Time-varying Minimal Model of HRV Control in Response to Cold Face Stimulus We have also employed the time-varying version of this minimal model to evaluate the changes in mechanisms which control heart rate in response to the cold face stimulus. This technique allows calculations of the ABR and RCC indices for every data sample. In this case, the data used in this model were re-sampled to 2 Hz for model computation. Therefore, the ABR and RCC impulse responses were computed every 0.5 seconds (details on this calculation have been presented in section 4.6). From these impulse responses, we calculated the corresponding transfer functions and subsequently all the descriptors shown in Figure 4.2 and Figure 4.3. Figure 5.28 and Figure 5.29 present the changes in the ABR and RCC descriptors in response to cold face stimulus, respectively. 161 Figure 5.28 Changes in descriptors extracted from time-varying impulse responses and transfer functions of the arterial baroreflex (ABR) gain from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. (A) time-course of ABR low-frequency gain (Habr_lf); (B) Habr_lf during cold face and recovery period; (C) time- course of ABR high-frequency gain (Habr_hf); (D) Habr_hf during cold face and recovery period; (E) time-course of ABR overall gain (Habr_overall); (F) Habr_overall during cold face and recovery period; (G) time-course of negative peak of ABR (abr_min); (H) abr_min during cold face and recovery period; (I) time-course of positive peak of ABR (abr_max); (J) abr_max during cold face and recovery period. Time t = 0 indicates the starting time of the 60-second cold face stimulus. Each plot shows group medians with either 75th or 25th percentiles. tim e (se con ds) -100 0 100 200 Habr_lf (%baseline) -1 0 0 0 100 200 300 CT L (n = 9 ) n o nT ra ns fu s ed (n = 9) T rans fu s ed (n = 6) Control Transfused nonTransfused Habr_lf (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (s ec onds ) -100 0 100 200 Habr_hf (%baseline) -1 0 0 0 100 200 300 C T L (n = 9) nonT rans fu s ed (n = 9) T rans fu s ed (n = 6) Control Transfused nonTransfused Habr_hf (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (s e c on ds ) -100 0 1 00 20 0 Habr_overall (%baseline) -1 0 0 0 100 200 300 C T L (n = 9) nonT rans fu s ed (n = 9) T rans fu s ed (n = 6) C ontrol Transfused nonTransfused Habr_overall (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (s ec on ds ) -100 0 100 200 abr_min (%baseline) -100 0 100 200 300 C T L (n = 9) nonTrans fu s ed (n = 9 ) Trans fu s e d (n = 6) C ontrol Transfused nonTransfused abr_min (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (s ec o nds ) -100 0 100 200 abr_max (%baseline) -1 0 0 0 100 200 300 C T L (n = 9) n onT rans fu s e d (n = 9) T rans fu s ed (n = 6) C ontrol Transfused nonTransfused abr_max (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) A) B) C) D) E) F) G) H) I) J) 162 Figure 5.29 Changes in descriptors extracted from time-varying impulse responses and transfer functions of the respiratory-cardiac coupling (RCC) gain from baselines following cold face stimulus in control (CTL), chronically transfused SCD, and non-chronically transfused SCD groups. (A) time-course of ABR low-frequency gain (Hrcc_lf); (B) Hrcc_lf during cold face and recovery period; (C) time-course of RCC high-frequency gain (Hrcc_hf); (D) Hrcc_hf during cold face and recovery period; (E) time-course of RCC overall gain (Hrcc_overall); (F) Hrcc_overall during cold face and recovery period; (G) time-course of negative peak of RCC (rcc_min); (H) rcc_min during cold face and recovery period; (I) time-course of positive peak of RCC (rcc_max); (J) rcc_max during cold face and recovery period. Time t = 0 indicates the starting time of the 60-second cold face stimulus. Each plot shows group medians with either 75th or 25th percentiles. + indicates a significant difference from the control group (p<0.05) tim e (s e c o n d s ) -100 0 100 200 rcc_min (%baseline) -100 -5 0 0 50 100 150 200 C T L (n = 9) n onTra ns fu s ed (n = 9 ) Trans fu s ed (n = 6) tim e (seco nds) -100 0 100 200 Hrcc_hf (%baseline) -10 0 -50 0 50 100 150 200 C TL (n = 9) nonTransfu sed (n = 9) Transfu sed (n = 6) tim e (seconds) -100 0 100 200 Hrcc_lf (%baseline) -100 0 100 200 CT L (n = 9 ) non Transfu se d (n = 9) T ran sfuse d (n = 6 ) Control Transfused nonTransfused Hrcc_lf (% Change from Baseline) -50 0 50 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) + + C ontrol T ransfused nonT ransfused Hrccr_hf (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 B aseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (s ec on ds ) -100 0 100 200 Hrcc_overall (%baseline) -1 0 0 0 100 200 CT L (n = 9 ) n onTra n s fu s ed (n = 9) Trans fu s ed (n = 6 ) Control Transfused nonTransfused Hrccr_overall (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) + + Control Transfused nonTransfused rcc_min (% Change from Baseline) -60 -40 -20 0 20 40 60 80 100 120 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) tim e (seconds ) -100 0 100 200 rcc_max (%baseline) -100 0 100 200 300 CT L (n = 9 ) no nT rans fu s ed (n = 9 ) T ran s fused (n = 6) Control Transfused nonTransfused rcc_max (% Change from Baseline) -50 0 50 100 150 Baseline Level cold (0 - 60 sec) recovery (180 - 240 sec) A) B) C) D) E) F) G) H) I) J) 163 By inspection, the left panels of Figure 5.28 show increases of all the ABR descriptors in the CTL subjects but not as pronounced in the two SCD groups. These increases indicate an increase of the ABR sensitivity of CTL subjects during the application of the cold pack. This agrees with the results from (Stemper et al., 2002)) that CFT increase baroreflex sensitivity. Furthermore, to statistically compare these descriptors across the subjects, we extracted median values of each descriptor during t = 0 to 60 and t = 180 to 240 from the onset of the 60-second CFT to represent each subject during the CFT and the recovery periods, respectively. Then, we applied the repeated-measures analysis of variances (RM ANOVA) with the descriptors extracted from each subject group to determine if there was any change from baseline during the CFT and the recovery periods. In addition, we applied the ANOVA on Ranks test to determine if the changes in these descriptors from baseline during both periods were different among the three subject groups. We also applied the post-hoc analyses after using these ANOVA tests to identify the sources of the differences, both from baselines and between groups. In contrast to our earlier inspection, the statistical analysis results on the right panels of Figure 5.28 show that none of these increases in the ABR descriptors was significant from their baselines, nor was there any difference among the 3 groups. This is likely to be because of the high variability in the all the descriptors, possibly due to the recordings or our model computation technique. 164 Similar to the time-courses of the ABR descriptors, the time-courses of the RCC descriptors showed increases (but not significant) in RCC sensitivity in CTL subjects during the CFT, which did not occur in the SCD subjects (Figure 5.29). This also agrees with the results from (Stemper et al., 2002). Moreover, we observed increases in the descriptors in CTL group during the recovery period, about 2 minutes after the removal of the cold pack. We do not have an explanation for this second wave of the RCC sensitivity spike yet. To further analyze the responsiveness to CFT in each group of subjects, we applied the signed-rank test to detect the changes of each descriptor from its own baseline during the CFT and during the recovery period. The changes of each descriptor from baseline in both groups of SCD subjects were also compared to those from the CTL subjects, using the rank-sum test. Results of these statistical analyses are shown in Table 5.8. 165 Table 5.8 p-values from pairwise comparisons of the changes in the ABR and RCC descriptors in response to the CFT Compared to baseline (signed-rank test) Compared to CTL (rank-sum test) parameters cold recovery cold recovery CTL Transfused Non- Transfused CTL Transfused Non- Transfused Transfused Non- Transfused Transfused Non- Transfused Habr_lf 0.098 0.688 0.910 0.820 0.563 0.910 0.328 0.222 0.456 0.546 Habr_hf 0.098 0.6875 0.820 0.359 0.0938 0.910 0.328 0.436 0.088 0.605 Habr_overall 0.098 0.688 1 1 0.156 1 0.272 0.258 0.272 0.796 abr_min 0.027 0.438 0.910 0.359 0.063 0.910 0.456 0.094 0.114 0.297 abr_max 0.359 1 0.426 0.570 0.219 1 0.776 0.796 0.955 0.605 Hrcc_lf 0.250 1 0.910 0.039 0.094 0.203 0.388 0.340 0.012 0.019 Hrcc_hf 0.074 1 0.426 0.164 0.094 0.734 0.328 0.032 0.036 0.136 Hrcc_overall 0.074 1 0.734 0.027 0.063 0.426 0.328 0.258 0.002 0.032 rcc_min 0.203 0.844 0.250 0.055 0.313 0.359 0.272 0.063 0.036 0.063 rcc_max 0.098 0.844 0.6520.164 0.0313 0.426 0.224 0.340 0.036 0.258 166 From these pairwise comparisons, we can see that both ABR and RCC descriptors showed a tendency for changes from baseline only in the CTL group but not in both of the SCD groups, confirming our visual inspection that the ABR and RCC responses were absent in SCD patients in comparison to the normal controls. Note that some of the parameters which show p-values of less than 0.05 in Table 5.8 did not appear as a marker in Figure 5.28 and Figure 5.29. This was because the post-hoc multiple pair-wise comparisons represent in these figures discount the significant level from the standard p- value of 0.05. 5.6.4. Discussion Model-based analysis of the heart rate control showed no significant improvement after transfusion in any descriptor extracted from the ABR and RCC impulse responses as well as their transfer functions. This suggests that blood transfusion did not alter the baroreflex or the respiratory-cardiac coupling mechanisms. Nonetheless, a larger sample size is still needed to confirm this finding. Comparisons among the three independent subject groups (normal controls, non- chronically transfused SCD patients, and chronically transfused SCD patients) also showed no difference. This suggests that there may not be any abnormality in the autonomic control of heart rate in SCD patients during baseline resting condition. However, abnormalities in the autonomic control in SCD patients might exist in response 167 to external stimuli. Therefore, we further investigated the changes of ABR and RCC in response to cold face stimulation, as it is one of the widely used autonomic stimuli. The results of this time-varying analysis of ABR and RCC in response to cold face stimulation show a tendency for increases in sensitivity of both baroreflex and respiratory cardiac coupling during the application of a cold pack in the normal controls, agreeing with the results from (Stemper et al., 2002)), who also showed increase in sensitivity of both mechanism during CFT. The variations across individuals of all ARB and RCC descriptors in all subject groups were very high. Therefore we did not detect a significant change in any of the parameters, nor did we detect any difference between any groups during the application of the cold pack. The results from both stationary and time-invariant minimal model of cardiovascular control presented in this section showed very high variability of all parameters. Nonetheless, the graphs shown in this section (Figure 5.28 and Figure 5.29) suggest suppressed regulations of both ABR and RCC mechanisms in both SCD groups during the CFT in comparison to the control subjects. We speculate that the impairment of the ABR response to the cold face stimulus might be responsible for the absence of cardiac sympathetic withdrawal or the shift of the ANS to parasympathetic dominance seen in normal controls (represented by LHR and LHR ra in Figure 5.20). Abnormalities of these mechanisms in SCD subjects could play a role in the initiation of vaso-occlusion in cold weather, which is a known risk of VOC. A larger sample size and further 168 refinements in the methodology for estimating time-varying model parameters are likely needed to improve the detection of the effects of the CFT or any other transient stimuli on ABR and RCC. 5.7. Relationship between HRV and Laboratory Blood Test To better understand the relationship between autonomic response and conventional blood tests normally used to diagnose sickle cell disease, we generated scatter plots between all pairs HRV parameters measured in response to hypoxia and blood test data. We selected the HRV response to hypoxia because, when compared to other stimuli, its ANS responses were significant and more pronounced. Parameters used in this analysis are listed below. 169 Table 5.9 Blood test data and HRV parameters in response to hypoxia that used to generate scatter plots Parameters from Blood test HRV response to hypoxia  Hemoglobin  Reticulocyte count  White blood cell (WBC) count  Red blood cell (RBC) count  Hematocrit  Mean Corpuscular Volume (MCV)  Platelets  Absolute Neutrophil Count (ANC)  High-sensitivity C-Reactive Protein (hs-CRP)  Hemoglobin A2  Hemoglobin F  RRI (% change from baseline)  log 10 (HFP %baseline)  log 10 (LHR %baseline)  log 10 (HFP ra %baseline)  log 10 (LHR ra %baseline)  G rsa (% change from baseline) 5.7.1. Results Scatter plots and correlation analysis of these 11 blood test parameters vs. 6 HRV parameters in response to hypoxia (total of 11 x 6 = 66 plots) show significant relationships between 17 pairs of parameters (p<0.05), and tendency for relationships between 7 pairs of parameters (0.05<p<0.1). Table 5.10 and Table 5.11 show results from correlation analysis and scattered plots of these relationships, respectively. Note from the graphs in Table 5.11 that some HRV indices are expressed as log 10 for normalization. 170 Table 5.10 Summary of p-value from the correlation results 11 lab parameters vs. 6 HRV parameters in response to hypoxia. Each cell shows correlation coefficient (p-value); * indicates P<0.05. † indicates 0.05<P<0.1. RRI (% change from baseline) log 10 (HFP %baseline) log 10 (LHR %baseline) log 10 (HFP ra %baseline) log 10 (LHR ra %baseline) G rsa (% change from baseline) Hemoglobin (g/dL) 0.5433 (0.0074)* 0.4428 (0.0302)* 0.3989 (0.0535) † 0.3732 (0.0725) † 0.1421 (0.5078) 0.4517 (0.0267)* Reticulocyte count (%) -0.4956 (0.0162)* -0.3437 (0.1001) -0.1836 (0.3904) -0.4084 (0.0476)* -0.0208 (0.9231) -0.2456 (0.2474) WBC count (x10 9 /L) -0.2988 (0.1661) -0.3465 (0.0972) † -0.2527 (0.2334) -0.4661 (0.0217)* 0.0170 (0.9370) -0.3824 (0.0652) † RBC count (x10 12 /L) 0.6090 (0.0034)* 0.4632 (0.0299)* 0.3145 (0.1540) 0.3858 (0.0762) † 0.0092 (0.9968) 0.4520 (0.0347)* Hematocrit (%) 0.6118 (0.0032)* 0.4943 (0.0194)* 0.2795 (0.2078) 0.4152 (0.0547) † 0.0406 (0.8577) 0.4718 (0.0266)* MCV (femtoliter) -0.1213 (0.5814) -0.0541 (0.8017) -0.0799 (0.7104) -0.0777 (0.7181) 0.1555 (0.4680) -0.1515 (0.4798) Platelets (x10 9 /L) -0.0251 (0.9096) -0.1057 (0.6231) -0.2637 (0.2131) -0.3760 (0.0701) † -0.0693 (0.7476) -0.1925 (0.3674) ANC (/mm 3 ) -0.2417 (0.3500) -0.3856 (0.1140) -0.1433 (0.5707) -0.4818 (0.0429)* -0.0819 (0.7466) -0.2587 (0.3000) hs-CRP (mg/L) -0.2495 (0.2628) -0.3317 (0.1220) 0.0788 (0.7207) -0.4516 (0.0305)* 0.2608 (0.2294) -0.3423 (0.1098) Hemoglobin A2 (%) -0.6772 (0.1395) -0.8752 (0.0099)* 0.2173 (0.6398) -0.8322 (0.0202)* 0.4556 (0.3043) -0.4936 (0.2603) Hemoglobin F (%) -0.4590 (0.0418)* -0.3349 (0.1378) -0.1553 (0.5015) -0.2455 (0.2834) 0.1151 (0.6193) -0.4187 (0.0589) † HRV response to hypoxia Lab data 171 Table 5.11 Scatter plots between HRV responses to hypoxia and lab data RRI (% change from baseline) log 10 (HFP %baseline) log 10 (LHR %baseline) log 10 (HFP ra %baseline) log 10 (LHR ra %baseline) G rsa (% change from baseline) Hemoglobin (g/dL) Reticulocyte count (%) WBC count (x10 9 /L) RBC count (x10 12 /L) Hematocrit (%) MCV (femtoliter) Platelets (x10 9 /L) ANC (/mm 3 ) hs-CRP (mg/L) Hemoglobin A2 (%) Hemoglobin F (%) log10(HFP%baseline) log10(LHR%baseline) log10(HFP%baseline) log10(HFPra%baseline) log10(LHRra%baseline) log10(HFP%baseline) log10(LHRra%baseline) log10(HFP%baseline) log10(HFPra%baseline) log10(HFP%baseline) log10(HFP%baseline) log10(LHRra%baseline) log10(HFP%baseline) log10(HFP%baseline) log10(HFP%baseline) log10(HFP%baseline) log10(LHRra%baseline) log10(HFP%baseline) Lab data HRV response to hypoxia 172 Measures of changes in parasympathetic modulations (HFP, HFP ra , and G rsa ) showed positive correlations to parameters related to red blood content (hemoglobin, hematocrit, RBC count), except reticulocyte count, which is usually high in SCD patients due to compensation of the bone marrow for their anemia. This suggests that the degree of anemia (decreased hemoglobin, RBC count, and hematocrit) is associated with the sensitivity of hypoxia response. This means that, following a hypoxic stimulus, a more anemic patient will become more tachycardia and had a stronger parasympathetic withdrawal. On the other hand, parameters related to white blood content (white blood cell count, and ANC), inflammation (hs-CRP), and hemoglobin A 2 show negative correlations to HRV parameters related to changes in parasympathetic modulations. In other words, parameters related to white blood cells were correlated to RRI drops and parasympathetic withdrawal. Since SCD is known to be an inflammatory disease, i.e. sickle cell patients are constantly in an inflammation state (Hebbel et al., 2004), this result suggests that the more inflamed the SCD patient, the more sensitive his/her ANS response to hypoxia will be. However, the decrease in RBC and the increase in WBC contents in SCD patients might be correlated. Therefore, one may have a confounding effect on the others. As a result, we created a scatter plot between RBC count and WBC count (Figure 5.30) which shows a significant correlation between RBC count and WBC count (R 2 = 0.522, p < 173 0.001). To separate the effect of RBC and WBC counts, we computed partial correlation between both blood counts and HRV parameter changes following hypoxia as shown in Table 5.12. Figure 5.30 Scatter plot between RBC count and WBC count Table 5.12 Summary of p-value from the partial correlation results RBC and WBC counts vs. 6 HRV parameters in response to hypoxia. Each cell shows partial correlation coefficient (p-value); * indicates P<0.05. † indicates 0.05<P<0.1. RRI (% change from baseline) log 10 (HFP %baseline) log 10 (LHR %baseline) log 10 (HFP ra %baseline) log 10 (LHR ra %baseline) G rsa (% change from baseline) WBC count, controlling for RBC count 0.0976 (0.6821) -0.0639 (0.7833) -0.0751 (0.7464) -0.3159 (0.1630) 0.0166 (0.9432) -0.1167 (0.6144) RBC count, controlling for WBC count 0.5265 (0.0171) * 0.3080 (0.1743) 0.1779 (0.4403) 0.0743 (0.7489) 0.0124 (0.9575) 0.2663 (0.2433) 2 4 6 8 10 12 14 16 WBC count 2 2.5 3 3.5 4 4.5 5 5.5 RBC count HRV response to hypoxia Lab data 174 The partial correlation results in Table 5.12 show that the degrees of the partial correlations between the HRV responses to hypoxia and WBC count controlling for the RBC counts, as well as RBC count controlling for WBC counts, are reduced compared to the correlation without the adjustment as shown in Table 5.10. These decreased correlations suggest that the association between blood counts and HRV responses to hypoxia are associated with the levels of both RBC and WBC counts; the level of one could not be considered separately from the other. 5.7.2. Discussion From these above results, we speculate that the sensitivity of the ANS responses to hypoxia in SCD patients were partly due to their anemia (associated with low RBC counts) and their level of inflammation (associated with high WBC counts and high hs- CRP). Their causal relationships, as well as the mechanism behind these relationships, are still the topic of investigation. Researchers have demonstrated a neural control of inflammation, especially through the parasympathetic branch of the ANS; i.e. the cholinergic anti-inflammation mechanisms inhibit the activation of macrophages and the release of cytokines (Tracey, 2002). Inflammation has been shown to alter neural activations (Besedovsky et al., 1983; Watkins, & Maier, 2002). On the other hand, there is evidence that stimulation of the vagus nerve prevents inflammation (Borovikova et al., 2000). Borovikova et al. described the ‘cholinergic anti-inflammatory pathway’ as shown in Figure 5.31 (from (Tracey, 175 2002)). More recently, another study has shown that vagus nerve stimulation could potentially improve heart failure as well as reduce inflammation (Zhang et al., 2009). In the present study, we suspect that, because of the constant inflammation state in SCD, the parasympathetic system in this group of patients is constantly being activated; and thus, it might be hypo-sensitive to an extra external perturbation. The results from this study suggest the ANS as another possible pharmacological target for preventing events that may trigger VOC. Figure 5.31 Wiring of the inflammatory reflex (figure from (Tracey, 2002)) 176 A low level of total hemoglobin, fetal hemoglobin, and an elevated WBC counts have been associated with the risk of premature death in SCD patients (Platt et al., 1994). The mechanism which link between the blood counts and the complications that leads to the death is probably related in part to the ANS. Moreover, up to 23% of deaths in adults with sickle cell disease were reported as sudden death events with no detectable cause found at autopsy (Darbari et al., 2006; Perronne et al., 2002; Platt et al., 1994), while a study (Darbari et al., 2006) showed that about 12% of deaths in SCD patients were caused by cardiac diagnoses, including myocardial infarction, heart failure, cardio- myopathy, and cardiac arrhythmia. Dysfunctional of the ANS in particular is a significant risk factor for cardiovascular adverse events, including sudden death in the general population (Curtis, & O'Keefe, 2002) . This abnormality is very likely to add stress to the cardiovascular systems and subsequently increase the risk of crisis in SCD patients. 177 Chapter 6. Future Research Opportunities This chapter presents possible future work. Some literature reviews and preliminary tests on these proposed future works has been performed and are presented in this chapter. 6.1. Closed-loop Minimal Model of Heart Rate Control Besides studying the reflex mechanism of ABR and RCC described in Chapter 4, we could also study the other two limbs of the closed-loop cardiovascular control system to determine how blood pressure (as assessed by the change in mean arterial pressure, ΔMBP) is influenced by fluctuations in RRI and Respiration (Khoo, 2008). A control block diagram which describes this closed-loop model is shown in Figure 6.1. Figure 6.1 Control box diagram of the minimal close-loop model (from (Khoo, 2008)). 178 We also plan to apply this model to track the changes in each mechanism shown in figure 6.1 in response to the autonomic stimuli. With this time-varying model, all impulse responses are assumed to be changing with time. This would allow us to track the changes in each extracted feature as a response to autonomic perturbations, such as sighs, hypoxia, or the cold pressure test. Both time-invariant and time-varying analysis will be applied to data from both SCD and control groups. 6.2. Minimal Model of Peripheral Perfusion Control 6.2.1. Introduction This is another minimal model which is designed to explore what controls peripheral perfusion. Perfusion controls are achieved thought either intrinsic or extrinsic pathways. While the intrinsic pathway is mainly controlled by the local effect of Nitric Oxide, NO, the extrinsic pathway is controlled through sympathetic nerve signaling from the central neural system (CNS) (Berne, & Levy, 2001). NO is an endothelium-derived vasodilator, which helps maintain the baseline vasodilation, allowing sufficient perfusion to tissue beds. It also plays a vital role in adapting vasculature to different stressors, such as oxidative stress and shear stress. While NO plays an essential role in vasodilation, SCD patients commonly have NO depletion. Multiple factors contribute to its depletion in SCD patients, including their increased level of NO-scavenging cell free hemoglobin, lower bioavailability of L-arginine for NO production, and inflammation (Gladwin et al., 2003; Nath et al., 2004). 179 To dissociate the effect of central sympathetic control from the effect of the endothelial local vasodilator, especially NO, to the vascular tone, we will use FMD to stimulate the NO production on the brachial artery of one arm and measure the subsequent peripheral perfusion responses on both arms. Changes in peripheral perfusion observed by laser Doppler could be due to either the NO or the CNS effect. To be able to dissociate the two effects, we propose a model which employs continuous measurement of blood pressure as well as perfusion from 2 locations; ipsi- and contra-lateral to the cuffed arm in the FMD experiment (experimental details can be found in Chapter 2). On the contra-lateral limb, change in perfusion is assumed to be a function of peripheral sympathetic modulation alone as there was minimal (or no) change in the shear rate thus the change in NO level on this limb can be considered negligible. On the other hand (literally), when the blood pressure cuff is released from the arm, the change in perfusion on the ipsi-lateral limb is assumed to be a function of both peripheral sympathetic modulation and change in NO level due to change in shear rate. 6.2.2. Computational Technique For each arm, the measured perfusion is assumed to be a summation of the sympathetic effect and the endothelial effect, while blood pressure is treated as a driving input (Figure 6.2). On the ipsi-lateral side of the FMD, the effect of endothelial is strongly presented since the change in shear rate caused by increase in flow from the release of pressure cuff will cause an increase in NO production, which results in 180 vasodilation (Figure 6.2A). On the contra-lateral side, however, the change in shear rate is assumed to be negligible, thus any changes in perfusion are based mainly on the effect of sympathetic nerve innervations (Figure 6.2B). Since sympathetic system is controlled from the medulla of the CNS, we can estimate the effect of this system to be similar on both sides. Figure 6.2 Diagram of the minimum model describing fluctuation of peripheral perfusion. (A) Additional effect of endothelial function becomes significant on the ipsi-lateral side of FMD, (B) The effect of endothelial function is considered negligible on the contra-lateral side. On the contra-lateral side, the fluctuations in PU, ΔPU, are assumed to be a function of fluctuation of blood pressure, ΔSBP, alone. Therefore, the impulse response A) B) 181 between ΔSBP and ΔPU (h Symp ) on this side represents the resistance of the vessels. In the other words, the changes in pressure applying to the resistance will cause a change in current or perfusion. This resistance can be altered when smooth muscles on the wall of large capillaries or arterioles constrict, in response to changes of the firing rate of the peripheral sympathetic neurons. This relationship is also assumed to be present on the ipsi-lateral side. In addition, on this side, the effect of the change in peripheral resistance due to the NO effect is modeled as another additive input to ΔPU. We use ΔSBP as the driving force for this part, and the impulse response which relates ΔSBP to ΔPU is denoted as h Endo . The stochastic and other influences on ΔSBP are denoted as ω PUcontra and ω PUcontra for contra-lateral ΔPU and ipsi-lateral ΔPU, respectively. To obtain both impulse responses, we will first use the measurement from the contra lateral side to quantify h Symp (t): ∆PU contra t h Symp i ∆SBPtiT Symp M1 i0 ω PUcontra t Equation 6.1 Substitute h Symp (t) into Equation 6.2 for the ipsi-lateral side, we can obtain the impulse response for the endothelial effect, h Endo (t). ∆PU ipsi th Symp i ∆SBP tiT Symp M1 i0 ∑ h Endo i ∆SBP tiT Endo M1 i0 ω PUipsi t Equation 6.2 182 where T Symp and T Endo are latencies associated with h Endo (t) and h Symp (t). For causality of the system, we only allow positive numbers for both T Symp and T Endo . For each model, we will extract the peak of the transfer functions, and the gain of the transfer function in different frequency bands to compare between subject groups. These parameters will allow us to have a better understanding of the degree of control from the central and local endothelial cells to vasodilation response observed in FMD. As a result the effect of endothelial or NO can be more effectively assessed and compared between cohorts of subjects. For each step mentioned above, we will use a basis function fitting method to estimate the impulse responses. More details about the estimation technique were described in Chapter 4. In brief, we estimate each transfer function to be a linear combination of Laguerre’s or Meixner’s orthonormal basis functions through the Least Square estimator (Lin et al., 2004; Chaicharn et al., 2008; Asyali, & Juusola, 2005). We will compute the transfer functions with both time-invariant and time-varying dynamic assumptions. We have noticed during the baseline measurement, that PU contra and PU ipsi sometimes fluctuate in different ranges. This due to the amount of total back scattered power received by each transducer. Even in the same subject, this power may vary from day to day based on various factors including transducer placement, connection, and lack of calibration. To circumvent this issue, we normalize both PU contra and PU ipsi by their 183 own means and baselines during stable baseline recording. Therefore, for the purpose of modeling, we use ∆PU contra and ∆PU ipsi calculated from Equation 6.3 and Equation 6.4 respectively. ∆PU contra t PU contra meanPU contra ,baseline stdPU contra ,baseline Equation 6.3 ∆PU ipsi t PU ipsi meanPU ipsi ,baseline stdPU ipsi ,baseline Equation 6.4 6.2.3. Preliminary Results Ten minute data segment obtained from SCD patients pre- and post-transfusion that contain an FMD maneuver are extracted from each patient. A raw time series is shown in Figure 6.3. Time t = 0 was set to be at the release of the blood pressure cuff, and data from t = 0 to 300 seconds was used to obtain a set of time-invariant impulse responses, h Endo (t) and h Symp (t), as detailed in previous section. A sample of the h Endo (t) and h Symp (t) is shown in Figure 6.4. 184 Figure 6.3 Time series of (A) systolic blood pressure (SBP), (B) peripheral perfusion (PU contra and PU ipsi ) during an FMD experiment. Figure 6.4 Impulse responses which represent (A) the effect of sympathetic control (h Symp ) and, (B) the effect of endothelial function (h Endo ) to the change in peripheral resistance. A) B) -300 -200 -100 0 100 200 300 60 70 80 90 100 SBP (mmHg) -300 -200 -100 0 100 200 300 -10 0 10 20 normalized PU time (seconds) ipsi contra A) B) 0 5 10 15 20 25 -0.02 0 0.02 0.04 h symp time (seconds) 0 5 10 15 20 25 -0.03 -0.02 -0.01 0 0.01 h endo time (seconds) 185 The transfer functions between blood pressure and flow represents the peripheral resistance. Our preliminary analysis from a real data from one subject shows that h Symp has a clear positive peak (Figure 6.4). This was calculated from the systemic blood pressure and perfusion from the arm with no stimulus. This can be interpreted that the resistance from the sympathetic control and other central control have a positive resistant to blood flow. On the other hand, h Endo shows a negative peak with smaller magnitude than h Symp . This suggests that the endothelial effect from FMD causes the lowering of the peripheral resistance. However the magnitude of the decrease in resistance is still smaller than the combination of the baseline resistance and change in resistance from the central control. Therefore the total peripheral resistance still has a positive value in this case. We also noticed that the time lags h Symp and h Endo are 1.5 and 0.5 seconds respectively, as shown by the time that the impulse responses start to take a non-zero value. This suggests that the effect of the endothelial function reacts faster than the sympathetic or central effect. We plan to apply this technique to datasets from SCD patients before and after their chronic transfusion. We will also compare the transfer functions and the frequency responses computed from this model between subject groups. We hope that this will give us some insight on the effect of blood transfusion on the regulation of peripheral perfusion both by the central neural system and the endothelial function. 186 6.3. Non-linear Model of Vascular Resistance 6.3.1. Introduction Our results from section 5.3 show that spontaneous sighs could induce sharp drops in microvascular blood flow in SCD patients, and that the likelihood that a sigh will cause the flow to drop is significantly higher in SCD subjects than CTLs (Figure 5.6). We also found that only inspirations with large volumes caused this blood flow drop event; clearly the inspiration volume needs to exceed some threshold for this event to take place. This led to a proposal of this non-linear model of peripheral vaso-constriction. We expected that the threshold component of the non-linearity may help explaining this reflex in SCD subjects. This post-deep-inspiration vasoconstriction reflex was observed in normal subjects as early as 1895 (Binet, & Sollier, 1895). It was later shown to be mediated by the sympathetic nerve activation (Bolton et al., 1936; Baron et al., 1996). Constriction of peripheral blood vessel is controlled by smooth muscle lining of the arterioles (Berne, & Levy, 2001). A constriction of these muscles would lead to a decrease of blood flow. Recordings of muscle sympathetic nerve activity (MSNA) by Eckberg et al. showed clearly that muscle sympathetic nerve activity is in sync with respiration; i.e. maximum activity occurred at end-expiration and minimum activity occurred at end-inspiration as shown in Figure 6.5 (Eckberg et al., 1985; Eckberg, 1995). 187 Figure 6.5 Physiological Signal recorded from a healthy subject breathing spontaneously (Figure from (Eckberg, 1995)). The same group later reported that the strength of the association between respiration and MSNA was a function of blood pressure (Eckberg et al., 1988; Eckberg, 1995). In particular, their results showed a decreased MSNA with an increased in diastolic blood pressure (Figure 6.6). 188 Figure 6.6 Muscle sympathetic nerve activity and respiration at different diastolic pressures (Figure from (Eckberg, 1995)) Although, these studies present the influences of respiration and blood pressure on peripheral blood flow through MSNA, they did not provide a clear quantitative relationship that would allow us to predict the occurrence of peripheral perfusion drops in SCA subjects. Therefore, we propose to explore these relationships using a non-linear, threshold-based mathematical model, using empirical data from our experiments. 6.3.2. Preliminary Results To quantify the relationship between respiration, blood pressure, and perfusion, we first visualized the breath-by-breath relation between inspired tidal volume (ViVol), diastolic blood pressure (DBP), and systemic vascular resistance (SVR). We used SVR to 189 represent the change in blood flow in response to change in blood pressure. In other words, at the same blood pressure level, a higher SVR would result in a lower perfusion. We measured the 3 parameters above continually, and compute a lag-adjusted breath-by- breath values of all parameters for the scatter plot shown by the dots in Figure 6.8. We found the destitution of SVR to be highly non-Gaussian. Therefore we used y = 1/SVR to compute the relationship. We used a 2 nd order regression between the inputs (x 1 = ViVol and x 2 = DBP) and the output (y). The plain in Figure 6.7 was computed by assuming y to be a linear function of x 1 , x 2 , x 1 2 , x 2 2 and x 1 ·x 2 . Figure 6.7 Relationship between x 1 = ViVol, x 2 = DBP and y = 1/SVR. Each point represents values of x 1 , x 2 , and y from a breath. The plane was computed from a 2 nd order regression between inputs (x 1 and x 2 ) and output (y). We then converted this regression plane back to present ViVol, DBP, and SVR as shown in Figure 6.8. 0 2 4 6 8 x 10 -3 30 40 50 60 0 0.5 1 1.5 2 x 10 -3 x 1 x 2 y 190 Figure 6.8 Relationship between ViVol, DBP and SVR. Each point represents values of ViVol, DBP and SVR from a breath. The plane was converted from the plain in Figure 6.7. The sample plot in Figure 6.7 shows that at a high ViVol value, SVR increase with a decrease in DBP. However, this relationship does not seem to be as prominent at a lower ViVol. A non-linear, threshold-based algorithm will be needed to distinguish this relationship in low ViVol and high ViVol regions. 0 1000 2000 3000 4000 30 35 40 45 50 55 60 2000 4000 6000 8000 ViVol DBP SVR 191 Chapter 7. Conclusion This study proposes computational techniques and experimental protocols for non-invasive modeling of the cardiovascular autonomic control in SCD patients. One of the leading causes of death and hospitalization in SCD is VOC, its mechanism is not yet fully understood. Nonetheless, it is know that anything that reduces the transit time of RBCs through the microvasculature or reduce the flow would be likely to increase the risk of VOC. In this study, we hypothesized that the reductions in flow that trigger VOCs could be partially due to the possible abnormality in ANS. Consequently, we set out to explore this rarely investigated area in SCD. To better understand the ANS, we stimulate the subjects’ autonomic control with external perturbations from experimental maneuvers; this included hypoxia and the cold face test. During each experiment the subject’s cardiovascular functions were monitored continuously. Then, we applied our off-line computational techniques to track the autonomic responses to experimental maneuvers. Among these, our time-varying HRV technique estimated moment-to-moment changes in activities of the sympathetic and parasympathetic divisions of the ANS (Chapter 3). Although, this technique is based on what had been reported in the literature (Bianchi et al., 1993; Blasi et al., 2003), we enhanced its computational method to compensate for the respiration patterns, as well as to improve the estimation speed (Sangkatumvong et al., 2008; Sangkatumvong et al., 192 2010). This allowed tracking of fast ANS responses to our autonomic stimuli, while still preserved the robustness of the estimation. While HRV analysis allowed us to observe the balance between the sympathetic and parasympathetic divisions of the ANS, it provides no information regarding the origins of variations in this balance. Researchers have found that the main reflex mechanisms continuously controlling the neural outputs from these two ANS divisions are the arterial baroreflex and the respiratory-cardiac coupling (Khoo, 2008; Belozeroff et al., 2003; Katona, & Jih, 1975; ESC/NASPE, 1996). To continuously monitor the modulations of these mechanisms, we proposed a time-varying minimal model of cardiovascular control (Chapter 4). The results of this model depict continuous changes in the sensitivity of these two mechanisms in response to any perturbations. We applied both of these techniques with experimental data from SCD subjects during autonomic perturbations. Our first perturbation was hypoxia. As nighttime hypoxia has been identified as a risk factor for VOC, we sought to understand the relationship between hypoxia and the possible ANS response that may contribute to VOC by establishing a protocol to mimic transient nighttime hypoxia and continuously monitored the cardiovascular functions in SCD subjects. While we did not find a direct relationship between hypoxia and decreased blood flow, we did find that the ANS response was significantly abnormal in SCD subjects, as assessed using our proposed HRV technique. In particular, compared to the normal control, the parasympathetic 193 withdrawal response in the SCD subjects was hyper-sensitive (Section 5.2). This would make it more likely for SCD subjects to have tachycardia, which could cause more stress to their cardiovascular systems. Moreover, during our hypoxia experiments, we observed that the sighs induced decreases in perfusion, more so in the SCD patients than the normal controls. Literature reviews reveal that this sigh-vasoconstriction event is a reflex which was mediated by the activation of the peripheral sympathetic nervous system (Binet, & Sollier, 1895; Bolton et al., 1936; Baron et al., 1996). Besides this activation of in the periphery, HRV analysis also showed an activation of the cardiac or central sympathetic nervous system (Section 5.3). In particular, our results show a greater sympathetic activation in the SCD patients compared to the CTLs, consistent with our other observation that a sigh is more likely to stimulate a peripheral microvascular perfusion drop in SCD than in normal subjects. The peripheral perfusion and ANS responses to hypoxia and spontaneous sigh presented here suggest an alternative mechanism regarding the relation of hypoxia to VOC. The data suggest that the neural-mediated mechanical signal, and not global hypoxia, is the primary triggering event. We speculate the hypoxia stimulates the chemoreflex activity, leading to hyperventilation, which, in turn, activates the peripheral sympathetic activity. This activation could lead to a decreased peripheral perfusion, which could eventually become a full VOC. 194 These findings of abnormal ANS responses prompted us to further explore their autonomic response to other external stimuli. We started by measuring the autonomic response to the cold face test, as it is one of the most widely used tests of the ANS; in addition, cold weather is also known to trigger VOC. Our results from this test show that the ANS balance shifted in the CTL subjects to become more parasympathetic dominant (or less sympathetic) (Section 5.5). This response is to be expected as the cold face test stimulates the diving reflex, which is known to activate the parasympathetic system. Nonetheless, we did not observe this response in SCD patients, regardless whether they had been treated with chronic transfusion or not. This suggests impairment in their cardiac parasympathetic nervous systems. This result agrees with our results from the controlled hypoxia and spontaneous sigh experiments. In addition to assessing the cardiovascular control of SCD subjects using the HRV analysis, we also applied the minimal model to assess the reflex mechanisms which control the HRV. We applied this model to assess the source of the abnormality of the HRV responses to cold face stimulus (Section 5.6). This model showed a tendency for both ABR and RCC sensitivities to increase during the CFT in the CTLs, but not in the SCD subjects. We speculate that this absence in ABR and RCC responses in SCD subjects may be responsible for their absence of the parasympathetic shift in HRV, as shown in normal controls. However, due to the large inter-subject variability, statistical significance of these differences was not attained. 195 Although the central ANS showed an increase in the parasympathetic activity in response to all of our stimuli (hypoxia, sigh, and cold face), the peripheral ANS showed instead an increase in the peripheral sympathetic nerve activity, as presented by reduced peripheral perfusion during sigh and CFT (Section 5.3 and 5.5). Although immediate CFT perfusion response in SCD subjects did not differ from the controls, we observed another episode of reduced perfusion only in SCD subjects (Figure 5.15A). We suspect that this second episode of perfusion drop after the cold pack was removed might be due to a vasoconstriction response to hyperventilation. In addition to the ANS responses to the stimuli, we assessed the HRV as well as the ABR and RCC sensitivities during the subjects’ resting baseline conditions. Although no difference in any parameters was observed between the SCD patients and the controls (Section 5.1 and 5.4), we found that blood transfusion increased the parasympathetic modulation in SCD patients. This suggests that, during their stable baselines, SCD patients are able to compensate their ANS for the anemia and other pathologies, and that their ANS functions are improved after their blood transfusions. Even though their ANS balances appeared to be normal during baselines, the ANS responses in SCD subjects during the application of all three stimuli (hypoxia, sigh, and cold face) were abnormal. To better understand the pathological consequences of SCD to these abnormal ANS responses, we correlated some parameters from blood tests regularly used to diagnose SCD with ANS responses to hypoxia (Section 5.7). Our results suggest that the 196 degree of hyper-sensitivity of the ANS responses to hypoxia in SCD patients is related to both their anemia and their level of inflammation. We speculate that this relationship might be the result of an over-activation of the ‘cholinergic anti-inflammatory pathway’ (Tracey, 2002; Borovikova et al., 2000). In summary, our results suggest a link in SCD patients between the ANS response to perturbations and drops in perfusion that could increase microvascular transit time and thereby promote VOC. The ANS responses to stimuli in SCD patients also tended to shift more easily toward the sympathetic division, i.e. parasympathetic withdrawal and sympathetic activation, compared to the normal controls. This is likely to cause their cardiovascular system to constantly be in a hyperactive state, adding more stress to and promoting other cardiovascular complications in SCD patients. 197 References AHMED, S., SIDDIQUI, A. K., SADIQ, A., SHAHID, R. K., PATEL, D. V., and RUSSO, L. A. (2004) Echocardiographic abnormalities in sickle cell disease. Am J Hematol, 76, 195-8. ALLEN, J., FRAME, J. R., and MURRAY, A. (2002) Microvascular blood flow and skin temperature changes in the fingers following a deep nspiratory gasp. Physiol Meas, 23, 365-73. ANGELONE, A., and COULTER, N. A., JR. (1964) Respiratory Sinus Arrhythemia: a Frequency Dependent Phenomenon. J Appl Physiol, 19, 479-82. ASSANASEN, C., QUINTON, R. A., and BUCHANAN, G. R. (2003) Acute myocardial infarction in sickle cell anemia. J Pediatr Hematol Oncol, 25, 978-81. ASYALI, M. H., and JUUSOLA, M. (2005) Use of meixner functions in estimation of Volterra kernels of nonlinear systems with delay. IEEE Trans Biomed Eng, 52, 229-37. BARON, R., FELDMANN, R., and LINDNER, V. (1993) Small fibre function in primary autonomic failure. J Neurol, 241, 87-91. BARON, R., HABLER, H. J., HECKMANN, K., and PORSCHKE, H. (1996) Respiratory modulation of blood flow in normal and sympathectomized skin in humans. J Auton Nerv Syst, 60, 147-53. BARRON, A., RISSANEN, J., and YU, B. (1998) The Minimum Description Length Principle in Coding and Modeling. IEEE TRANSACTIONS ON INFORMATION THEORY, 44, 2743-2760. BATRA, A. S., ACHERMAN, R. J., WONG, W. Y., WOOD, J. C., CHAN, L. S., RAMICONE, E., EBRAHIMI, M., and WONG, P. C. (2002) Cardiac abnormalities in children with sickle cell anemia. Am J Hematol, 70, 306-12. BAUM, K. F., DUNN, D. T., MAUDE, G. H., and SERJEANT, G. R. (1987) The painful crisis of homozygous sickle cell disease. A study of the risk factors. Arch Intern Med, 147, 1231-4. 198 BECKLAKE, M. R., GRIFFITHS, S. B., MC, G. M., GOLDMAN, H. I., and SCHREVE, J. P. (1955) Oxygen dissociation curves in sickle cell anemia and in subjects with the sickle cell trait. J Clin Invest, 34, 751-5. BELOZEROFF, V., BERRY, R. B., and KHOO, M. C. (2003) Model-based assessment of autonomic control in obstructive sleep apnea syndrome. Sleep, 26, 65-73. BERNE, R. M., and LEVY, M. N. (2001) Cardiovascular physiology, St. Louis, MO, Mosby. BESEDOVSKY, H., DEL REY, A., SORKIN, E., DA PRADA, M., BURRI, R., and HONEGGER, C. (1983) The immune response evokes changes in brain noradrenergic neurons. Science, 221, 564-6. BIANCHI, A. M., MAINARDI, L., PETRUCCI, E., SIGNORINI, M. G., MAINARDI, M., and CERUTTI, S. (1993) Time-variant power spectrum analysis for the detection of transient episodes in HRV signal. IEEE Trans Biomed Eng, 40, 136- 44. BINET, A., and SOLLIER, P. (1895) Recherches sur le pouls cérébral. Archives de physiol. Normale et Pathol., 7, 719-734. BLASI, A., JO, J., VALLADARES, E., MORGAN, B. J., SKATRUD, J. B., and KHOO, M. C. (2003) Cardiovascular variability after arousal from sleep: time-varying spectral analysis. J Appl Physiol, 95, 1394-404. BLASI, A., JO, J. A., VALLADARES, E., JUAREZ, R., BAYDUR, A., and KHOO, M. C. (2006) Autonomic cardiovascular control following transient arousal from sleep: a time-varying closed-loop model. IEEE Trans Biomed Eng, 53, 74-82. BOLTON, B., CARMICHAEL, E. A., and STURUP, G. (1936) Vaso-constriction following deep inspiration. J Physiol, 86, 83-94. BOROVIKOVA, L. V., IVANOVA, S., ZHANG, M., YANG, H., BOTCHKINA, G. I., WATKINS, L. R., WANG, H., ABUMRAD, N., EATON, J. W., and TRACEY, K. J. (2000) Vagus nerve stimulation attenuates the systemic inflammatory response to endotoxin. Nature, 405, 458-62. BRINKER, A. C. D. (1995) Meixner-like functions having a rational z-transform. International Journal of Circuit Theory and Applications, 23, 237-246. 199 BUCHHEIT, M., RICHARD, R., DOUTRELEAU, S., LONSDORFER-WOLF, E., BRANDENBERGER, G., and SIMON, C. (2004) Effect of acute hypoxia on heart rate variability at rest and during exercise. Int J Sports Med, 25, 264-9. CERUTTI, S., BIANCHI, A. M., and MAINARDI, L. T. (2001) Advanced spectral methods for detecting dynamic behaviour. Auton Neurosci, 90, 3-12. CHAICHARN, J., CARRINGTON, M., TRINDER, J., and KHOO, M. C. (2008) The effects on cardiovascular autonomic control of repetitive arousal from sleep. Sleep, 31, 93-103. CHAICHARN, J., LIN, Z., CHEN, M. L., KEENS, T. G., DAVIDSON WARD, S. L., and KHOO, M. K. (2006) Time-varying closed-loop modeling of autonomic control in pediatric obstructive sleep apnea syndrome during cold face stimulation. Conf Proc IEEE Eng Med Biol Soc, 1, 3569-71. COHEN, J. (1988) Statistical power analysis for the behavioral sciences, Hillsdale, N.J., L. Erlbaum Associates. CONNES, P., HUE, O., TRIPETTE, J., and HARDY-DESSOURCES, M. D. (2008) Blood rheology abnormalities and vascular cell adhesion mechanisms in sickle cell trait carriers during exercise. Clin Hemorheol Microcirc, 39, 179-84. CONNES, P., MARTIN, C., BARTHELEMY, J.-C., MONCHANIN, G., ATCHOU, G., FORSUH, A., MASSARELLI, R., WOUASSI, D., TRIRIET, P., and PICHOT, V. (2006) Nocturnal autonomic nervous system activity impairment in sickle cell trait carriers. Clin Physiol Funct Imaging, 26, 87-91. COX, D. A., VITA, J. A., TREASURE, C. B., FISH, R. D., ALEXANDER, R. W., GANZ, P., and SELWYN, A. P. (1989) Atherosclerosis impairs flow-mediated dilation of coronary arteries in humans. Circulation, 80, 458-65. CURTIS, B. M., and O'KEEFE, J. H., JR. (2002) Autonomic tone as a cardiovascular risk factor: the dangers of chronic fight or flight. Mayo Clin Proc, 77, 45-54. DAMPNEY, R. A. (1994) Functional organization of central pathways regulating the cardiovascular system. Physiol Rev, 74, 323-64. DARBARI, D. S., KPLE-FAGET, P., KWAGYAN, J., RANA, S., GORDEUK, V. R., and CASTRO, O. (2006) Circumstances of death in adult sickle cell disease patients. Am J Hematol, 81, 858-63. 200 DE MONTALEMBERT, M., AGGOUN, Y., NIAKATE, A., SZEZEPANSKI, I., and BONNET, D. (2007) Endothelial-dependent vasodilation is impaired in children with sickle cell disease. Haematologica, 92, 1709-10. DEYMANN, A. J., and GOERTZ, K. K. (2003) Myocardial infarction and transient ventricular dysfunction in an adolescent with sickle cell disease. Pediatrics, 111, E183-7. DUFFIN, J., MOHAN, R. M., VASILIOU, P., STEPHENSON, R., and MAHAMMED, S. (2000) A model of the chemoreflex control of breathing in humans: model parameters measurement. Respiratory Physiology. DYSON, K. S., SHOEMAKER, J. K., and HUGHSON, R. L. (2006) Effect of acute sympathetic nervous system activation on flow-mediated dilation of brachial artery. Am J Physiol Heart Circ Physiol, 290, H1446-53. EATON, W. A., and HOFRICHTER, J. (1995) The biophysics of sickle cell hydroxyurea therapy. Science, 268, 1142-3. ECKBERG, D. L. (1983) Human sinus arrhythmia as an index of vagal cardiac outflow. J Appl Physiol, 54, 961-6. ECKBERG, D. L. (1995) Respiratory sinus arrhythmia and other human cardiovascular neural periodicities. IN DEMPSEY, J. A. & PACK, A. I. (Eds.) Regulation of breathing. 2nd ed. New York:, Marcel Dekker. ECKBERG, D. L. (1997) Sympathovagal balance: a critical appraisal. Circulation, 96, 3224-32. ECKBERG, D. L., NERHED, C., and WALLIN, B. G. (1985) Respiratory modulation of muscle sympathetic and vagal cardiac outflow in man. J Physiol, 365, 181-96. ECKBERG, D. L., REA, R. F., ANDERSSON, O. K., HEDNER, T., PERNOW, J., LUNDBERG, J. M., and WALLIN, B. G. (1988) Baroreflex modulation of sympathetic activity and sympathetic neurotransmitters in humans. Acta Physiol Scand, 133, 221-31. ESC/NASPE (1996) Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Eur Heart J, 17, 354-81. 201 FRENETTE, P. S., and ATWEH, G. F. (2007) Sickle cell disease: old discoveries, new concepts, and future promise. J Clin Invest, 117, 850-8. GLADWIN, M. T., SCHECHTER, A. N., OGNIBENE, F. P., COLES, W. A., REITER, C. D., SCHENKE, W. H., CSAKO, G., WACLAWIW, M. A., PANZA, J. A., and CANNON, R. O., 3RD (2003) Divergent nitric oxide bioavailability in men and women with sickle cell disease. Circulation, 107, 271-8. GROSSMAN, P., KAREMAKER, J., and WIELING, W. (1991) Prediction of tonic parasympathetic cardiac control using respiratory sinus arrhythmia: the need for respiratory control. Psychophysiology, 28, 201-16. HAINSWORTH, R. (1991) Reflexes from the heart. Physiol Rev, 71, 617-58. HARGRAVE, D. R., WADE, A., EVANS, J. P., HEWES, D. K., and KIRKHAM, F. J. (2003) Nocturnal oxygen saturation and painful sickle cell crises in children. Blood, 101, 846-8. HATHAWAY, D. K., CASHION, A. K., MILSTEAD, E. J., WINSETT, R. P., COWAN, P. A., WICKS, M. N., and GABER, A. O. (1998) Autonomic dysregulation in patients awaiting kidney transplantation. Am J Kidney Dis, 32, 221-9. HEATH, M. E., and DOWNEY, J. A. (1990) The cold face test (diving reflex) in clinical autonomic assessment: methodological considerations and repeatability of responses. Clin Sci (Lond), 78, 139-47. HEBBEL, R. P., OSAROGIAGBON, R., and KAUL, D. (2004) The endothelial biology of sickle cell disease: inflammation and a chronic vasculopathy. Microcirculation, 11, 129-51. HEDREVILLE, M., BARTHELEMY, J. C., TRIPETTE, J., ROCHE, F., HARDY- DESSOURCES, M. D., PICHOT, V., HUE, O., and CONNES, P. (2008) Effects of strenuous exercise on autonomic nervous system activity in sickle cell trait carriers. Auton Neurosci, 143, 68-72. HEISTAD, D. D., ABBOUND, F. M., and ECKSTEIN, J. W. (1968) Vasoconstrictor response to simulated diving in man. J Appl Physiol, 25, 542-9. HIJMERING, M. L., STROES, E. S., OLIJHOEK, J., HUTTEN, B. A., BLANKESTIJN, P. J., and RABELINK, T. J. (2002) Sympathetic activation markedly reduces endothelium-dependent, flow-mediated vasodilation. J Am Coll Cardiol, 39, 683- 8. 202 INGRAM, V. M. (1959) Abnormal human haemoglobins. III. The chemical difference between normal and sickle cell haemoglobins. Biochim Biophys Acta, 36, 402-11. INWALD, D., HATHORN, M. K., and COSTELOE, K. (1996) The deep breath vasoconstriction reflex--a new tool for autonomic assessment in infancy? Early Hum Dev, 45, 55-61. ITANO, H. A., and PAULING, L. (1949) A rapid diagnostic test for sickle cell anemia. Blood, 4, 66-8. IWASAKI, K., OGAWA, Y., AOKI, K., SAITOH, T., OTSUBO, A., and SHIBATA, S. (2006) Cardiovascular regulation response to hypoxia during stepwise decreases from 21% to 15% inhaled oxygen. Aviat Space Environ Med, 77, 1015-9. JAJA, S. I., KEHINDE, M. O., and OGUNGBEMI, S. I. (2008) Cardiac and autonomic responses to change in posture or vitamin C supplementation in sickle cell anemia subjects. Pathophysiology, 15, 25-30. JAMES, T. N., RIDDICK, L., and MASSING, G. K. (1994) Sickle cells and sudden death: morphologic abnormalities of the cardiac conduction system. J Lab Clin Med, 124, 507-20. JO, J. A., BLASI, A., VALLADARES, E., JUAREZ, R., BAYDUR, A., and KHOO, M. C. (2003) Model-based assessment of autonomic control in obstructive sleep apnea syndrome during sleep. Am J Respir Crit Care Med, 167, 128-36. JO, J. A., BLASI, A., VALLADARES, E. M., JUAREZ, R., BAYDUR, A., and KHOO, M. C. (2007) A Nonlinear Model of Cardiac Autonomic Control in Obstructive Sleep Apnea Syndrome. Ann Biomed Eng. JOANNIDES, R., HAEFELI, W. E., LINDER, L., RICHARD, V., BAKKALI, E. H., THUILLEZ, C., and LUSCHER, T. F. (1995) Nitric oxide is responsible for flow- dependent dilatation of human peripheral conduit arteries in vivo. Circulation, 91, 1314-9. KANDEL, E. R., SCHWARTZ, J. H., and JESSELL, T. M. (2000) Principle of Neural Science, McGraw-Hill. KATONA, P. G., and JIH, F. (1975) Respiratory sinus arrhythmia: noninvasive measure of parasympathetic cardiac control. J Appl Physiol, 39, 801-5. 203 KATONA, P. G., POITRAS, J. W., BARNETT, G. O., and TERRY, B. S. (1970) Cardiac vagal efferent activity and heart period in the carotid sinus reflex. Am J Physiol, 218, 1030-7. KAWAKAMI, Y., NATELSON, B. H., and DUBOIS, A. R. (1967) Cardiovascular effects of face immersion and factors affecting diving reflex in man. J Appl Physiol, 23, 964-70. KAY, S. M. (1988) Modern spectral estimation : theory and application, Englewood Cliffs, N.J., Prentice Hall. KELM, M. (2002) Flow-mediated dilatation in human circulation: diagnostic and therapeutic aspects. Am J Physiol Heart Circ Physiol, 282, H1-5. KENNEDY, A. P., WILLIAMS, B., MEYDRECH, E. F., and STEINBERG, M. H. (1988) Regional and temporal variation in oscillatory blood flow in sickle cell disease. Am J Hematol, 28, 92-4. KHOO, M. C. (2008) Modeling of autonomic control in sleep-disordered breathing. Cardiovasc Eng, 8, 30-41. KHOO, M. C., KIM, T. S., and BERRY, R. B. (1999) Spectral indices of cardiac autonomic function in obstructive sleep apnea. Sleep, 22, 443-51. KHURANA, R. K., and WU, R. (2006) The cold face test: a non-baroreflex mediated test of cardiac vagal function. Clin Auton Res, 16, 202-7. KIRKHAM, F. J., HEWES, D. K., PRENGLER, M., WADE, A., LANE, R., and EVANS, J. P. (2001) Nocturnal hypoxaemia and central-nervous-system events in sickle-cell disease. Lancet, 357, 1656-9. LESTER, L. A., SODT, P. C., HUTCHEON, N., and ARCILLA, R. A. (1990) Cardiac abnormalities in children with sickle cell anemia. Chest, 98, 1169-74. LIN, Z., CHEN, M. L., KEENS, T. G., WARD, S. L., and KHOO, M. C. (2004) Noninvasive assessment of cardiovascular autonomic control in congenital central hypoventilation syndrome. Conf Proc IEEE Eng Med Biol Soc, 5, 3870-3. LIPOWSKY, H. H., SHEIKH, N. U., and KATZ, D. M. (1987) Intravital microscopy of capillary hemodynamics in sickle cell disease. J Clin Invest, 80, 117-27. LJUNG, L. (1999) System Identification: Theory for the User. Upper Saddle River, NJ, Prentice Hall PTR. 204 MANCI, E. A., CULBERSON, D. E., YANG, Y. M., GARDNER, T. M., POWELL, R., HAYNES, J., JR., SHAH, A. K., and MANKAD, V. N. (2003) Causes of death in sickle cell disease: an autopsy study. Br J Haematol, 123, 359-65. MARMARELIS, V. Z. (1993) Identification of nonlinear biological systems using Laguerre expansions of kernels. Ann Biomed Eng, 21, 573-89. MARMARELIS, V. Z. (1997) Modeling methodology for nonlinear physiological systems. Ann Biomed Eng, 25, 239-51. MARMARELIS, V. Z. (2004) Nonlinear dynamic modeling of physiological systems, Hoboken, N.J., Wiley-Interscience. MARSHALL, J. M. (1994) Peripheral Chemoreceptors and Cardiovascular Regulation. Physiological Reviews, 74. MILLER, S. T., SLEEPER, L. A., PEGELOW, C. H., ENOS, L. E., WANG, W. C., WEINER, S. J., WETHERS, D. L., SMITH, J., and KINNEY, T. R. (2000) Prediction of adverse outcomes in children with sickle cell disease. N Engl J Med, 342, 83-9. MUKKAMALA, R., KIM, J. K., LI, Y., SALA-MERCADO, J., HAMMOND, R. L., SCISLO, T. J., and O'LEARY, D. S. (2006) Estimation of arterial and cardiopulmonary total peripheral resistance baroreflex gain values: validation by chronic arterial baroreceptor denervation. Am J Physiol Heart Circ Physiol, 290, H1830-6. NATH, K. A., KATUSIC, Z. S., and GLADWIN, M. T. (2004) The perfusion paradox and vascular instability in sickle cell disease. Microcirculation, 11, 179-93. NEFF, R. A., WANG, J., BAXI, S., EVANS, C., and MENDELOWITZ, D. (2003) Respiratory sinus arrhythmia: endogenous activation of nicotinic receptors mediates respiratory modulation of brainstem cardioinhibitory parasympathetic neurons. Circ Res, 93, 565-72. NORRIS, S., JOHNSON, C. S., and HAYWOOD, L. J. (1991) Sickle cell anemia: does myocardial ischemia occur during crisis? J Natl Med Assoc, 83, 209-13. O'BRIEN, L. M., and GOZAL, D. (2005) Autonomic dysfunction in children with sleep- disordered breathing. Sleep, 28, 747-52. OGURA, H. (1985) Estimation of Wiener kernels of a nonlinear system and a fast algorithm using digital Laguerre filters. 15th NIBB Conference. Okazaki, Japan. 205 PARKER, R. S., DOYLE, F. J., 3RD, and PEPPAS, N. A. (1999) A model-based algorithm for blood glucose control in type I diabetic patients. IEEE Trans Biomed Eng, 46, 148-57. PAULING, L., ITANO, H. A., and ET AL. (1949) Sickle cell anemia, a molecular disease. Science, 109, 443. PAVLU, J., AHMED, R. E., O'REGAN, D. P., PARTRIDGE, J., LEFROY, D. C., and LAYTON, D. M. (2007) Myocardial infarction in sickle-cell disease. Lancet, 369, 246. PEARSON, S. R., ALKON, A., TREADWELL, M., WOLFF, B., QUIROLO, K., and BOYCE, W. T. (2005) Autonomic reactivity and clinical severity in children with sickle cell disease. Clin Auton Res, 15, 400-7. PERRONNE, V., ROBERTS-HAREWOOD, M., BACHIR, D., ROUDOT- THORAVAL, F., DELORD, J. M., THURET, I., SCHAEFFER, A., DAVIES, S. C., GALACTEROS, F., and GODEAU, B. (2002) Patterns of mortality in sickle cell disease in adults in France and England. Hematol J, 3, 56-60. PERSSON, P. B. (1996) Modulation of cardiovascular control mechanisms and their interaction. Physiol Rev, 76, 193-244. PIEPOLI, M. F., and CAPUCCI, A. (2007) Autonomic nervous system in the genesis of arrhythmias in chronic heart failure: implication for risk stratification. Minerva Cardioangiol, 55, 325-33. PLATT, O. S., BRAMBILLA, D. J., ROSSE, W. F., MILNER, P. F., CASTRO, O., STEINBERG, M. H., and KLUG, P. P. (1994) Mortality in sickle cell disease. Life expectancy and risk factors for early death. N Engl J Med, 330, 1639-44. QUINN, C. T., and AHMAD, N. (2005) Clinical correlates of steady-state oxyhaemoglobin desaturation in children who have sickle cell disease. Br J Haematol, 131, 129-34. QUINN, C. T., SHULL, E. P., AHMAD, N., LEE, N. J., ROGERS, Z. R., and BUCHANAN, G. R. (2007) Prognostic significance of early vaso-occlusive complications in children with sickle cell anemia. Blood, 109, 40-5. RAUH, R., POSFAY, A., and MUCK-WEYMANN, M. (2003) Quantification of inspiratory-induced vasoconstrictive episodes: a comparison of laser Doppler fluxmetry and photoplethysmography. Clin Physiol Funct Imaging, 23, 344-8. 206 RODGERS, G. P., SCHECHTER, A. N., NOGUCHI, C. T., KLEIN, H. G., NIENHUIS, A. W., and BONNER, R. F. (1990) Microcirculatory adaptations in sickle cell anemia: reactive hyperemia response. Am J Physiol, 258, H113-20. ROMERO-VECCHIONE, E., VASQUEZ, J., PEREZ, O., and ANTEQUERA, R. (2002) Blunted hemodynamic and hormonal responses to tilt-test in patients with sickle- cell disease. International Congress Series, 1237, 169-177. ROMERO MESTRE, J. C., HERNANDEZ, A., AGRAMONTE, O., and HERNANDEZ, P. (1997) Cardiovascular autonomic dysfunction in sickle cell anemia: a possible risk factor for sudden death? Clin Auton Res, 7, 121-5. SANGKATUMVONG, S., COATES, T. D., and KHOO, M. C. (2008) Abnormal autonomic cardiac response to transient hypoxia in sickle cell anemia. Physiol Meas, 29, 655-68. SANGKATUMVONG, S., COATES, T. D., WOOD, J. C., MEISELMAN, H. J., KATO, R., DETTERICH, J. A., BUSH, A., and KHOO, M. C. (2010) Time-varying analysis of autonomic control in response to spontaneous sighs in sickle cell anemia. Conf Proc IEEE Eng Med Biol Soc, 1, 1626-9. SATHYAPRABHA, T. N., SATISHCHANDRA, P., NETRAVATHI, K., SINHA, S., THENNARASU, K., and RAJU, T. R. (2006) Cardiac autonomic dysfunctions in chronic refractory epilepsy. Epilepsy Res, 72, 49-56. SAUL, J. P., BERGER, R. D., CHEN, M. H., and COHEN, R. J. (1989) Transfer function analysis of autonomic regulation. II. Respiratory sinus arrhythmia. Am J Physiol, 256, H153-61. SERJEANT, G. R., CEULAER, C. D., LETHBRIDGE, R., MORRIS, J., SINGHAL, A., and THOMAS, P. W. (1994) The painful crisis of homozygous sickle cell disease: clinical features. Br J Haematol, 87, 586-91. SHIAVI, R. (2007) Introduction to applied statistical signal analysis : guide to biomedical and electrical engineering applications, Oxford, Academic. SHOEMAKER, J. K. (2004) Peripheral Chemoreceptor Contributions to Cardiovascular Regulation. Heart Drug. SLEIGHT, P., LA ROVERE, M. T., MORTARA, A., PINNA, G., MAESTRI, R., LEUZZI, S., BIANCHINI, B., TAVAZZI, L., and BERNARDI, L. (1995) Physiology and pathophysiology of heart rate and blood pressure variability in 207 humans: is power spectral analysis largely an index of baroreflex gain? Clin Sci (Lond), 88, 103-9. SOMERS, V. K., DYKEN, M. E., CLARY, M. P., and ABBOUD, F. M. (1995) Sympathetic neural mechanisms in obstructive sleep apnea. J Clin Invest, 96, 1897-904. SOMERS, V. K., MARK, A. L., ZAVALA, D. C., and ABBOUD, F. M. (1989) Influence of ventilation and hypocapnia on sympathetic nerve responses to hypoxia in normal humans. J Appl Physiol, 67, 2095-100. STAUSS, H. M., ANDERSON, E. A., HAYNES, W. G., and KREGEL, K. C. (1998) Frequency response characteristics of sympathetically mediated vasomotor waves in humans. Am J Physiol, 274, H1277-83. STEMPER, B., HILZ, M. J., RAUHUT, U., and NEUNDORFER, B. (2002) Evaluation of cold face test bradycardia by means of spectral analysis. Clin Auton Res, 12, 78-83. THARAUX, P. L., GIROT, R., KANFER, A., DUSSAULE, J. C., GAITZ, J. P., TRIBOUT, L., BAUDOT, N., and VAYSSAIRAT, M. (2002) Cutaneous microvascular blood flow and reactivity in patients with homozygous sickle cell anaemia. Eur J Haematol, 68, 327-31. TRACEY, K. J. (2002) The inflammatory reflex. Nature, 420, 853-9. UONG, E. C., BOYD, J. H., and DEBAUN, M. R. (2006) Daytime pulse oximeter measurements do not predict incidence of pain and acute chest syndrome episodes in sickle cell anemia. J Pediatr, 149, 707-9. VALADI, N., SILVA, G. S., BOWMAN, L. S., RAMSINGH, D., VICARI, P., FILHO, A. C., MASSARO, A. R., KUTLAR, A., NICHOLS, F. T., and ADAMS, R. J. (2006) Transcranial Doppler ultrasonography in adults with sickle cell disease. Neurology, 67, 572-4. WARNER, H. R., and COX, A. (1962) A mathematical model of heart rate control by sympathetic and vagus efferent information. J Appl Physiol, 17, 349-55. WATKINS, L. R., and MAIER, S. F. (2002) Beyond neurons: evidence that immune and glial cells contribute to pathological pain states. Physiol Rev, 82, 981-1011. 208 WEISKOPF, R. B., FEINER, J., HOPF, H., VIELE, M. K., WATSON, J. J., LIEBERMAN, J., KELLEY, S., and TOY, P. (2003) Heart rate increases linearly in response to acute isovolemic anemia. Transfusion, 43, 235-40. WIENER, N. (1958) Nonlinear problems in random theory, [Cambridge], Technology Press of Massachusetts Institute of Technology. ZAPOL, W. M., HILL, R. D., QVIST, J., FALKE, K., SCHNEIDER, R. C., LIGGINS, G. C., and HOCHACHKA, P. W. (1989) Arterial gas tensions and hemoglobin concentrations of the freely diving Weddell seal. Undersea Biomed Res, 16, 363- 73. ZHANG, Y., POPOVIC, Z. B., BIBEVSKI, S., FAKHRY, I., SICA, D. A., VAN WAGONER, D. R., and MAZGALEV, T. N. (2009) Chronic vagus nerve stimulation improves autonomic control and attenuates systemic inflammation and heart failure progression in a canine high-rate pacing model. Circ Heart Fail, 2, 692-9. ZOCCAL, D. B., BONAGAMBA, L. G. H., OLIVEIRA, F. R. T., ANTENES- RODRIGUES, J., and MACHADO, B. H. (2007) Increased sympathetic activity in rats submitted to chronic intermittent hypoxia. Experimental Physiology, 79-85.

Asset Metadata

Creator Sangkatumvong, Suvimol "Ming" (author)

Core Title Modeling of cardiovascular autonomic control in sickle cell disease

Contributor Electronically uploaded by the author (provenance)

School Andrew and Erna Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Biomedical Engineering

Publication Date 04/26/2011

Defense Date 03/03/2011

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag autonomic nervous system,baroreflex,cardiovascular autonomic control,cold face test,heart rate variability,hypoxia,minimal model,OAI-PMH Harvest,parasympathetic,physiological system modeling,respiration,respiratory-cardiac coupling,sickle cell disease,sigh,sympathetic

Place Name California (states), Los Angeles (city or populated place), medical facilities: Childrens' Hospital Los Angeles (geographic subject), USA (countries)

Language English

Advisor Coates, Thomas (committee chair), Khoo, Michael C.K. (committee chair), Meiselman, Herbert J. (committee member), Wood, John C. (committee member)

Creator Email mingsuvimol@hotmail.com,sangkatu@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3781

Unique identifier UC1136204

Identifier etd-Sangkatumvong-4436 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-468285 (legacy record id),usctheses-m3781 (legacy record id)

Legacy Identifier etd-Sangkatumvong-4436.pdf

Dmrecord 468285

Document Type Dissertation

Rights Sangkatumvong, Suvimol "Ming"

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract Sickle cell disease (SCD) is a genetic disorder that is characterized by recurrent episodes of vaso-occlusive crisis (VOC) from the sickling behavior of red blood cells. Currently, no technique can distinguish the cause or predict the occurrence of a crisis accurately and reliably. One area which has rarely been studied in SCD patients is their autonomic nervous system (ANS). Since the ANS is responsible for the moment-to-moment control of the vascular tone, we hypothesized that the ANS plays an important role in the initiation of their VOC. Computational techniques, including spectral analysis of HRV and a model which characterizes the dynamics of baroreflex and respiratory-cardiac coupling, were used to assess cardiovascular autonomic control in SCD patients and normal control (CTL) subjects. These analysis techniques were applied to responses elicited from the subjects during the application of non-invasive and easily reproducible physiological interventions, such as transient-controlled hypoxia and the cold face test.