University of Southern California Dissertations and Theses

Post-intensive care unit mechanical ventilation: Relationship of infections to outcomes of weaning from prolonged mechanical ventilation

(USC Thesis Other)

Post-intensive care unit mechanical ventilation: Relationship of infections to outcomes of weaning from prolonged mechanical ventilation

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content POST-INTENSIVE CARE UNIT MECHANICAL VENTILATION:
RELATIONSHIP OF INFECTIONS TO OUTCOMES OF WEANING FROM
PROLONGED MECHANICAL VENTILATION
By
Hanjoo Kim
A Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(APPLIED BIOSTATISTICS AND EPIDEMIOLOGY)
August 2005
Copyright 2005 Hanjoo Kim
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 1430394
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
®
UMI
UMI Microform 1430394
Copyright 2006 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DEDICATION
For my parents, unconditional love.
For my wife, Seunghye, and my daughter, Doyeon, fountain of love.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
I would like to acknowledge my thesis committee for their sincere and dedicated
guidance and discipline on my young endeavor.
Stanley P. Azen, Ph.D.
Arif M. Ansari, Ph.D.
David J. Scheinhom, M.D., FACP, FCCP
I would also like to express my much gratitude for their devoted teachings and
extensive support.
Meg A. Hassenpflug, M.S., R.D.
Donguk Kim, Ph.D.
Laurie D. LaBree, M.A.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List of Tables v
Abstract vi
1. Overview 1
2. Traditional Analytic Approach
2.1 Introduction 2
2.2 Methods 3
2.3 Results 6
2.4 Discussion 13
3. Data Mining Approach
3.1 Overview of Decision Tree Methodology 18
3.2 Application to Barlow Respiratory Hospital Dataset 20
3.3 Discussion 24
4. Bibliography 25
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Table 2-la. Baseline Sociodemographic and Clinical Variables at 6
Admission
Table 2-lb. Outcome Variables during Hospitalization 7
Table 2-2. Length of Stay and Time to Wean by Treatment for Primary 8
Infection
Table 2-3. The Association between Death in Relation to the a) Number 9
of Treated Infections, b) Type of Infection, and c) Type of
Isolation
Table 2-4. The Association between Diabetes Mellitus and 10
Treatment of Primary Infections
Table 2-5. The Risk Factors for Treated a) Primary and b) Sepsis 11
Table 2-6. The Association between Primary Infections and Sepsis 12
Table 3-1. Comparison of Likelihood by Split Point 22
Figure 3-1. JMP Decision Tree Output 21
V
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
Objective: To report the effects on outcome of selected infections in ventilator-
dependent patients transferred to Barlow Respiratory Hospital for weaning from
prolonged mechanical ventilation in the post-intensive care unit setting.
Methods: The occurrence and type of clinical infections were obtained from 186
patients enrolled from 3/1/02 -2/28/03. The data were abstracted from patients’
medical records. Statistical analyses were conducted to investigate 1) the
differences in length of stay and time to wean between patients with or without
infections, 2) the associations between death and patients with or without infections,
and 3) risk factors with occurrence of infections.
Results: Length of stay (p < 0.05) and time to wean (p = 0.03) were longer for
patients with infections than for those without. The number of deaths increased in
patients with pneumonia (p < 0.01), or line sepsis (p < 0.01).
Conclusions: Infections negatively impacted outcome, length of stay, and time to
wean.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1. OVERVIEW
This study utilizes a data set from Barlow Respiratory Hospital on 186
ventilator-dependent patients receiving invasive mechanical ventilation admitted for
weaning. The study presents two analytic approaches. First is a traditional
analytic approach utilizing logistic regressions in the style of a peer review paper.
The second introduces an overview of data mining methodology, and presents an
example utilizing decision tree strategies for two of the risk factors found in the
traditional analytic approach.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2. TRADITIONAL ANALYTIC APPROACH
2.1 INTRODUCTION
Approximately 40% of patients who enter an intensive care unit in North
America require mechanical ventilatory support to treat respiratory failure;1 only 5-
20% require prolonged mechanical ventilation.2 ,3 Factors that influence
transferring patients out of the intensive care unit are cost and bed scarcity.
Infection in the intensive care unit, often with antibiotic resistant organisms, has
been shown to adversely affect numerous intensive care unit outcomes. When
ventilator-dependent patients are transferred to a post-intensive care unit for
continued weaning efforts, the risk of complications continues. There is
preliminary data from a multicenter study that the most prevalent complications at
long-term acute-care hospitals are infectious.4
This study was undertaken at Barlow Respiratory Hospital to measure the
effect of infectious complications on outcomes in an active post-intensive care unit
weaning program, and to seek clinically relevant risk indicators for infection.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.2 METHODS
Study cohort: The study cohort consisted o f adult (^1 8 years of age) ventilator-
dependent patients receiving invasive mechanical ventilation admitted for weaning.
Data were collected as part of an observational multicenter study without
interventions, conducted from March 1, 2002 to February 28, 2003. Patients were
excluded if they were ventilator-dependent patients not admitted for weaning: end-
of-life care or terminal weaning, home ventilator training, chronically ventilated
patients admitted for treatment of an intercurrent medical problem, or not a weaning
candidate as documented by the physician on admission.
The Institutional Review Board of Barlow Respiratory Hospital approved
this study, for which David J. Scheinhom, M.D., FACP, FCCP as Principal
Investigator, on October 25, 2001. The need for informed consent was waived, as
this is an observational study, with no new treatments or interventions offered to
patients. The issues of de-identification of data, and patient confidentiality, were
addressed to the complete satisfaction of the Board.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Study data: The initial datasets were collected at the admission to Barlow
Respiratory Hospital. Baseline sociodemographic and clinical variables at
admission included: gender, age, prior episode of mechanical ventilation, premorbid
functional status, diagnosis of diabetes mellitus prior to admission, pressure ulcers,
measures of white blood cell count, hematocrit, and serum: sodium, glucose, blood
urea nitrogen, creatinine, albumin, and bilirubin.
Outcome variables were collected during the Barlow Respiratory Hospital
hospitalization. Variables included: death, septic shock death, length of stay (in
days), time to wean (in days), types of infections treated, number of treated
infections from the five most common and important infections - referred in this
study as prim ary infections (urinary tract infection, Clostridium difficile colitis,
pneumonia or tracheobronchitis, aspiration pneumonia, and line sepsis), treatment of
diabetes mellitus during the hospitalization, and type of isolation.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Statistical analyses: The Wilcoxon rank sum test was used to investigate the
differences in length of stay and time to wean between patients treated and not
treated, separately for each of the five primary infections. The Yates-corrected chi-
square test was used to investigate 1) the associations between death and each of the
primary infections, 2) the association between death and the number o f treated
infections (single vs. multiple), 3) the association between death and the type of
isolation precautions, and 4) the association between the status of diabetes mellitus
and the number of treated primary infections. Furthermore, univariate and
stepwise logistic regression analyses were conducted to identify risk factors for each
of the primary infections and sepsis (sepsis with shock and sepsis without shock).
Univariate and stepwise analyses were also used to investigate the associations
between the primary infections and sepsis. Risk factors significant at p < 0.20
value from the univariate logistic regressions were included in the subsequent
stepwise model selection method for each infection. The stepwise model was
adjusted for gender and age, and variables were added to the model if p-value < 0.05.
All analyses were conducted at the 0.05 significance level and utilized SAS
(Cary, NC: Version 8.02) and STATA (College Station, TX: Version 8.1).
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.3 RESULTS
Of the 222 patients treated during the study period, 186 (83.8%) were eligible and
included in the study. Table 2-la presents the baseline sociodemographic and
clinical variables of the patients at admission, and Table lb presents the outcome
variables during follow-up. As shown in Table 2-la, of the 186 patients, 98 were
males and 88 were females. The average age was 71.9 years. Diabetes mellitus
was diagnosed in 134 patients prior to admission, 36 patients had a prior episode of
mechanical ventilation, and 58 had multiple pressure ulcers.
Table 2-1a. Baseline Sociodemographic and Clinical Variables at Admission. (n=186)
G en d er: M ale I i 98 (4 7 .3 % ) i
A ge (y ears) 7 1 .9 ± 1 3 .7
P rior E p is o d e of M echanical V entilation
No 53 (28.5% )
Y es 3 6 (1 9 .4 % )
U nknow n 97 (52.1% )
P re m o rb id F unctional S ta tu s
^Totally b e d rid d e n a n d d is a b le d , no s e lf c a re 34 (18.3% )
s B ed rid d en 50% o r m o re of th e tim e, lim ited s e lf c a re 2 7 (14.5% )
A m bulatory, c a p a b le of se lf-c a re b u t n o t w ork 6 2 (33.4% )
R e stric te d in s tre n u o u s activity I 3 2 (1 7 .2 % ):
Fully active I 2 5 (1 3.4% )
iU nknow n 6 (3.2% ) I
D ia b e te s M ellitus P rior to A d m issio n 5 2 (2 8 .0 % ):
P r e s s u r e U lcers
N o n e I 7 7 (4 1 .4 % ):
S in g le 51 (27.4% )
t Multiple 58 (31.2% )
W hite B lood C ell (th o u sa n d /m l) 1 1.58 ± 6 .2 6 i
H em ato c rit (%) 3 0 .4 4 ± 4.11 I
S o d iu m (m Eq/L) I 137.11 ± 5.37
G lu c o se (m g/dL ) 1 4 1 .3 7 ± 57 8 7 '
B lood U rea N itrogen (m g/dL) 3 1 .3 7 ± 19 23
C re a tin in e (m g/dL) | 1.07 ± 0 78
A lbum in (gram /dL ) I 2.26 t 0 55
Bilirubin (m g/dL) I 0.84 + 1 4 9
* F re q u e n c y a n d % for d is c re te v a ria b le s, m e a n (SD ) for c o n tin u o u s v a ria b le s.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The majority of patients (n=62) were ambulatory, capable of self-care, but could not
work. The mean white blood cell count was 11.58 103 /ml and the mean bilirubin
was 0.84 mg/dL.
Table 2-1 b. Outcome Variables during Hospitalization (n =186).
Death 54 (29.0%)
Septic Shock Death 29(15.6% )
Length of Stay (days) 43.2 ± 30.8!
Num ber of Patients W eaned 95(5 1 1 % )
Time to W ean (days) 26.3 ± 24 4
i Number of Primary Infections Treated
0 57(30.6% )
1 1 44(23.7% )
2-5 85(45.7% )
Types of Infections Treated**
Primary Infections:
Urinary tract infection 85 (45.7%)
Clostridium difficile colitis 74(39.8% )
Pneum onia or tracheobronchitis 76(40.9% )!
Aspiration pneum onia 10 (5 4%)
Line sep sis 28 (15.1%)
Sepsis:
S epsis with shock 29(15.6% )
S epsis without shock 13 (7.0%)!
Treated for Diabetes Mellitus during Hospitalization 70(37.6% )
Type of Isolation
Methicillin-resistant staphylococcus aureu 62 (33 3%)
i Vancomycin-resistant enterococcus 9 (4 8%)
Frequency and % for discrete variables, m ean (SD)fbr continuous variables
Multiple infections possible
As shown in Table 2-lb, 54 patients died of whom 29 died of septic shock.
The mean length of stay was 43.2 days. Of the 186 patients, 95 (51.1%) were
weaned; 91 were not weaned (49 died, and 42 were discharged without weaning).
The mean time to wean was 26.3 days. O f the five primary infections, 44 patients
were treated for a single infection and 85 were treated for multiple infections while
57 were not treated for any infections.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The number of patients treated for diabetes during follow up was 70, and 62 were
isolated for methicillin-resistant staphylococcus aureus.
T ab le 2 -2 . L en g th o f S ta y a n d T im e to W e a n by t r e a t m e n t fo r P rim a ry in fectio n .
| 1 L ength o f S tay (d ay s) T im e to W ea n (d ays), n=95
f Infection IM ean (SD )!M edian (R a n g e)! p-value* iM ean (SD) M edian (R a n g e ) p-value*
U rinary tract infection i
No (n= 10 1) 32.1 (24.8) 2 7 .0 (3 ,1 5 0 ) < 0 0 1 No (n=54) 1 9 4 ( 1 5 .1 ) 14.0 (3 ,7 4 ) < 0 .0 1
I Y es (n = 8 5 )j5 6 .4 (32.2)1 4 9 .0 (1 1 ,1 6 7 ) Y es (n=41) 135.5 (30.7) 2 2 .0 (8 ,1 2 7 )
C lostridium difficile colitis t
No (n = 1 12) 3 3 .5 (26.4) 27.0 (3 ,1 6 7 ) < 0 .0 1 No (n=57) 1 9 9 ( 1 6 3 ) 15.0 (3 ,1 0 1 ) < 0 .0 1
i Y es (n= 74)l 5 7 .9 (31.4). 4 7 .0 (1 5 ,1 3 6 ) Y es (n=38) •36.0 (30.7) 2 3 .5 (7 ,1 2 7 )
P n e u m o n ia o r trac h eo b ro n c h itis
No (n = 1 10)131.8 (20.1) 2 8 .5 (3 ,1 0 9 ) j < 0 .0 1 No (n=68) 2 1 .9 (1 7 .8 ) 1 7 .0 (3 , 101) 0.03
Y es (n= 76)15 9.7 (35.9) 5 0 .0 (7 ,1 6 7 ) I Y es (n=27) 37.6 (33.9) 2 2 .0 (7 ,1 2 7 )
A spiration p n e u m o n ia
No (n = 1 7 6 ) 42.1 (30 1) 35 0 ( 3 ,1 6 7 ) j 0.05 No (n=89) 24 8 ( 2 2 5) 18.0 (3, 127) 0.03
Y es (n = 1 0 )|6 3 .1 (38.9) 5 3 .5 (1 4 ,1 3 1 ) i Y es (n=6) 50.0 (39.3) 3 4 .5 (1 9 ,1 1 9 )
Line s e p s i s i
( No (n = 1 5 8 ) 3 9.5 (28.1) 3 2 .5 (3 ,1 5 0 ) < 0 .0 1 No (n= 86) 23 8 (2 0 .5 ) 18.0 (3 ,1 1 9 ) 0 03
j Y es (n= 28) 64.1 (37.4) 5 6 .5 (7 ,1 6 7 ) ; Y es (n=9) i 50.7 (42.1) 3 8 .0 (9, 127)
I * C o m p a riso n of m e d ia n length o f sta y utilized W ilcoxon ran k su m test.
Table 2-2 presents the relationships between length of stay (from admission
to discharge) of patients and the primary infections, and between time to wean (from
admission to weaned date) and the primary infections, respectively. Those patients
treated for any one of the primary infections spent more days at Barlow Respiratory
Hospital and took more days to wean than those who were not treated for the
infection (all p ’s < 0.05).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2-3 presents the associations between death in relation to the a)
number of treated primary infections, b) type of primary infection, and c) type of
isolation. The likelihood of death if treated for multiple primary infections was
3.19 times greater than if treated for a single primary infection (p < 0.01). The
likelihood of death increased if treated for pneumonia or tracheobronchitis (OR =
2.61), and line sepsis (OR - 3.52), respectively (p < 0.01). The type of isolation
was not related to death (p > 0.49).
Table 2-3. The Association between Death in Relation
to the a) Number of Treated Infections, b) Type of
Primary Infection, and c) Type of Isolation. [ ______
Infection OR (95% C.l.) i D-value*
a) Number of Treated Infections
1 1 1.00
! 2-5 3.19 (1.27,8.00) 0.01
b) Type of Primary Infection
Urinary tract infection
No 1 00
Yes 1.15 (0.61,2.17) j 0.67
Clostridium difficile colitis
No 1.00 !
Yes 0.95 (0.50,1.81) 0.87
Pneumonia or tracheobronchitis s
No 1.00
i Yes 2.61 (1.36,4.98) <0.01
Aspiration pneumonia
No 1.00
Yes 1.05 (0.26,4.22)? >0.99
Line sep sis ;
No 1.00 j
Yes 3.52 (1.54,8.04) <0.01
c)Type of Isolation
Isolated for Methicillin-resistant staphylococcus aureus
No 1.00 s
i Yes 0.79 (0.40,1.56) 0.49
Isolated for Vancomycin-resistant enterococcus
No 1.00
Yes 1.24 (0.30,5 13) 0.72
!* Analyses utilized either the chi-square or Fisher's exact test
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2-4 presents the association between status of diabetes mellitus and
the multiple primary infections during the hospitalization. Patients with diabetes
mellitus prior to admission were not strongly related to the number of treated
primary infections (p = 0.09). In contrast, patients with diabetes mellitus prior to
admission and treated for diabetes mellitus during the hospitalization were 9.64
times more likely to be treated for two to five primary infections (p = 0.01). Also
patients treated for diabetes mellitus during the hospitalization were 3.84 times more
likely to be treated for two to five primary infections (p < 0.01).
Table 2.4. The Association between Diabetes Mellitus Status and Number of Treated Primary
[In fection s._______________ ] ________________________________________________ [ _______
Number of Primary Infections OR (95% C.l.) p-value*
Diabetes Priorto Admission (n=52/186)
0 1.00
1 1.25(0.49,3 20) 0 64
2-5 1.94(0.88,4.28) 0.09
Treatment during Hospitalization of Patients with Diabetes Mellitus Priorto Admission (n=44/52)
0 1.00...
1 7.14(0.53,95.38) 0.08
2-5 9.64(1.24,75.01) 0.01 [
0
Treatment during Hospitalization (n=70/186)
1.00
!
1
2-5
1.94(0.78,4.80)
3.84(1.72,8.59)
0.14
<0.01
* Analyses utilized chi-square.
Table 2-5 presents the results of the stepwise logistic regression analyses.
Shown in the table are independent risk factors for the primary infections after
adjusted for gender and age.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Patients who were treated for diabetes mellitus or had a white blood cell count
higher than 12.3 103 /ml were 2.14 times and 1.97 times more likely to be treated for
pneumonia or tracheobronchitis, respectively (p = 0.03 and p = 0.04). Higher
bilirubin level 0.5 mg/dL) was associated with aspiration pneumonia (p < 0.01).
Table 2-5. The Risk factors for Treated a) Primary and b) Sepsis/
Infection Risk Factors OR (95% C.l.) I p-value*
j a) Primary Infections
Urinary tract infection
None j 1
I Clostridium difficile colitis :
I None
i Pneum onia or tracheobronchitis
Diabetes
No 1.00 0.03
Yes 2.14(1.09,4.17)
White blood cell (thousand/ml)
! <12.3 1.00 0.04
| | >= 12.3 1.97(1.04,3.73) I
Aspiration pneum onia
Bilirubin (mg/dL)
< 0.5 1.00 <0.01
; >= 0.5 0.12(0.03,0.51) :
Line sep sis
Diabetes
I i No 1.00 0.06
j Yds 4.57 (0 .96,21.78)!
b) S epsis
S epsis with shock
Prior functional status: Totally bedridden/disabled/no self care
; No 1.00 0.04
Yes i 2.81 (1.04,7.56) i
Glucose (mg/dL)
r < 149 1.00 0.02
. >= 149 2.86 (1.17,6.98) j
Sep sis without shock
1 ............................None { _________________ j _____________ |
:* Analyses utilized logistic regression (p < 0.20), stepw ise model (p < 0.15), and !
adjusting for gender and age logistic regression (chi-square p < 0.05)
Those patients who were treated for diabetes mellitus prior to admission were 4.57
times more likely to be treated for line sepsis (p = 0.06).
1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Those patients with the lowest prior functional status, ‘Totally bedridden and
disabled, no self care’ were 2.81 times more likely to be treated for sepsis with
shock than those with a better prior functional status (p = 0.04). No risk factors
were found for urinary tract infection, Clostridium difficile colitis, and sepsis
without shock.
Table 2-6. The Association between the Primary infections
find Sepsis. i_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Risk Infection • OR (95% C.l.) p-value*
a) S epsis with Shock
i Pneum onia or tracheobronchitis
No i 1.00 <0.01
Yes 4.13(1.64,10.40)
b) S epsis without Shock
(Aspiration pneum onia
No 1.00 0 .0 2
Yes 7.64(1.33,43.78)
(Clostridium difficile colitis
i No 1.00 0 .0 1
( Yes i 9.09(1.65,50.06)
* Analyses utilized logistic regression (p < 0.20), stepw ise
i model (p < 0.15), and adjusting for gender and age
: logistic regression (chi-square p < 0.05) : ~
Table 2-6 presents the association between the primary infections and sepsis after
adjusting for gender and age. Those patients treated for pneumonia or
tracheobronchitis were 4.13 times more likely to be treated for sepsis with shock (p
< 0.01). Those patients treated for aspiration pneumonia were 7.64 times more
likely (p = 0.02), and those treated for Clostridium difficile colitis were 9.09 times
more likely (p = 0.01) to be treated for sepsis without shock.
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.4 DISCUSSION
Summary of Findings: O f the 186 ventilation-dependent patients admitted for
weaning, 52 patients were treated for diabetes mellitus prior to admission (Table 2-
la), and 70 patients were treated for diabetes mellitus during the hospitalization.
The most prevalent primary infection was urinary tract infection, and 129 (69.6%)
patients were treated for at least one of the five primary infection. Of the 54
patients who died, 29 (53.7%) died of septic shock (Table 2-lb). Length of stay
and time to wean in this cohort were known to be prolonged when compared to an
historical control5 because patients have been admitted older and more ill in recent
years. Median length of stay and time to wean from mechanical ventilation were
statistically significantly longer in patients sustaining any of the five primary
infections than in non-infected patients (Table 2-2). A prior study showed that the
mean time that a weaning protocol used at Barlow Respiratory Hospital was halted
by signs and symptoms of infection was 7.7 days6 which supports these findings.
As expected, pulmonary and more serious infections, and more than one infection
had a significant impact on mortality (Table 2-3).
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Treatment of diabetes mellitus during the hospitalization statistically significantly
increased the likelihood of treatment for multiple primary infections (Table 2-4).
Diabetes mellitus is known to impair immune function in the chronically critically
ill.7 In stepwise analyses, diabetes mellitus and high white blood cell were found
to be risk factors for pneumonia or tracheobronchitis, high bilirubin for aspiration
pneumonia, and diabetes mellitus for line sepsis (Table 2-5). The worst functional
status and high glucose were risk factors for sepsis with shock. In another set of
stepwise analyses, pneumonia or tracheobronchitis was found to be a risk factor for
sepsis with shock (Table 2-6). Aspiration pneumonia and Clostridium difficile
colitis were risk factors for sepsis without shock. Any similar studies in the past
have been on intensive care unit patients. The Multicenter study5 that consists of
23 sites including Barlow Respiratory Hospital found the increase in length of stay
and time to wean when treated for the infections in its 1,410 patient population.
Comparison with Other Studies: There is no direct comparison because this study
and the Multicenter study5 were the first studies of these associations in post
intensive care unit patients.
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Study Limitations: The results were limited to the variables studied since the data of
the patients were collected from retrospective chart review, and not designed
prospectively. The study cohort is a special cohort that is not representative of all
hospitals since the patients were from a specialized hospital. The sample from one
specialized hospital may not be representative of the specialization. A total of 186
patients may be considered a small sample size.
Risk factors associated with specific infections may be contributing factors
in the occurrence of the infection, or the result of it. It follows from this that 1) if
the risk factor is a contributor, 2) the mechanism resulting in the contribution is
known, and 3) that mechanism can be modified with demonstrable good effect, then
the risk factor is clinically significant, as well as statistically significant. Further
study is needed to investigate which of these risk factors for the specific infections
during the Barlow Respiratory Hospital hospitalization meet the three criteria. An
example of a risk factor that meets the three criteria might be diabetes mellitus,
although infection can unmask diabetes mellitus as well.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Diabetes mellitus may meet criteria numbers one and two, but may not meet number
« • R
three. If in the intensive care unit tighter glucose control decreases infection,
criterion number three might be met with tighter glucose control in prolonged
mechanical ventilation patients post-intensive care unit.
Summary and Conclusion: O f complications that typically befall ventilator-
dependent patients (infectious, cardiovascular, mechanical), common complications
treated in patients weaning from mechanical ventilation in the post-intensive care
unit setting are infectious. These patients have a host of conditions that may make
them particularly susceptible to infections including: age, multiple organ
dysfunction, intensive care unit exposure to broad spectrum antibiotics with
resultant antibiotic resistance,9 impaired mental status, incontinence, indwelling
lines (venous catheters, Foley catheters), aspiration, and tracheostomy. In this
cohort of 186 post-intensive care unit patients studied at Barlow Respiratory
Hospital; Primary infection, on admission or nosocomial, was treated in 69.4% of
patients. Multiple primary infections were treated in 45.7% of patients. The prior
diagnosis and concurrent treatment of diabetes mellitus was associated with sepsis
and treatment of multiple episodes of infection.
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Treatment for all five primary infections was associated with prolongation of length
of stay and time to wean. Treatment of lower respiratory infection was associated
with sepsis with shock while treatment of aspiration pneumonia and treatment of
Clostridium difficile colitis were associated with sepsis without shock. Death as an
outcome was associated with treatment for lower respiratory infection, line sepsis,
and sepsis with shock. Risk indicators associated with specific infections and
sepsis were found. Further study is needed to determine the clinical relevance of
the risk factors.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3. DATA MINING APPROACH
3.1 OVERVIEW OF DECISION TREE METHODOLOGY
One of the popular methods of data mining to explore large amounts of data
is a decision tree. A decision tree is a tree that recursively partitions data,
automatically splitting the data at optimum points.1 0 The root and each internal
node of a tree are labeled with a question. The arcs emanating from each node
represent each possible answer to the associated question. Each leaf node
represents a prediction of a solution to the problem under consideration.1 1 The
amount of information associated with an attribute value is related to the probability
of occurrence. T he basic strategy used by JMP (Cary, NC: Version 5.1) is to
choose splitting attributes with the highest information gain first. Gain is defined
as the difference between how much information is needed to make a correct
classification before the split versus how much information is needed after the split.
Certainly, the split should reduce the information needed by the largest amount.
This is calculated by determining the differences between the entropies of the
original dataset and the weighted sum of the entropies from each of the subdivided
datasets.
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Entropy is a concept used to quantify information. Entropy is used to measure the
amount of uncertainty or surprise or randomness in a set of data. When all data in
a set belong to a single class, there is no uncertainty. In this case the entropy is
zero.1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.2 APPLICATION TO BARLOW RESPIRATORY HOSPITAL DATASET
Figure 3-1 consists of three outputs of JMP using the same data on 186
patients. The picture a) presents the relationship between pneumonia or
tracheobronchitis and its risk factor, white blood cell count. To divide a continuous
variable, white blood cell count into high and low white blood cell counts, a split
point was necessary. In this example, when asked to perform a split, JMP
automatically split white blood cell count at 16.9 103 /ml as shown in the picture a).
•i
The most significant split such as 16.9 10 /ml is determined by the largest
likelihood-ratio chi-square statistic.1 0 In the logistic regression analyses reported in
section 2.3, one of the variable’s tertile, 12.3 103 /ml was used to divide the data into
two categories as high and low. Tertiles or quartiles such as 12.3 10 /ml are
arbitrary split numbers that create categories with equal numbers of subjects in each
category. A decision tree presents a more meaningful split point than a tertile or a
quartile.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3-1. JMP Decision Tree Output.
a) Relationship between Pneumonia or tracheobronchitis and White Blood Cell Count
All Rows
I
J
Count GA 2 Level Prob
186 251.60061 0 0.5914
1 0.4086
WBC>=16.9
I
WBC<16.9
!
Count GA 2 Level Prob Count GA 2 Level Prob
21 26.733595 0 0.3333 165 218.44317 0 0.6242
1 0.6667 1 0.3758
b) Relationship between Sepsis with Shock and Its First Three Risk Factors
All Row s
Count GA 2 Level Prob
186 161.01342 0 0.8441
1 0.1559
Levelod Glucose(O)
I
Count GA 2 Level Prob
124 83.22544 0 0.8952
1 0.1048
Leveled G lucose(1)
Count GA 2 Level Prob
62 70.806816 0 0.7419
1 0.2581
Leveled Zubrod(O)
Count GA 2 Level Prob
48 43.253876 0 0.8333
1 0.1667
Leveled Zubrod(1)
Count GA 2 Level Prob
14 19.121427 0 0.4286
1 0.5714
Leveled Albumln(O)
............ '
Leveled Albumin(1)
Count GA 2 Level Prob Count GA 2 Level Prob
25 29.647666 0 0.7200 23 8.226866 0 0.9565
1 0.2800 1 0.0435
c) Relationship between Sepsis with Shock and Glucose as a Continuous Variable
All Rows
)
Count GA 2 Level Prob
186 161.01342 0 0.8441
1 0.1559
Glucose>=175 Glucose<175 j
Count GA 2 Level Prob
41 52.644113 0 0.6585
1 0.3415
Count GA 2 Level Prob
145 96.452322 0 0.8966
1 0.1034
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
As shown in Table 3-1, the white blood cell count split at 16.9 103 /ml showed a
stronger likelihood in relation to pneumonia or tracheobronchitis than the white
blood count split at 12.3 103 /ml (OR = 3.63 vs. OR = 1.97).
Table 3-1. Comparison of Likelihood by Split Point
j Split Point by Risk Factor j OR (95% C.l.) i p-value* j
| a) Pneumonia or tracheobronchitis
(White Blood Cell (103 /mi)
r 123 1 97(1.04,3.73) 0 04
r 16.9 3.63(1.35,9.76) 0 01
j b) Sepsis with Shock
Glucose (mg/dL)
r 149 2.86(1.17,6.98) 0 02 ■
f 175 3.81 (1.54,9.40) [ < 0.01 |
|* Analyses utilized logistic regression (p < 0.20), j
stepwise model (p <0.15), and adjusting for gender ™ 1
;and age logistic regression (chi-square p < 0.05) j
The picture b) presents the splits of risk factors for sepsis with shock. To find risk
factors for sepsis with shock wit the traditional approach, a series of logistic and
stepwise regressions was necessary. However, the decision tree automatically
chooses risk factors and begins splitting. To determine the optimum split when
there are multiple variables, each inputted variable is considered. The one that
results in the highest reduction in total sum of squares is the optimum split, and is
used to create a new branch of the tree.1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the picture b), JMP chose glucose and prior functional status as first two risk
factors to be split just as the logistic and stepwise regressions have chosen them as
risk factors. Splitting can continue in the same manner until user is satisfied with
the predictive power of the model.1 1 After glucose and prior functional status, the
next variable chosen and split was albumin in the picture b). For any reason three
risk factors were needed, albumin would be the third risk factor for sepsis with
shock. Although the same first two risk factors were selected, the picture c)
presents a different split point for glucose at 175 mg/dL. As shown in Table 7,
again the glucose split at 175 mg/dL presented a stronger likelihood of sepsis with
shock than the glucose split at a tertile, 149 mg/dL (OR = 3.81 vs. OR = 2.86).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.3 DISCUSSION
Data mining approach using a basic decision tree was introduced to
reinforce the study. E xperts’ prior knowledge is necessary for choosing which
variables to include as potential risk factors before running a decision tree analysis
and when to stop splitting the variables any further. Experts may be able to
estimate a split point of variables based on their empirical experience when there are
only a few variables involved. However, when there are many variables, a decision
tree approach is helpful finding the significant split points for the many variables.
In the example of glucose, experts now knowing the significant glucose
level of 175 mg/dL may turn their attention directly to sepsis with shock when the
next patient comes in with his or her glucose level higher than 175 mg/dL or, by
clinical extrapolation, during the hospitalization. Knowing this significant level
may also lead to further studies into not only biological causality between glucose
level higher than 175 mg/dL and sepsis with shock but also other characteristics that
patients with glucose level higher than 175 mg/dL possess.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
BIBLIOGRAPHY
11. Dunham Margaret H. Data Mining: Introductory and Advanced Topics
Prentice Hall 2003.
1. Esteban A, Anzueto A, Frutos F, et al. Characteristics and outcomes in adult
patients receiving mechanical ventilation: a 28-day international study.
JAMA. 2002 Jan 16;287(3):345-55
6. Kalb TH, Lorin S. Infection in the chronically critically ill: unique risk
profile in a newly defined population. Crit Care Clin 2002; 18:529-552
2. Kurek CK, Cohen IL, Lambrinos J, et al: Clinical and economic outcome of
patients undergoing tracheostomy for prolonged mechanical ventilation in
New York State during 1993; analysis of 6,353 cases under diagnostic
related group 483. Crit Care Med; 1997;25:983-988
10. Sail J, Creighton L, Lehman A. JMP Start Statistics. Third Edition.
Thomson Learning. 2005.
9. Mylotte JM, Goodnough S, Tayara A. Antibiotic-resistant organisms among
long-term care facility residents on admission to an inpatient geriatrics unit:
Retrospective and prospective surveillance. Am J Infect Control 2001;
29(3): 139-44
4. Scheinhorn DJ, Chao DC, Steam-Hassenpflug MA, et al: Infectious
Complications in Weaning from Prolonged Mechanical Ventilation at Long
Term Hospitals: Preliminary Report from a Multicenter Study. Am J
Respir Crit Care Med 2004; 169(7):A44
5. Scheinhorn DJ, Chao DC, Steam-Hassenpflug MA, et al: Post-ICU
mechanical ventilation: Treatment of 1,123 patients. Am J Resp Crit Care
Med 1999; 159:1568-1573
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7. Scheinhorn DJ, Chao DC, Steam-Hassenpflug MA, et al. Outcomes in post-
ICU mechanical ventilation: A therapist-implemented weaning protocol.
Chest 2001; 119:236-242
3. Seneff MG, Zimmerman JE, Knaus WA, et al: Predicting the duration of
mechanical ventilation: The importance of disease and patients
characteristics. Chest 1996; 110:469
8. Van den Berghe G. How does blood glucose control with insulin save lives
in intensive care? J Clin Invest 2004; 114:1187-95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

P53 and bladder cancer outcome: A combined analysis from the Keck School of Medicine

PDF

Predictors of mammography use among California teachers

PDF

Metabolic effects of magnesium supplementation in women with a history of gestational diabetes mellitus

PDF

Validation of serum cotinine as a biomarker of environmental tobacco smoke exposure: Validation with self-report and association with subclinical atherosclerosis in non-smokers

PDF

Recreational physical activity and risk of breast cancer: The California Teachers Study

PDF

Predictive value of CT coronary artery calcium scanning for coronary heart disease in asymptomatic subjects with diabetes mellitus

PDF

The effects of tobacco smoke on respiratory symptoms among young adults in the Children's Health Study II

PDF

Cigarettes and alcohol in relation to colorectal cancer within the Singapore Chinese Health Study

PDF

Polymorphisms in genes involved in steroid hormone metabolism and mammographic density changes in women randomized to menopausal estrogen and progesterone therapy

PDF

The association between recreational physical activity and mammographic density

PDF

The influence of family structures on adolescent smoking among multicultural adolescents in Hawaii

PDF

Progression of carotid intima-media thickness and plasma antioxidants: The Los Angeles Atherosclerosis Study

PDF

A case-control study of passive smoking and bladder cancer risk in Los Angeles

PDF

Selective laser trabeculoplasty for the treatment of glaucoma

PDF

Randomized comparison of oral mercaptopurine versus oral thioguanine for standard risk acute lymphoblastic leukemia

PDF

Development and evaluation of standardized stroke outcome measures in a population of stroke patients in rural China

PDF

Effect of hormone therapy on the progression of carotid-artery atherosclerosis in postmenopausal women with and without established coronary artery disease

PDF

Risk factors for diabetic retinopathy in Latinos: Los Angeles Latino eye study

PDF

Agreement between administrative and clinical data for term newborns with congenital malformations

PDF

Lymphedema: Impact on breast cancer survivors

Asset Metadata

Creator Kim, Hanjoo (author)

Core Title Post-intensive care unit mechanical ventilation: Relationship of infections to outcomes of weaning from prolonged mechanical ventilation

Degree Master of Science

Degree Program Applied Biostatistics and Epidemiology

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

Tag biology, biostatistics,health sciences, public health,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-45919

Unique identifier UC11337875

Identifier 1430394.pdf (filename),usctheses-c16-45919 (legacy record id)

Legacy Identifier 1430394.pdf

Dmrecord 45919

Document Type Thesis

Rights Kim, Hanjoo

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA