University of Southern California Dissertations and Theses

Predicting neonatal outcomes among women diagnosed with severe preeclampsia and HELLP syndrome: a comparison of models

(USC Thesis Other)

Predicting neonatal outcomes among women diagnosed with severe preeclampsia and HELLP syndrome: a comparison of models

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

PREDICTING NEONATAL OUTCOMES AMONG WOMEN DIAGNOSED WITH SEVERE
PREECLAMPSIA AND HELLP SYNDROME:
A COMPARISON OF MODELS

by
Isabella Sarah Hauptman

A Thesis Presented to the
FACULTY OF THE USC KECK SCHOOL OF MEDICINE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
APPLIED BIOSTATISTICS AND EPIDEMIOLOGY

May 2021

Copyright 2021 Isabella Sarah Hauptman
ii
Acknowledgements
First and foremost, I wish to express my sincerest gratitude for my mentors and my thesis
committee for their support and guidance during the development of this thesis. Specifically, I would like
to thank Melissa Wilson, Ph.D. for her generosity, patience, and mentorship throughout this process. I
would also like to thank Patrick Mullin, M.D. and Wendy Mack, Ph.D. for serving on my committee and
providing extremely thoughtful feedback on my project. Further, I would like to thank the amazing
professors whom I have had the privilege of learning from at USC. I cannot overstate my appreciation for
their willingness to share their knowledge and excitement about their respective fields. To the Department
of Preventive Medicine at Keck School of Medicine of USC, thank you for the opportunity to fulfill my
lifelong dream of studying epidemiology and biostatistics.
Of course, the completion of this thesis and my degree would not have been possible without the love
and support of my friends and family. I most certainly would not be in the position I am today without
them. Mom, Dad, Natalie, and Brooke: thank you for everything.

iii
Table of Contents
Acknowledgements ........................................................................................................... ii
List of Tables .................................................................................................................... vi
List of Figures ................................................................................................................. vii
Abstract .......................................................................................................................... viii
Introduction ..................................................................................................................... 1
Methods ............................................................................................................................ 3
Results .............................................................................................................................. 8
Discussion ...................................................................................................................... 31
References ...................................................................................................................... 37
Appendices ..................................................................................................................... 40
Appendix A. ............................................................................................................................ 40
Figure A1. Residual Analysis: Delta c
2
for Model 1A ............................................................................... 40
Figure A2. Residual Analysis: Delta Deviance for Model 1A .................................................................... 40
Figure A3. Residual Analysis: Delta Beta for Model 1A ............................................................................ 41
Figure A4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 1B .......................... 41
Appendix B. ............................................................................................................................ 42
Figure B1. Residual Analysis: Delta c
2
for Model 1B ................................................................................ 42
Figure B2. Residual Analysis: Delta Deviance for Model 1B .................................................................... 42
Figure B3. Residual Analysis: Delta Beta for Model 1B ............................................................................ 43
iv
Figure B4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 1B .......................... 43
Appendix C. ............................................................................................................................ 44
Figure C1. Residual Analysis: Delta c
2
for Model 2A ................................................................................ 44
Figure C3. Residual Analysis: Delta Beta for Model 2A ............................................................................ 45
Figure C4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 2A .......................... 45
Appendix D. ............................................................................................................................ 46
Figure D1. Residual Analysis: Delta c
2
for Model 2B ................................................................................ 46
Figure D2. Residual Analysis: Delta Deviance for Model 2B .................................................................... 46
Figure D3. Residual Analysis: Delta Beta for Model 2B ............................................................................ 47
Figure D4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 2B .......................... 47
Appendix E. ............................................................................................................................ 48
Figure E1. Residual Analysis: Delta c
2
for Model 3A ................................................................................ 48
Figure E2. Residual Analysis: Delta Deviance for Model 3A .................................................................... 48
Figure E3. Residual Analysis: Delta Beta for Model 3A ............................................................................ 49
Figure E4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 3A .......................... 49
Appendix F. ............................................................................................................................ 50
Figure F1. Residual Analysis: Delta c
2
for Model 3B ................................................................................ 50
Figure F2. Residual Analysis: Delta Deviance for Model 3B ..................................................................... 50
Figure F3. Residual Analysis: Delta Beta for Model 3B ............................................................................. 51
Figure F4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 3B .......................... 51
Appendix G. ............................................................................................................................ 52
Figure G1. Residual Analysis: Delta c
2
for Model 4A ............................................................................... 52
Figure G2. Residual Analysis: Delta Deviance for Model 4A .................................................................... 52
Figure G3. Residual Analysis: Delta Beta for Model 4A ............................................................................ 53
Figure G4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 4A ......................... 53
v
Appendix H. ............................................................................................................................ 54
Figure H1. Residual Analysis: Delta c
2
for Model 4B ................................................................................ 54
Figure H2. Residual Analysis: Delta Deviance for Model 4B .................................................................... 54
Figure H3. Residual Analysis: Delta Beta for Model 4B ............................................................................ 55
Figure H4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 4B .......................... 55
Appendix I. ............................................................................................................................. 56
Figure I1. Residual Analysis: Delta c
2
for Model 5A ................................................................................. 56
Figure I2. Residual Analysis: Delta Deviance for Model 5A ...................................................................... 56
Figure I3. Residual Analysis: Delta Beta for Model 5A ............................................................................. 57
Figure I4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 5A ........................... 57
Appendix J. ............................................................................................................................. 58
Figure J1. Residual Analysis: Delta c
2
for Model 5B ................................................................................. 58
Figure J2. Residual Analysis: Delta Deviance for Model 5B ...................................................................... 58
Figure J3. Residual Analysis: Delta Beta for Model 5B ............................................................................. 59
Figure J4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 5B ........................... 59

vi
List of Tables
Table 1. Model Definitions ............................................................................................................. 4
Table 2. Demographic and Clinical Characteristics of the Study Population Categorized by
ACOG Diagnosis .......................................................................................................................... 11
Table 3. Demographic and Clinical Characteristics of the Study Population Categorized by
Gestational Age at Delivery .......................................................................................................... 13
Table 4. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria (Model
1A) ................................................................................................................................................ 15
Table 5. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 1B) .... 16
Table 6. Proportion of Low Birthweight, SGA, IUGR, and Apgar Score in the Composite
Neonatal Outcome ........................................................................................................................ 16
Table 7. Predictive Model for Neonatal Death using ACOG Diagnostic Criteria (Model 2A) ... 18
Table 8. Predictive Model for Neonatal Death using Gestational Age (Model 2B) ..................... 19
Table 9. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria (Model
3A) ................................................................................................................................................ 21
Table 10. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 3B) ... 22
Table 11. Proportion of Low Birthweight, SGA, IUGR, Apgar Score, and Neonatal Death in the
Composite Neonatal Outcome ...................................................................................................... 22
Table 12. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 4A) .................................................................................................................................... 24
Table 13. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 4B) ... 25
Table 14. Proportion of IUGR and Apgar Score in the Composite Neonatal Outcome ............... 25
Table 15. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 5A) .................................................................................................................................... 27
Table 16. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 5B) ... 28
Table 17. Proportion of IUGR, Apgar Score of 4 or less, and Neonatal Death in the Composite
Neonatal Outcome ........................................................................................................................ 28
Table 18. Model Sensitivity, Specificity, Correction Classification, AUC, and Model
Comparisons ................................................................................................................................. 30
vii
List of Figures
Figure 1. Area Under ROC Curve Model 1A ............................................................................... 15
Figure 2. Area Under ROC Curve Model 1B ............................................................................... 16
Figure 3. Area Under ROC Curve Model 2A ............................................................................... 18
Figure 4. Area Under ROC Curve Model 2B ............................................................................... 19
Figure 5. Area Under ROC Curve Model 3A ............................................................................... 21
Figure 6. Area Under ROC Curve Model 3B ............................................................................... 22
Figure 7. Area Under ROC Curve Model 4A ............................................................................... 24
Figure 8. Area Under ROC Curve Model 4B ............................................................................... 25
Figure 9. Area Under ROC Curve Model 5A ............................................................................... 27
Figure 10. Area Under ROC Curve Model 5B ............................................................................. 28

viii
Abstract
Background: Preeclampsia with severe features (sPE) and Hemolysis, Elevated Liver
Enzymes, and Low Platelet Count Syndrome (HELLP) are multisystemic gestational
complications that can have serious maternal and neonatal implications. Objective: We aim to
examine whether the clinical diagnosis of HELLP Syndrome or sPE is more predictive of
neonatal outcomes than gestational age alone. Methods: We developed a total of 10 predictive
models using clinical and laboratory data obtained from the medical records of 165 women with
self-identified HELLP syndrome recruited from 2 online HELLP resources. We defined 5
separate neonatal outcomes consisting of combinations of the following: small for gestational
age (SGA) defined by Olsen et al (2010), very low birthweight (<1500 g), intrauterine growth
restriction, neonatal death, and/or a 1 minute Apgar score of 4 or less. A predictive model was
developed for each neonatal outcome (5) for each main exposure of interest (The American
College of Obstetricians and Gynecologists (ACOG) diagnosis of HELLP versus sPE and
gestational age at delivery > 34 weeks versus ≤ 34 weeks) using multivariable logistic
regression. For each outcome, we compared the area under the curve between the ACOG
diagnostic model and the gestational age at delivery model. If the models were significantly
different (p<0.05), the model with higher AUC was identified as the preferred model. Results:
We found that gestational age was a better predictor than ACOG diagnosis of 1) SGA, very low
birthweight, IUGR, and/or an Apgar score of 4 or less and 2) SGA, very low birthweight, IUGR,
an Apgar score of 4 or less, and/or neonatal death (p=0.002 and p=0.031, respectively). There
was not a significant difference between the exposures for the following adverse neonatal events:
1) neonatal death (p=0.18), 2) IUGR and/or an Apgar of 4 or less (p=0.80), and 3) neonatal
death, IUGR and/or an Apgar of 4 or less (p=0.46). Conclusion: The results of our exploratory
ix
study support the use of GA as a predictor of adverse neonatal outcomes over the diagnosis of
HELLP Syndrome vs. sPE. Specifically, we observed that models developed with GA as a
predictor were equally as good as or better at predicting adverse neonatal outcomes compared to
models developed with diagnosis of HELLP and sPE.

1
Introduction
Preeclampsia with severe features (sPE) and Hemolysis, Elevated Liver Enzymes, and
Low Platelet Count Syndrome (HELLP) are serious gestational complications that can pose a
threat to both mother and child (Haram , Svendsen, & Abildgaard, 2009). Both are multisystemic
gestational disorders considered by most to be on the same clinical spectrum. Recent findings,
however, may suggest sPE and HELLP are independent conditions that arise from a materno-
fetal imbalance (Aloizos, et al., 2013). Preeclampsia occurs in 3-6% of pregnancies while
HELLP Syndrome occurs in 0.5 to 0.9% of pregnancies and in 20-25% of preeclamptic
pregnancies (Aloizos et al., 2013). sPE/HELLP are leading causes of maternal and neonatal
mortality in low- and middle-income countries (Myatt & Roberts, 2015). Most neonates born to a
mother with HELLP syndrome or sPE require extended hospitalization in neonatal intensive care
units (Aloizos et al., 2013).
The diagnostic criteria for preeclampsia were updated by The American College of
Obstetricians and Gynecologists in 2013 and The International Society for the Study of
Hypertension in Pregnancy (ISSHP) in 2014 to include hypertension in the absence of
proteinuria if hematologic complications, impairment of renal or liver function, neurological
symptoms, or uteroplacental dysfunction is present (Kallela, Jääskeläinen & Kortelainen et al.,
2016). Nevertheless, it remains unclear whether the ACOG-defined diagnoses are predictive of
neonatal outcomes or if other factors (i.e. gestational age) are equally or even more informative
(Abramovici, Friedman, Mercer, Audibert , Kao & Sibai, 1999; Harms, Rath, Herting & Kuhn,
1995). Gestational age (GA) is an established risk factor for poor neonatal and maternal
outcomes and the risk of adverse outcomes is negatively associated with increasing GA until up
to 40 weeks (Neggers, 2018; Platt, 2014).
2
In the present study, we aim to investigate the clinical utility of the ACOG diagnoses for
HELLP Syndrome and sPE in predicting neonatal outcomes. Specifically, we aim to examine
whether the clinical diagnosis of HELLP Syndrome or sPE is more predictive of neonatal
outcomes than GA alone.
3
Methods
To investigate the clinical utility of the ACOG diagnoses for HELLP syndrome and severe
PE in predicting neonatal outcomes, we developed a total of 10 predictive models to compare the
exposure of the ACOG diagnoses to exposure of GA at delivery using Stata 16.0 (StataCorp,
2019).
The study sample (n=165) consisted of women with self-identified HELLP Syndrome who
were recruited online from two separate HELLP resource websites
(www.hellpsyndromesociety.org or https://www.facebook.com/pages/Hellp-Syndrome-
Research-at-USC/163745723652843). Women completed a standardized risk factor
questionnaire, which included questions about their medical history, reproductive and sexual
history, family history, and the affected pregnancy. Medical records were requested from the
delivery hospital and the obstetrician from all cases. The diagnosis of HELLP was subsequently
verified by medical record abstraction and reviewed by one of the investigators for confirmation
of the diagnosis. A standardized data abstraction form was used to abstract the records, which
included information about prenatal visits, comorbidities, obstetric history, and delivery. Missing
covariate data from case abstractions were not imputed, with the exception of neonatal death.
1

Participants were classified into 2 main exposures based on their clinical characteristics. The
first exposure was ACOG diagnosis. Participants were classified as having HELLP syndrome if
medical records were available to confirm the following criteria: 1. Hemolysis (schistocytes, burr
cells, or LDH > 600), 2. Elevated liver enzymes (AST >70 and/or ALT > 70), and 3. Low
platelets (platelets < 100 K). Women meeting two of the three criteria were classified as having

1
Only missing data regarding neonatal death (n=11) were verified by the investigator and subsequently updated in
the analysis. No neonatal deaths occurred from these observations. Other missing variable data were not imputed.
4
sPE. Women with significant hypertension (³160/110 on two occasions, at least 6 hours apart)
and proteinuria (500 mg/dL/24 h or +3 dipstick on two occasions at least 6 hours apart) were also
classified as having sPE, with or without one of the above criteria. The second exposure was
early delivery, defined as delivery at a GA of less than or equal to 34 weeks.
We defined 5 separate neonatal outcomes (Table 1). The first was a composite outcome
consisting of at least one of the following outcomes: small for gestational age (birthweight <10
th

percentile per gestational week and gender defined as by Olson et al. (2010)), a 1 minute Apgar
score of less than or equal to 4, intrauterine growth restriction (IUGR), and very low birthweight
defined as less than 1500 g. The second outcome was neonatal death. The third outcome was a
composite of at least one of the first and second outcomes: small for gestational age (birthweight
<10
th
percentile per gestational week and gender defined as by Olson et al. (2010), a 1 minute
Apgar score of less than or equal to 4, intrauterine growth restriction (IUGR), very low
birthweight defined as less than 1500 g, and neonatal death. The fourth neonatal outcome was a
composite outcome consisting of at least a 1 minute Apgar score of less than or equal to 4 or
intrauterine growth restriction (IUGR). The fifth neonatal outcome was a composite outcome
consisting of at least a 1 minute Apgar score of less than or equal to 4, intrauterine growth
restriction (IUGR), or neonatal death.
Table 1. Model Definitions
Model Outcome (at least one of the following)
1 Small for gestational age, 1 min Apgar score ≤ 4, IUGR, very low birthweight
2 Neonatal death
3 Small for gestational age, 1 min Apgar score ≤ 4, IUGR, and very low
birthweight, and neonatal death
4 1 min Apgar score ≤ 4 and IUGR
5 1 min Apgar score ≤ 4, IUGR, and neonatal death
5
A total of 10 predictive models were developed. A predictive model was developed for each
neonatal outcome (5) for each main exposure of interest (ACOG diagnosis and Early Delivery).
The following independent candidate covariates were considered: maternal age at diagnosis of
pregnancy, pre-pregnancy weight, gestational age at first visit, maternal history of asthma,
maternal history of diabetes, maternal history of chronic hypertension, delivery type, headache
during labor, epigastric pain, edema, nausea, visual symptoms, maximum LDH levels, maximum
bilirubin levels, maximum AST levels, maximum ALT levels, maximum creatinine levels,
minimum platelet levels, child birthweight, maximum systolic blood pressure, maximum
diastolic blood pressure, white blood cell count, nulliparity, maternal hemorrhage, blood type,
eclampsia, and placental abruption. The distributions of selected characteristics are reported in
Tables 2 and 3. Continuous variables are reported as means ± standard deviation and categorical
variables are reported as frequencies and counts. Statistical tests comparing sPE to HELLP and
GA >34 weeks to GA ≤ 34 weeks were performed using t-tests for normally distributed
continuous variables, Wilcoxon rank sum tests for non-normally distributed continuous
variables, Pearson’s chi-square tests for categorical variables with expected frequencies of >5 in
all cells, and Fischer’s exact chi-square tests for categorical variables with expected frequencies
of <5 in at least one cell.
Exploratory data analysis was conducted for both continuous and categorical variables with
respect to both main exposures of interest. The distributions of continuous variables were plotted
by exposure status via histogram, boxplot, and a lowess smoothed graph on the logit scale. An
initial visual assessment of linearity was conducted, but no transformations occurred at this
stage. Categorical variables were tabulated by their respective frequencies by exposure status.
Univariate analyses were performed with logistic regression on each neonatal outcome and each
6
candidate variable. Eligible variables for the preliminary model were defined as having a
univariate p-value of <0.25. Once determined, we performed multivariable logistic regression on
each neonatal outcome and the respective eligible candidate variables. Subsequent variables that
did not meet statistical significance of less than or equal to 0.05 were removed from the
preliminary model in order of decreasing significance. Variables with a Wald p-value of less
than 0.05 were maintained in the model.
After the preliminary main effects models were finalized, the linearity of the continuous
variables were assessed. Fractional polynomials were calculated from the adjusted preliminary
models. Considerations were made for both one and two term power functions in comparison to
linear models for each continuous variable. The functions were selected based on significance of
<0.05. If significant, the greater power term was selected. Once linearity was assessed and
transformations were made, the preliminary models were updated and finalized. Although each
outcome had 2 separate models, each respective outcome model was matched to include the
same independent variables (save for the two separate main exposures of interest). Variables that
were forced into each model were noted. Goodness of fit was determined using the Hosmer-
Lemeshow test set to 4 groups. We chose this group selection in order to maximize the number
of models with expected values of 5 or greater in each group.
Model diagnostics of the final preliminary model were assessed by examining residuals (see
Apendices A-J). We identified covariate patterns and subsequent observations with potential
influence defined as a change in the regression estimates of greater than 1, a change in Pearson’s
Goodness of Fit of greater than 10, or a change in Deviance of greater than 10 when covariate
patterns were removed. Subsequent identified covariate patterns were assessed, but none were
removed from the models because each observation was clinically relevant. The resulting
7
regression estimates were reported on the odds ratio scale. The ROC curve, the area under the
ROC curve (AUC), and the classification table formed at the cutpoint deemed to maximize the
sensitivity and specificity of the model were reported as assessments of the predictive power of
the model. The maximized cutpoint was assessed by graphing sensitivity and specificity versus
probability cutoff to determine where both specificity and sensitivity were maximal. For each
outcome, we compared the area under the curve between each model to determine which model
was a better fit. If the models were statistically significantly different (p<0.05), the model with
higher AUC was identified as the preferred model. Subsequent findings and conclusions are
reported.
This study was conceived as an exploratory study and thus no a priori power calculation was
made to detect differences in AUC. Subsequently, we calculated power to detect the width of the
two-sided 95% confidence interval for the AUC of both the ACOG diagnostic criteria models
and the GA models using PASS Software (2020). Widths were calculated using 80% power and
AUC of 0.700. The models developed using the ACOG diagnostic criteria are powered to detect
a two-sided 95% confidence interval on AUC with a width of 0.166 when there were 68 subjects
from the positive population and 97 subjects from the negative population. The models
developed using GA are powered to detect a 95% confidence interval on AUC with a width of
0.157 when there were 93 subjects from the positive population and 72 subjects from the
negative population.
8
Results
The total study population consisted of 165 individuals, of which 68 were formally diagnosed
with HELLP Syndrome and 97 were diagnosed with PE with severe features (Table 2).
Additionally, 93 individuals gave birth at or below a GA of 34 weeks and 72 gave birth after 34
weeks (Table 3). The total sample size in each multivariable prediction model varies based on
the available data for each covariates.
When observing the clinical characteristics of the study population categorized by ACOG
Diagnosis, we do not observe a statistically significant difference between the demographic
features of mothers diagnosed with sPE and HELLP. The average maternal age of the study
population was around 30 years (p=0.21), with over 98% of participants white (p=0.81), and
over 87% nulliparous (p=0.70) (Table 2). No statistically significant difference was observed
between mothers diagnosed with sPE and HELLP with the following variables related to medical
history: prior history of hypertension (p=0.10), prior history of diabetes (p=0.93), mean pre-
pregnancy weight (lbs) (p=0.47), maximum systolic blood pressure (mmHg) p=(0.77), and
maximum diastolic blood pressure (mmHg) (p=0.90) (Table 2). Statistically significant
differences were observed between mothers diagnosed with sPE and HELLP with the following
variables related to laboratory measurements: higher maximum LDH (units/L) in the HELLP
group (2155.7 ± 3337.9) compared to the sPE group (509.3 ± 411.7) (p<0.001), higher
maximum bilirubin (mg/dL) in the HELLP group (4.1 ± 12.5) compared to the sPE group (0.7 ±
0.4) (p<0.001), higher maximum AST (units/L) in the HELLP group (625.8 ± 905.9) compared
to the sPE group (290.5 ± 394.4) (p<0.001), higher maximum ALT (units/L) in the HELLP
group (580.5 ± 1316.0) compared to the sPE group (262.5 ± 341.5) (p=0.001), and lower
minimum platelet count ( x 10
9
/ L) in the HELLP group (47.5 ± 20.2) compared to the sPE group
9
(101.9 ± 68.1) (p<0.001) (Table 2). Maximum creatinine levels (mg/dL) were not statistically
significantly different between the two groups (p=0.78). Additionally, no statistically significant
difference was detected between mothers diagnosed with sPE and HELLP with regard to the
following perinatal outcomes: delivery type (p=0.16), maternal hemorrhage (p=0.39),
birthweight (g) (p=0.56), small for gestational age (0.86), IUGR, Apgar Score (p=0.08), and
neonatal death (p=0.06) (Table 2).
When observing the clinical characteristics of the study population categorized by GA at
delivery, we do not observe a statistically significant difference between the demographic
features of mothers who gave birth at or below 34 weeks and those who gave birth later than 34
weeks. The average maternal age of the study population was around 30 years (p=0.31), with
over 98% of participants white (p=0.24), and over 87% nulliparous (p=0.59) (Table 3). No
statistically significant difference was observed between mothers who gave birth at or below 34
weeks and those who gave birth later than 34 weeks with the following variables related to
medical history: prior history of hypertension (p=0.60), prior history of diabetes (p=0.36), mean
pre-pregnancy weight (lbs) (p=0.90), and maximum diastolic blood pressure (mmHg) (p=0.39)
(Table 3). However, a statistically significant difference was noted for maximum systolic blood
pressure (mm Hg), with a higher SBP in the earlier delivery group (165.5 ± 21.7) compared to
the later delivery group (157.3 ± 25.5) (p=0.030) (Table 3). No statistically significant
differences were observed between mothers who gave birth at or below 34 weeks and those who
gave birth later than 34 weeks with the following variables related to laboratory measurements:
maximum LDH (units/L) (p=0.09), maximum bilirubin (mg/dL) (p=0.74), maximum AST
(units/L) (p=0.83), maximum ALT (units/L)(p=0.70), maximum creatinine levels (mg/dL)
(p=0.43), and minimum platelet count ( x 10
9
/ L) (p=0.39) (Table 3). Statistically significant
10
differences were detected for the following perinatal outcomes: delivery type, with less vaginal
and vacuum assisted and more Cesarean Section deliveries in the earlier delivery group
compared to the later delivery group (p=0.010), lower birthweight (g) in the earlier delivery
group (1327.1 ± 590.5) compared to the later delivery group (2769.5 ± 699.7) (p<0.001), greater
frequency of small for gestational age in the earlier delivery group (36.3%) compared to the later
delivery group (8.5%) (p<0.001), greater frequency of IUGR in the earlier delivery group
(21.1%) compared to the later delivery group (4.5%) (p=0.003), and greater frequency of
neonatal death in the earlier delivery group (17.2%) compared to the later delivery group (4.2%)
(p=0.009) (Table 3). No significant differences were observed for maternal hemorrhage (p=0.17)
or Apgar Score (p=0.11) (Table 3).
11
Table 2. Demographic and Clinical Characteristics of the Study Population Categorized by
ACOG Diagnosis
Variable
1
N Severe PE N HELLP p-value
2

Demographic Characteristics
Maternal age, years 97 30.8 (± 3.7) 68 30.1 (± 4.0) 0.21
White 90 89 (98.9%) 64 63 (98.4%) >0.99
Nulliparity 96 84 (87.5%) 66 59 (89.4%) 0.71
Medical History
History of hypertension 91 11 (12.1%) 67 3 (4.5%) 0.10
History of diabetes 94 6 (6.4%) 66 4 (6.1%) >0.99
Pre-pregnancy weight, lbs 90 147.3 (± 29.9) 57 148.1 (± 38.1) 0.47
Maximum systolic blood
pressure (mm Hg)
93 161.5 (± 24.3) 67 162.6 (± 23.0) 0.77
Maximum diastolic blood
Pressure (mm Hg)
93 98.7 (± 12.0) 67 99.2 (± 15.0) 0.90
Laboratory Measurements
Maximum LDH (units/L) 40 509.3 (± 411.7) 50 2155.7 (± 3337.9) <0.001
Maximum Bilirubin
(mg/dL)
56 0.7 (± 0.4) 56 4.1 (± 12.5) <0.001
Maximum AST (units/L) 87 290.5 (± 394.4) 67 625.8 (± 905.9) <0.001
Maximum ALT (units/L) 78 262.5 (± 341.5) 66 580.5 (± 1316.0) 0.001
Maximum Creatinine
(mg/dL)
74 3.6 (± 15.1) 59 5.4 (± 19.2) 0.78
Minimum Platelet Count
(x 10
9
/ L)
89 101.9 (± 68.1) 68 47.5 (± 20.2) <0.001

Lactate Dehydrogenase (LDH); Aspartate Aminotransferase (AST); Alanine Aminotransferase (ALT); Intrauterine
Growth Restriction (IUGR)
1
Continuous variables presented as mean ( ± standard deviation) and categorical variables presented as frequencies
(%)
2
P-values obtained by t-test or Wilcoxon rank sum for continuous variables as appropriate and by Pearson’s chi-
square test or exact test for categorical variables as appropriate

12

Perinatal Events
Delivery Type
Vaginal (spontaneous)
Cesarean Section
Vacuum-Assisted
87
20 (22.9%)
63 (72.4%)
4 (4.6%)
62
8 (12.9%)
53 (85.5%)
1 (1.6%)
0.17
Maternal Hemorrhage 90 4 (4.4%) 65 5 (7.7%) 0.49
Birthweight (g) 82 1989.9 (± 977.9) 61 1925.8 (± 945.4) 0.56
Small for Gestational Age 80 20 (25.0%) 59 14 (23.7%) 0.86
IUGR 92 12 (13.0%) 64 10 (15.6%) 0.65
Apgar Score
0-4
5-10
77
20 (20.6%)
77 (79.4%)
57
7 (10.3%)
61 (89.7%)
0.08
Neonatal Death 97 15 (15.5%) 68 4 (5.9%) 0.06
13
Table 3. Demographic and Clinical Characteristics of the Study Population Categorized by
Gestational Age at Delivery
Variable
1
N Gestational Age
> 34 weeks
N Gestational Age
≤ 34 weeks
p-value
2

Demographic Characteristics
Maternal Age, years 72 30.8 (± 3.7) 93 30.3 (± 4.0) 0.31
White 63 63 (100%) 91 89 (97.8%) 0.51
Nulliparity 69 62 (89.9%) 93 81 (87.1%) 0.59
Medical History
History of Hypertension 67 5 (7.5%) 91 9 (9.8%) 0.60
History of diabetes 70 3 (4.3%) 90 7 (7.8%) 0.52
Pre-pregnancy Weight,
lbs
63 146.6 (± 31.9) 84 148.4 (± 34.30) 0.90
Maximum Systolic Blood
Pressure mm Hg
70 157.3 (± 25.5) 90 165.5 (± 21.7) 0.030
Maximum Diastolic
Blood Pressure mm Hg
70 97.7 (± 14.0) 90 99.8 (± 2.8) 0.39
Laboratory Measurements
Maximum LDH (units/L) 35 867.5 (± 994.4) 55 1778.0 (±3224.4) 0.09
Maximum Bilirubin
(mg/dL)
52 1.8 (± 2.5) 60 2.4 (±12.0) 0.74
Maximum AST (units/L) 67 425.1 (± 620.8) 87 445.1 (± 734.0) 0.83
Maximum ALT (units/L) 65 354.4 (± 533.6) 79 452.6 (± 1169.0) 0.77
Maximum Creatinine
(mg/dL)
58 2.6 (± 9.7) 75 5.9 (± 20.9) 0.43
Minimum Platelet Count
(x 10
9
/ L)
70 84.2 (± 71.2) 87 73.6 (±47.7) 0.77

Lactate Dehydrogenase (LDH); Aspartate Aminotransferase (AST); Alanine Aminotransferase (ALT); Intrauterine
Growth Restriction (IUGR)
1
Continuous variables presented as mean ( ± standard deviation) and categorical variables presented as frequency
(%)
2
P-values obtained by t-test or Wilcoxon rank sum for continuous variables as appropriate and by Pearson’s chi-
square test or exact test for categorical variables as appropriate
14
Perinatal Events
Delivery Type
Vaginal (spontaneous)
Cesarean Section
Vacuum-Assisted
64
15 (23.4%)
44 (68.8%)
5 (7.8%)
85
13 (15.3%)
72 (84.7%)
0 (0%)
0.008
Maternal Hemorrhage 66 6 (9.1%) 89 3 (3.4%) 0.17
Birthweight (g) 63 2769.5 (± 699.7) 80 1327.1 (± 590.5) <0.001
Small for Gestational Age 59 5 (8.5%) 80 29 (36.3%) <0.001
IUGR 66 3 (4.5%) 90 19 (21.1%) 0.003
Apgar Score
0-4
5-10
72
8 (11.1 %)
64 (88.9 %)
93
19 (20.4%)
74 (79.6%)
0.11
Neonatal Death 72 3 (4.2%) 93 16 (17.2%) 0.009

15
Model 1. Composite Neonatal Outcome Including Low Birthweight, SGA, IUGR, and
Apgar Score

Table 4. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 1A)
Variable Odds Ratio 95% Confidence Interval p-value
Severe PE Ref. Ref. Ref. Ref.
HELLP 0.93 0.49 1.79 0.83
Maximum Systolic Blood
Pressure (mmHg)
1.02 1.01 1.04 0.002
n=160; LR c
2
(2)= 10.82; p=0.0045; Hosmer Lemeshow c
2
(2)= 0.14, p=0.9313
Figure 1. Area Under ROC Curve Model 1A

16
Table 5. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 1B)
n=160; LR c
2
(2)= 40.73; p<0.001; Hosmer Lemeshow c
2
(2)= c
2
=1.58, p=0.4536
Figure 2. Area Under ROC Curve Model 1B

Table 6. Proportion of Low Birthweight, SGA, IUGR, and Apgar Score in the Composite
Neonatal Outcome
2

Variable Cases (% total)
Small for Gestational
Age
36 (48.6%)
Intrauterine Growth
Restriction
22 (30.6%)
Apgar ≤ 4 27 (36.0%)
Low Birthweight 52 (70.7%)

2
Frequencies sum to more than 100% as some neonates had more than 1 outcome.
Variable Odds Ratio 95% Confidence Interval p-value
Gestational Age > 34 weeks Ref. Ref. Ref. Ref.
Gestational Age ≤ 34
weeks
6.91 3.31 14.44 <0.001
Maximum Systolic Blood
Pressure (mmHg)
1.02 1.01 1.04 0.012
17
The first set of models (1A and 1B) used a composite neonatal outcome including low
birthweight, SGA, IUGR, or 1 min APGAR Score of 4 or less. The proportions of low
birthweight, SGA, IUGR, and Apgar score are reported in Table 6. When using ACOG criteria
to define the outcome, we observed a nonsignificant reduction in adverse neonatal outcomes for
women diagnosed as HELLP compared to those with sPE (OR=0.93, 95% CI: 0.49, 1.79,
p=0.83, Table 4). We did note that maximum systolic blood pressure (SBP) is significantly
higher in women with the composite outcome compared to those without (OR=1.02, 95% CI:
1.01, 1.04, p=0.002, Table 4). The area under the ROC curve (AUC) for this model was
estimated to be 0.64 (95% CI: 0.55-0.73) (Table 18, Figure 1). As parameterized, the model had
a sensitivity of 49.3% and a specificity of 74.1% with a correct classification rate of 62.5%
(Table 18).
Using GA to predict risk of adverse neonatal outcomes, we found a significant increase in the
odds an adverse outcome associated with delivering at £34 weeks (OR=6.91, 95% CI: 3.31,
14.44, p<0.001, Table 5). As before, maximum SBP is associated with increased adverse
neonatal outcomes (OR=1.02, 95% CI: 1.01, 1.04, p=0.012 Table 5). When examining
predictive capacity of the model, we estimated the AUC to be 0.77 (95% CI: 0.69-0.84) (Table
18, Figure 2). As parameterized, the model had a sensitivity of 70.7% and a specificity of 68.2%
with a correct classification rate of 69.4%. A comparison of the area under the ROC curve
between Models 1A and 1B suggests that they are statistically significantly different, with the
GA-based model performing better than the one based on ACOG definitions (p=0.002) (Table
18).
18
Model 2. Neonatal Death Outcome

Table 7. Predictive Model for Neonatal Death using ACOG Diagnostic Criteria (Model 2A)
Variable Odds Ratio 95% Confidence Interval p-value
Severe PE Ref. Ref. Ref. Ref.
HELLP 0.47 0.11 1.98 0.30
Delivery Type
Vaginal (spontaneous)
Cesarean Section
Vacuum-Assisted

Ref.
0.60
0.44

Ref.
0.02
0.03

Ref.
0.24
5.54

Ref.
<0.001
0.53
Maximum Systolic Blood
Pressure
1.04 1.01 1.06 0.006
n=146; LR c
2
(4)= 26.76, p<0.001 ; Hosmer Lemeshow c
2
(2)=1.51, p=0.471
Figure 3. Area Under ROC Curve Model 2A

19
Table 8. Predictive Model for Neonatal Death using Gestational Age (Model 2B)
Variable Odds Ratio 95% Confidence Interval p-value
Gestational Age > 34
weeks
Ref. Ref. Ref. Ref.
Gestational Age ≤ 34
weeks
26.37 2.50 278.13 0.006
Delivery Type
Vaginal (spontaneous)
Cesarean Section
Vacuum-Assisted

Ref.
0.03
3.31

Ref.
0.01
0.14

Ref.
0.13
80.37

Ref.
<0.001
0.46
Maximum Systolic Blood
Pressure
1.03 1.01 1.06 0.040
n=146; LR c
2
(4)= 37.96, p<0.001 ; Hosmer Lemeshow c
2
(2)=0.19, p=0.908
Figure 4. Area Under ROC Curve Model 2B

The second set of models (2A and 2B) analyzed the outcome of neonatal death alone. When
using ACOG criteria to define the outcome, we observed a nonsignificant reduction in neonatal
death (OR=0.47 95% CI: 0.11-1.98, p=0.30, Table 7) associated with a diagnosis of HELLP
compared to sPE. Moreover, we found that neonatal death was significantly less likely for
Cesarean Section deliveries compared to vaginal deliveries (OR=0.60 95% CI: 0.02-0.24,
20
p<0.001 Table 7). Neonatal death for vacuum-assisted delivery was not significantly different
from vaginal delivery (OR=0.44 95% CI: 0.03-5.54, p=0.53 Table 7). We also found that
maximum systolic blood pressure (SBP) was significantly higher in those who had neonatal
death (OR=1.04 95% CI: 1.01-1.06, p=0.006 Table 7). The area under the ROC curve (AUC)
for this model was estimated to be 0.85 (95% CI: 0.70-0.99) (Table 18, Figure 3). As
parameterized, the model had a sensitivity of 71.4% and a specificity of 87.9% with a correct
classification rate of 86.3% (Table 18).
Using GA to predict risk of neonatal death, we found a significant increase in the odds an
adverse outcome associated with delivering at £34 weeks compared to >34 weeks (OR=26.37,
95% CI: 2.50- 278.13, p=0.006, Table 8). As before, neonatal death was significantly less likely
in Cesarean Section delivery compared to vaginal delivery (OR=0.03 95% CI: 0.01-0.13,
p<0.001 Table 8), but neonatal death was not significantly different for vacuum-assisted delivery
compared to vaginal delivery (OR=3.31 95% CI: 0.14-80.37, p=0.46 Table 7). Additionally,
maximum SBP was independently associated with neonatal death (OR=1.03, 95% CI: 1.01, 1.06,
p=0.040 Table 8). When examining predictive capacity of the model, the AUC was 0.89 (95%
CI: 0.78-0.99) (Table 18, Figure 4). As parameterized, the model had a sensitivity of 64.3% and
a specificity of 93.2% with a correct classification rate of 90.4% (Table 178. A comparison of
the area under the ROC curve between Models 2A and 2B suggests that they are not statistically
significantly different (p=0.18) (Table 18).
21
Model 3. Composite Neonatal Outcome Including Low Birthweight, SGA, IUGR, Apgar
Score, and Neonatal Death

Table 9. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 3A)
Variable Odds Ratio 95% Confidence Interval p-value
Severe PE Ref. Ref. Ref. Ref.
HELLP 0.52 0.25 1.08 0.08
Maximum
Systolic Blood
Pressure
1.03 1.01 1.04 0.001
Edema 2.58 1.25 5.33 0.010
Male 2.09 1.01 4.35 0.048
n=147; LR c
2
(4)= 23.03; p=0.0001; Hosmer Lemeshow c
2
(2) =0.83, p=0.66
Figure 5. Area Under ROC Curve Model 3A
22
Table 10. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 3B)
Variable Odds Ratio 95% Confidence Interval p-value
Gestational Age > 34
weeks
Ref. Ref. Ref. Ref.
Gestational Age ≤ 34
weeks
8.35 3.70 18.86 <0.001
Maximum Systolic
Blood Pressure
1.03 1.01 1.04 0.008
Edema 2.53 1.14 5.65 0.023
Male 1.47 0.66 3.26 0.35
n=147 LR c
2
(4)= 50.07 ; p<0.001 Hosmer Lemeshow c
2
(2)=0.83, p=0.66
Figure 6. Area Under ROC Curve Model 3B

Table 11. Proportion of Low Birthweight, SGA, IUGR, Apgar Score, and Neonatal Death
in the Composite Neonatal Outcome
3

Variable Cases (%)
Small for Gestational Age 34 (48.6)
Intrauterine Growth Restriction 22 (28.6%)
Apgar ≤4 27 (33.8%)
Low Birthweight 53 (66.4%)
Neonatal Death 19 (25.0%)

3
Frequencies sum to more than 100% as some neonates had more than 1 outcome.
23
The third set of models (3A and 3B) analyzed the neonatal outcome including very low
birthweight, SGA, IUGR, Apgar Score, and neonatal death. The proportions of low birthweight,
SGA, IUGR, Apgar score, and neonatal death are reported in Table 11. When using ACOG
criteria to predict risk, we observed a nonsignificant reduction in neonatal death (OR=0.52 95%
CI: 0.25-1.08, p=0.08, Table 9) for those with HELLP compared to those with sPE. Once again,
maximum SBP was significantly higher in those with the adverse neonatal outcomes (O R=1.03
95% CI: 1.01-1.04, p=0.001 Table 9). Additionally, those with edema were significantly more
likely to have the outcome (OR=2.58 95% CI: 1.25-5.33, p=0.010 Table 9) along with male
neonates (OR=2.09 95% CI:1.01-4.53, p=0.048 Table 9). The area under the ROC curve (AUC)
for this model was estimated to be 0.73 (95% CI: 0.65-0.81) (Table 18. Figure 5). As
parameterized, the model had a sensitivity of 70.3% and a specificity of 64.4% with a correct
classification rate of 67.4% (Table 18).
Using GA to predict risk of neonatal outcomes including low birthweight, SGA, IUGR,
Apgar Score, and neonatal death, we found a significant increase in the odds of an adverse
outcome associated with delivering at £34 weeks compared to those who delivered >34 weeks
(OR=8.53, 95% CI: 3.70- 18.86, p<0.001, Table 10). As before, we noted that maximum SBP is
significantly higher in those with the adverse neonatal outcomes, independent of GA (OR=1.03
95% CI: 1.01-1.04, p=0.008 Table 10). Additionally, those with edema were significantly more
likely to have a neonatal complication, independent of GA, gender, or blood pressure. (OR=2.53
95% CI: 1.14-5.65, p=0.023 Table 10). However, male neonates were not significantly more
likely to have an adverse outcome when using GA as the exposure (OR=1.47 95% CI:0.66-3.26,
p=0.35 Table 10). When examining predictive capacity of the model, we estimated the AUC to
be 0.82 (95% CI: 0.75-0.89) (Table 18, Figure 6). As parameterized, the model had a sensitivity
24
of 75.7% and a specificity of 76.7% with a correct classification rate of 76.2% (Table 18). A
comparison of the area under the ROC curve between Models 3A and 3B suggests that they are
statistically significantly different (p=0.031), with the GA-based model showing better predictive
ability than the model based on ACOG definitions (Table 18).
Model 4. Composite Neonatal Outcome Including IUGR and Apgar Score

Table 12. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 4A)
Variable Odds Ratio 95% Confidence Interval p-value
Severe PE Ref. Ref. Ref. Ref.
HELLP 0.30 0.11 0.79 0.014
Maximum
Creatinine
1.03 1.00 1.06 0.032
Maximum
Systolic Blood
Pressure
1.04 1.02 1.06 0.001
n=130; LR c
2
(3)=22.96, p<0.001; Hosmer Lemeshow c
2
(2)= 1.02, p=0.6019

Figure 7. Area Under ROC Curve Model 4A

25
Table 13. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 4B)
Variable Odds Ratio 95% Confidence Interval p-value
Gestational Age >
34 weeks
Ref. Ref. Ref. Ref.
Gestational Age ≤
34 weeks
2.68 1.04 6.92 0.041
Maximum
Creatinine
1.04 1.00 1.05 0.06
Maximum Systolic
Blood Pressure
1.03 1.02 1.06 0.002
n=130; LR c
2
(3)=20.6 p<0.001; Hosmer Lemeshow c
2
(2) =1.34, p=0.5130

Figure 8. Area Under ROC Curve Model 4B

Table 14. Proportion of IUGR and Apgar Score in the Composite Neonatal Outcome
4

Variable Cases (%)
Intrauterine Growth Restriction 22 (52.4%)
Apgar ≤ 4 27 (62.8%)

4
Frequencies sum to more than 100% as some neonates had more than 1 outcome.
26
The fourth set of models (4A and 4B) analyzed the neonatal outcome including IUGR and
Apgar Score. The proportions of IUGR and Apgar score are reported in Table 14. When using
ACOG criteria to predict risk, we observed a significant reduction in the odds of adverse
neonatal events with the diagnosis of HELLP compared to sPE (OR=0.30 95% CI: 0.11-0.79,
p=0.014, Table 12). We noted that both maximum creatinine levels (OR=1.03 95% CI: 1.00-
1.06, p=0.032, Table 12) and maximum SBP (OR=1.04 95% CI: 1.02-1.06, p=0.001, Table 12)
were significantly and independently associated with higher risk of neonatal complications. The
area under the ROC curve (AUC) for this model was estimated to be 0.77 (95% CI: 0.67-0.87)
(Table 18, Figure 7). As parameterized, the model had a sensitivity of 72.7% and a specificity of
67.0% with a correct classification rate of 68.5% (Table 18).
Using GA to predict risk of adverse outcomes, including IUGR and Apgar Score of 4 or less,
we found a significant increase in the odds having a neonatal complication associated with
delivering at £34 weeks (OR=2.68, 95% CI: 1.04- 6.92, p=0.041, Table 13). As before, we noted
a significant and independent increase in the odds of an adverse outcome associated with
maximum SBP (OR=1.03 95% CI: 1.02-1.06, p=0.002, Table 13). In contrast to the ACOG
model, there is not a significant increase in the odds of an adverse neonatal outcome associated
with maximum creatinine levels (OR=1.04 95% CI: 1.00-1.05, p=0.06 Table 13). When
examining predictive capacity of the model, we estimated the AUC to be 0.76 (95% CI: 0.67-
0.86) (Table 18, Figure 8). As parameterized, the model had a sensitivity of 78.8% and a
specificity of 65.0% with a correct classification rate of 68.5% (Table 18). A comparison of the
area under the ROC curve between Models 4A and 4B suggests that they are not statistically
significantly different (p=0.81) (Table 18).
27
Model 5. Composite Neonatal Outcome Including IUGR, Apgar Score, and Neonatal
Death

Table 15. Predictive Model for Adverse Neonatal Events using ACOG Diagnostic Criteria
(Model 5A)
Variable Odds Ratio 95% Confidence Interval p-value
Severe PE Ref. Ref. Ref. Ref.
HELLP 0.23 0.09 0.60 0.003
Maximum
Creatinine
1.03 1.00 1.06 0.041
Maximum
Systolic Blood
Pressure
1.04 1.01 1.06 0.001
n=130; LR c
2
(3)=27.10, p<0.001 ; Hosmer Lemeshow c
2
(2)=1.59, p=0.4525
Figure 9. Area Under ROC Curve Model 5A

28
Table 16. Predictive Model for Adverse Neonatal Events using Gestational Age (Model 5B)
Variable Odds Ratio 95% Confidence Interval p-value
Gestational Age
> 34 weeks
Ref. Ref. Ref. Ref.
Gestational Age
≤ 34 weeks
2.81 1.14 6.94 0.024
Maximum
Creatinine
1.02 1.00 1.05 0.08
Maximum
Systolic Blood
Pressure
1.03 1.01 1.05 0.004
n=130; LR c
2
(3)=22.08 p=0.0001; Hosmer Lemeshow c
2
(2)=2.57, p=0.2771.
Figure 10. Area Under ROC Curve Model 5B

Table 17. Proportion of IUGR, Apgar Score of 4 or less, and Neonatal Death in the
Composite Neonatal Outcome
5

Variable Cases (%)
Intrauterine Growth
Restriction
22 (44.9%)
APGAR ≤4 27 (54.0%)
Death 19 (39.6%)

5
Frequencies sum to more than 100% as some neonates had more than 1 outcome.
29
The fifth set of models (5A and 5B) analyzed the neonatal outcome including IUGR,
Apgar score, and neonatal death. The proportions of IUGR, Apgar score, and neonatal are
reported in Table 17. When using ACOG criteria to predict risk, we observed a significant
reduction in the odds of an adverse neonatal outcome for women who met all criteria for HELLP
compared to those with sPE (OR=0.23 95% CI: 0.09-0.69, p=0.003, Table 15). Again, maximum
SBP is independently associated with increased risk (OR=1.04 95% CI: 1.01-1.06, p=0.001,
Table 15). Maximum creatinine levels were also associated with higher risk (OR=1.03 95% CI:
1.00-1.06, p=0.041, Table 15). The area under the ROC curve (AUC) for this model was
estimated to be 0.78 (95% CI: 0.66-0.84) (Table 18, Figure 9). As parameterized, the model had
a sensitivity of 73.0% and a specificity of 68.8% with a correct classification rate of 70.0%
(Table 18).
Using GA to predict risk of adverse neonatal outcomes including IUGR, Apgar Score of 4 or
less, and neonatal death, we found a significant increase in the odds an adverse outcome
associated with delivering at £34 weeks compared to those who delivered after 34 weeks
(OR=2.81, 95% CI: 1.14-6.49, p=0.023, Table 16). Consistent with other models, we noted a
significant increase in the odds of an adverse outcome associated with maximum SBP (OR=1.03
95% CI: 1.01-1.05, p=0.004, Table 16). Maximum creatine levels were independently associated
with an increase in the odds of an adverse neonatal outcome, however the OR did not reach
statistical significance (OR=1.02 95% CI: 1.00-1.05, p=0.08 Table 16). When examining
predictive capacity of the model, we estimated the AUC to be 0.75 (95% CI: 0.69-0.87) (Table
18, Figure 10). As parameterized, the model had a sensitivity of 72.9% and a specificity of
66.7% with a correct classification rate of 68.5% (Table 18).A comparison of the area under the
30
ROC curve between Models 5A and 5B suggests that they are not statistically significantly
different (p=0.46) (Table 18).
Table 18. Model Sensitivity, Specificity, Correction Classification, AUC, and Model
Comparisons

Model Cut
point
Sensitivity
(%)
Specificity
(%)
Correct
Classification
(%)
AUC 95%
Confidence
Interval
Comparison
p-value
Model
1A
0.50 49.3 74.1 62.5 0.64 0.55-0.73 0.002
Model
1B
0.55 70.7 69.4 68.2 0.77 0.69-0.84
Model
2A
0.15 71.4 87.9 86.3 0.85 0.70-0.99 0.18
Model
2B
0.15 64.3 93.2 90.4 0.89 0.78-0.99
Model
3A
0.50 70.3 64.4 67.4 0.73 0.65-0.81 0.031
Model
3B
0.55 75.7 76.7 76.2 0.82 0.75-0.89
Model
4A
0.25 72.7 67.0 68.5 0.77 0.67-0.87 0.80
Model
4B
0.25 78.8 65.0 68.5 0.76 0.67-0.86
Model
5A
0.3 73.0 68.8 70.0 0.78 0.69-0.87 0.46
Model
5B
0.3 72.9 66.7 68.5 0.75 0.66-0.84

31
Discussion
In this paper, we developed a total of 10 predictive models to investigate the clinical
utility of the ACOG diagnosis for sPE and HELLP Syndrome in predicting adverse neonatal
outcomes. In general, the GA-based exposure models performed better than the ACOG-defined
exposure models for several of the outcome definitions used (Models 1A/B and 3A/B).
Additionally, we found that HELLP Syndrome, compared to those with sPE, is associated with a
decreased risk of an adverse outcome, as defined in Models 4 and 5. Delivering at or below 34
weeks of gestation was a significant predictor of neonatal outcomes regardless of the definition
used. Additionally, maximum systolic blood pressure was associated with a significant increase
in risk of an adverse outcome independent of outcome definition. Of note, when evaluating the
risk of neonatal death, delivery by C-Section was associated with a decreased risk (Models
2A/B).
These findings are not commensurate with prior studies that evaluated neonatal outcomes
in women with sPE and HELLP Syndrome. Few studies attempted to model the predictive power
of the ACOG diagnoses and instead reported on associations with various adverse neonatal
outcomes between those with HELLP and sPE. Examining such associations, Gul et al. found
that neonatal and perinatal mortality was significantly higher in the HELLP group versus the sPE
group but did not find any significant difference between the groups with respect to IUGR and
Apgar score. Once controlling for GA (GA) at delivery, these differences were insignificant (Gul
et al., 2005). Similarly, Turget et al. (2010) found that neonates born to women with HELLP
Syndrome had significantly lower neonatal bodyweight and higher neonatal mortality compared
to women with sPE without HELLP. As with Gul et al., neonatal mortality and morbidity were
found to be more influenced by GA than diagnosis (Turget et al., 2010); when stratified by GA,
32
the association between neonatal adverse outcomes and diagnosis are attenuated and
nonsignificant. Abramovici et al. (1999) also found that neonates born to women with HELLP
had significantly lower birthweight, earlier GA at delivery, and a higher frequency of 5 minutes
Apgar scores less than 7 compared to neonates born to women with sPE, but the association
becomes null when stratified as <28 weeks, 29-32 weeks, and 33-36 weeks of GA (Abramovici
et al., 1999). Haddad et al., (2000) found no association between an increased risk in neonatal
adverse outcomes among women with HELLP syndrome diagnosed at or before 28 weeks of
gestation compared to women with severe PE diagnosed at or before 28 weeks.
Several differences between these studies may account for the discrepancy in findings.
Our study was comprised of an almost entirely white study population with ready internet access.
Unlike preeclampsia, which is more prevalent in black women, HELLP syndrome is more
common in white women (Vinnars, Wijnaendts, Westgren, et al., 2008). Both Abramovici et al.
(2009) and Haddad et al. (2000) examined HELLP and severe PE in a study population that
included a majority African American population. Further, the respective diagnoses of HELLP
and sPE used in this study were based upon the 2013 ACOG criteria. Gul et al (2005), Turgot et
al. (2010), Haddad et al. (2000), and Abramovici et al. (1999) defined sPE using the 1996 ACOG
criteria, which excluded severe gestational hypertension in the absence of proteinuria with other
clinical features. Historically, studies of PE regularly adjusted for GA. However, GA should not
be included in models examining risk factors for neonatal outcomes as GA can be an
intermediate or a collider, not a confounder (Wilcox, Weinberg, & Basso, 2011). In studies
where the exposure is PE and the outcome is a neonatal complication, GA is an intermediate
between PE and the complication, since early delivery is a result of developing PE. Adjusting for
an intermediate variable results in bias toward the null. Similarly, collider-stratification bias can
33
result when conditioning on a shared effect, such as GA, which effects both neonatal outcomes
and PE. Adjusting for a collider can lead to substantial negative bias (Schisterman, Cole, & Platt,
2009). Thus, we would expect to see bias toward the null when stratifying by GA, which was
observed by Abramovici et al (1999), Gul et al (2005), and Turgot et al (2010).
In the present study, we found that the models using ACOG diagnoses were generally
worse at predicting neonatal outcomes compared to models using GA (≤34 weeks vs. >34
weeks). Specifically, our GA models were significantly better at predicting the neonatal adverse
events of 1) low birthweight, SGA, IUGR, and 1 minute Apgar Score of 4 or less (Model 1) and
2) low birthweight, SGA, IUGR, 1 minute Apgar Score of 4 or less, and neonatal death (Model
3). We did not find a significant difference between the predictive power of the models that
examined 1) neonatal death (Model 2), 2) IUGR and 1 minute Apgar score of 4 or less (Model
4), and 3) IUGR, 1 minute Apgar score of 4 or less, and neonatal death (Model 5). Though the
ACOG-based models 2, 4 and 5 were not significantly different than the GA-based models 2, 4
and 5 with regard to predictive power, we found that the ACOG diagnosis was not significantly
associated with risk of adverse neonatal outcomes in contrast to GA, dichotomized at 34 weeks,
in these models. These findings suggest that neonatal morbidity and mortality are better
predicted by GA at delivery than to the presence of PE with severe features or HELLP
syndrome.
These results are supported by previous findings. Kinay et al. (2015) examined maternal
characteristics and perinatal outcomes between women with severe PE and HELLP in two
separate groups: women who gave birth at or less than 34 weeks’ gestation and more than 34
weeks’ gestation. They did not find a statistically significant difference in perinatal outcomes
between patients with severe preeclampsia and HELLP in either GA category, suggesting that
34
ACOG diagnosis may be a poor predictor of neonatal outcomes (Kinay et al., 2015). A study by
Menzies et al. (2007) examined the predictive power of preeclampsia severity using the
definitions provided by the Canadian Hypertension Society (CHS) and the National
Hypertension Blood Pressure. Education Program Working Group on Hypertension in Pregnancy
(NHBPEP) in the U.S in an international cohort. The study found little evidence that sPE
predicted adverse neonatal outcomes, with the exception of diastolic blood pressure greater than
110 mm Hg and suspected placental abruption (Menzies et al., 2007).
Our findings support the conclusion that the ACOG diagnosis of preeclampsia and PE
with severe features does not predict adverse neonatal outcomes better than GA alone, and in
fact, GA appears to be a better predictor in some outcomes. Although HELLP Syndrome and
severe PE have defined diagnostic criteria, the clinical utility of the diagnoses for predicting
adverse events are in question. The rigid nature of the definitions, the dynamic nature of
delivery, and interventions employed to manage symptoms can all impact the ultimate diagnosis.
To address some of these concerns, in 2013 ACOG revised the diagnosis of preeclampsia to
include clinical signs of organ dysfunction (e.g., renal or liver function) when proteinuria is not
present (American College of Obstetricians and Gynecologists, 2013). Most women who are
diagnosed with HELLP Syndrome also have either preeclampsia or eclampsia, though non-
hypertensive HELLP Syndrome has been described (Rath, Faridi, & Dudenhausen, 2000).
An evaluation for suspected HELLP diagnosis requires complete chemistry and blood
panels, including nitrogen, urea, creatinine, liver function enzymes, LDH, glucose, total
bilirubin, and uric acid, an assessment of red blood cells, and, in some cases, liver imaging. Yet
diagnosis of HELLP is often made even when all of the tests required to meet the definition have
not been done. Because the diagnosis of PE and HELLP Syndrome often occur simultaneously
35
and because all laboratory assays to definitively diagnose HELLP are often not performed, there
is potential for misclassification. Exactly how much overlap there is between PE with severe
features and HELLP Syndrome is an area of active research, with some studies suggesting
substantial overlap (Smulian, Shen-Schwarz, Scorza, Kinzler & Vintzileos, 2004; Weiner et al.,
2016) and others suggestive of differing underlying pathophysiology (Mehrabian,
Mohammadizadeh, Moghtaderi, & Najafian, 2012; Vinnars et al., 2008). Evidence suggests that
both preeclampsia and HELLP have origins in the placenta and both maternal and fetal genes
have a role in the etiology (Haram, Mortensen & Nagy, 2014; Wilson et al., 2009; Wilson et al,
2011).
Our findings suggest that the use of GA to predict neonatal outcomes in this population
has more predictive ability than the use of ACOG diagnoses. In contrast to the myriad
difficulties in accurately diagnosing a dynamic condition, GA has less potential for
misclassification. It’s possible that the increase in predictive ability for the GA-based models is a
result of misclassification when using the ACOG criteria.
This study has several strengths. First, a strength of the study is the variety of models
developed. We defined, a priori, different definitions of adverse neonatal outcome to thoroughly
investigate the predictive ability of ACOG diagnoses compared to GA. All outcomes consisted
of composite adverse neonatal events, with the exception of model 2, where neonatal death was
considered in isolation. We felt that since the relationship with sPE/HELLP and various
neonatal events was not understood from an etiologic perspective, using varying definitions of
neonatal adverse outcomes was an important first step. Second, medical laboratory data were
available through medical record abstraction. This allowed us to verify the diagnoses as well as
evaluate specific laboratory values as covariates.
36
This study also has several limitations. First, the study population is small (n=165),
which leads to reduced power to detect differences between the AUC curves, especially since
some models had fewer observations due to missing data on covariates. Second, participants
were self-identified and opted into the study from online resources. Therefore, it is unknown how
many women with HELLP Syndrome or severe PE accessed the websites and thus, we are
unable to calculate participation rates or evaluate selection bias. Third, the potential for
misclassification of HELLP syndrome vs. sPE is not insignificant, since diagnosis of HELLP
requires complete blood and chemistry panels and timing of the assays can determine whether a
diagnosis of HELLP is made. If these tests were not performed or did not meet the cut points set
for a diagnosis of HELLP, the participants were classified as having severe PE, potentially
leading to the underreporting of HELLP. If indeed some HELLP cases had been misclassified as
sPE, any observed differences between these groups would be attenuated, thus we do not
anticipate that misclassification can explain our results.
The results of our exploratory study support the use of GA as a predictor of adverse
neonatal outcomes over the diagnosis of HELLP Syndrome vs. sPE. Specifically, we observed
that models developed with GA as a predictor were equally as good as and/or better at predicting
adverse neonatal outcomes compared to models developed with diagnosis of HELLP and severe
PE. Further research is suggested to examine the clinical utility of these diagnoses with respect to
maternal outcomes and to refine our findings with a larger study population.
37
References
Abramovici, D., Friedman, S. A., Mercer, B. M., Audibert, F., Kao, L., & Sibai, B. M. (1999).
Neonatal outcome in severe preeclampsia at 24 to 36 weeks' gestation: does the HELLP
(hemolysis, elevated liver enzymes, and low platelet count) syndrome matter?. American
journal of obstetrics and gynecology, 180(1), 221-225.

Aloizos, S., Seretis, C., Liakos, N., Aravosita, P., Mystakelli, C., Kanna, E., & Gourgiotis, S.
(2013). HELLP syndrome: understanding and management of a pregnancy-specific
disease. Journal of obstetrics and gynaecology : the journal of the Institute of Obstetrics
and Gynaecology, 33(4), 331–337. https://doi.org/10.3109/01443615.2013.775231

American College of Obstetricians and Gynecologists. (2013). Hypertension in pregnancy.
Report of the American College of Obstetricians and Gynecologists’ task force on
hypertension in pregnancy. Obstetrics and gynecology, 122(5), 1122-1131.

Gul, A., Cebeci, A., Aslan, H., Polat, I., Ozdemir, A., & Ceylan, Y. (2005). Perinatal outcomes
in severe preeclampsia-eclampsia with and without HELLP syndrome. Gynecologic and
obstetric investigation, 59(2), 113–118. https://doi-
org.libproxy1.usc.edu/10.1159/000082648

Haddad, B., Barton, J. R., Livingston, J. C., Chahine, R., & Sibai, B. M. (2000). HELLP
(hemolysis, elevated liver enzymes, and low platelet count) syndrome versus severe
preeclampsia: onset at < or =28.0 weeks' gestation. American journal of obstetrics and
gynecology, 183(6), 1475–1479. https://doi-
org.libproxy1.usc.edu/10.1067/mob.2000.106975

Haram, K., Svendsen, E., & Abildgaard, U. (2009). The HELLP syndrome: clinical issues and
management. A Review. BMC pregnancy and childbirth, 9, 8.
https://doi.org/10.1186/1471-2393-9-8

Harms, K., Rath, W., Herting, E., & Kuhn, W. (1995). Maternal hemolysis, elevated liver
enzymes, low platelet count, and neonatal outcome. American journal of
perinatology, 12(01), 1-6.

Kallela, J., Jääskeläinen, T., Kortelainen, E., Heinonen, S., Kajantie, E., Kere, J., ... & Laivuori,
H. (2016). The diagnosis of pre-eclampsia using two revised classifications in the Finnish
Pre-eclampsia Consortium (FINNPEC) cohort. BMC pregnancy and childbirth, 16(1), 1-
7.

Kınay, T., Küçük, C., Kayıkçıoğlu, F., & Karakaya, J. (2015). Severe Preeclampsia versus
HELLP Syndrome: Maternal and Perinatal Outcomes at <34 and ≥34 Weeks'
Gestation. Balkan medical journal, 32(4), 359–363. https://doi-
org.libproxy1.usc.edu/10.5152/balkanmedj.2015.15777

38
Mehrabian, F., Mohammadizadeh, F., Moghtaderi, N., & Najafian, A. (2012). Comparison of
placental pathology between severe preeclampsia and HELLP syndrome. Archives of
gynecology and obstetrics, 285(1), 175–181. https://doi.org/10.1007/s00404-011-1948-0

Menzies, J., Magee, L. A., Macnab, Y. C., Ansermino, J. M., Li, J., Douglas, M. J., Gruslin, A.,
Kyle, P., Lee, S. K., Moore, M. P., Moutquin, J. M., Smith, G. N., Walker, J. J., Walley,
K. R., Russell, J. A., & von Dadelszen, P. (2007). Current CHS and NHBPEP criteria for
severe preeclampsia do not uniformly predict adverse maternal or perinatal
outcomes. Hypertension in pregnancy, 26(4), 447–462. https://doi-
org.libproxy1.usc.edu/10.1080/10641950701521742

Myatt, L., & Roberts, J. M. (2015). Preeclampsia: Syndrome or Disease?. Current hypertension
reports, 17(11), 83. https://doi.org/10.1007/s11906-015-0595-4

Neggers, Y. H. (2018). Gestational Age and Pregnancy Outcomes. Pregnancy and Birth
Outcomes, 9.

Olsen, I. E., Groveman, S. A., Lawson, M. L., Clark, R. H., & Zemel, B. S. (2010). New
intrauterine growth curves based on United States data. Pediatrics, 125(2), e214–e224.
https://doi.org/10.1542/peds.2009-0913

PASS 2020 Power Analysis and Sample Size Software (2020). NCSS, LLC. Kaysville, Utah,
USA, ncss.com/software/pass.

Platt, M. J. (2014). Outcomes in preterm infants. Public health, 128(5), 399-403.

Publications Committee, Society for Maternal-Fetal Medicine, & Sibai, B. M. (2011). Evaluation
and management of severe preeclampsia before 34 weeks' gestation. American journal of
obstetrics and gynecology, 205(3), 191–198. https://doi-
org.libproxy1.usc.edu/10.1016/j.ajog.2011.07.017

Rath, W., Faridi, A., & Dudenhausen, J. W. (2000). HELLP syndrome. Journal of perinatal
medicine, 28(4), 249–260. https://doi.org/10.1515/JPM.2000.033

Schisterman, E. F., Cole, S. R., & Platt, R. W. (2009). Overadjustment bias and unnecessary
adjustment in epidemiologic studies. Epidemiology (Cambridge, Mass.), 20(4), 488–495.
https://doi.org/10.1097/EDE.0b013e3181a819a1

Smulian, J., Shen-Schwarz, S., Scorza, W., Kinzler, W., & Vintzileos, A. (2004). A
clinicohistopathologic comparison between HELLP syndrome and severe
preeclampsia. The journal of maternal-fetal & neonatal medicine : the official journal of
the European Association of Perinatal Medicine, the Federation of Asia and Oceania
Perinatal Societies, the International Society of Perinatal Obstetricians, 16(5), 287–293.
https://doi.org/10.1080/14767050400018015

StataCorp. (2019). Stata Statistical Software: Release 16. College Station, TX: StataCorp LP.
39

Turgut, A., Demirci, O., Demirci, E., & Uludoğan, M. (2010). Comparison of maternal and
neonatal outcomes in women with HELLP syndrome and women with severe
preeclampsia without HELLP syndrome. Journal of prenatal medicine, 4(3), 51–58.

Vinnars, M. T., Wijnaendts, L. C., Westgren, M., Bolte, A. C., Papadogiannakis, N., & Nasiell,
J. (2008). Severe preeclampsia with and without HELLP differ with regard to placental
pathology. Hypertension, 51(5), 1295-1299.

Weiner, E., Schreiber, L., Grinstein, E., Feldstein, O., Rymer-Haskel, N., Bar, J., & Kovo, M.
(2016). The placental component and obstetric outcome in severe preeclampsia with and
without HELLP syndrome. Placenta, 47, 99–104.
https://doi.org/10.1016/j.placenta.2016.09.012
Wilcox, A. J., Weinberg, C. R., & Basso, O. (2011). On the pitfalls of adjusting for gestational
age at birth. American journal of epidemiology, 174(9), 1062–1068.
https://doi.org/10.1093/aje/kwr230

Wilson, M. L., Brueggmann, D., Desmond, D. H., Mandeville, J. E., Goodwin, T. M., & Ingles,
S. A. (2011). A fetal variant in the GCM1 gene is associated with pregnancy induced
hypertension in a predominantly hispanic population. International journal of molecular
epidemiology and genetics, 2(3), 196–206.

Wilson, M. L., Desmond, D. H., Goodwin, T. M., Miller, D. A., & Ingles, S. A. (2009). Maternal
and fetal variants in the TGF-beta3 gene and risk of pregnancy-induced hypertension in a
predominantly Latino population. American journal of obstetrics and gynecology, 201(3),
295.e1–295.e2955. https://doi.org/10.1016/j.ajog.2009.05.038

40
Appendices
Appendix A.
Figure A1. Residual Analysis: Delta c
2
for Model 1A

Figure A2. Residual Analysis: Delta Deviance for Model 1A

41
Figure A3. Residual Analysis: Delta Beta for Model 1A

Figure A4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 1B

42
Appendix B.
Figure B1. Residual Analysis: Delta c
2
for Model 1B

Figure B2. Residual Analysis: Delta Deviance for Model 1B

43

Figure B3. Residual Analysis: Delta Beta for Model 1B

Figure B4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 1B

44
Appendix C.
Figure C1. Residual Analysis: Delta c
2
for Model 2A

Figure C2. Residual Analysis: Delta Deviance for Model 2A

45
Figure C3. Residual Analysis: Delta Beta for Model 2A

Figure C4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 2A

46
Appendix D.
Figure D1. Residual Analysis: Delta c
2
for Model 2B

Figure D2. Residual Analysis: Delta Deviance for Model 2B

47
Figure D3. Residual Analysis: Delta Beta for Model 2B

Figure D4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 2B

48
Appendix E.
Figure E1. Residual Analysis: Delta c
2
for Model 3A

Figure E2. Residual Analysis: Delta Deviance for Model 3A

49
Figure E3. Residual Analysis: Delta Beta for Model 3A

Figure E4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 3A

50
Appendix F.
Figure F1. Residual Analysis: Delta c
2
for Model 3B

Figure F2. Residual Analysis: Delta Deviance for Model 3B

51
Figure F3. Residual Analysis: Delta Beta for Model 3B

Figure F4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 3B

52
Appendix G.
Figure G1. Residual Analysis: Delta c
2
for Model 4A

Figure G2. Residual Analysis: Delta Deviance for Model 4A

53

Figure G3. Residual Analysis: Delta Beta for Model 4A

Figure G4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 4A

54
Appendix H.
Figure H1. Residual Analysis: Delta c
2
for Model 4B

Figure H2. Residual Analysis: Delta Deviance for Model 4B

55
Figure H3. Residual Analysis: Delta Beta for Model 4B

Figure H4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 4B

56
Appendix I.
Figure I1. Residual Analysis: Delta c
2
for Model 5A

Figure I2. Residual Analysis: Delta Deviance for Model 5A

57

Figure I3. Residual Analysis: Delta Beta for Model 5A

Figure I4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 5A
58
Appendix J.
Figure J1. Residual Analysis: Delta c
2
for Model 5B

Figure J2. Residual Analysis: Delta Deviance for Model 5B

59
Figure J3. Residual Analysis: Delta Beta for Model 5B

Figure J4. Graph of Sensitivity and Specificity versus Probability Cutoff for Model 5B

Abstract (if available)

Abstract Background: Preeclampsia with severe features (sPE) and Hemolysis, Elevated Liver Enzymes, and Low Platelet Count Syndrome (HELLP) are multisystemic gestational complications that can have serious maternal and neonatal implications. Objective: We aim to examine whether the clinical diagnosis of HELLP Syndrome or sPE is more predictive of neonatal outcomes than gestational age alone. Methods: We developed a total of 10 predictive models using clinical and laboratory data obtained from the medical records of 165 women with self-identified HELLP syndrome recruited from 2 online HELLP resources. We defined 5 separate neonatal outcomes consisting of combinations of the following: small for gestational age (SGA) defined by Olsen et al (2010), very low birthweight (<1500 g), intrauterine growth restriction, neonatal death, and/or a 1 minute Apgar score of 4 or less. A predictive model was developed for each neonatal outcome (5) for each main exposure of interest (The American College of Obstetricians and Gynecologists (ACOG) diagnosis of HELLP versus sPE and gestational age at delivery > 34 weeks versus ≤ 34 weeks) using multivariable logistic regression. For each outcome, we compared the area under the curve between the ACOG diagnostic model and the gestational age at delivery model. If the models were significantly different (p<0.05), the model with higher AUC was identified as the preferred model. Results: We found that gestational age was a better predictor than ACOG diagnosis of 1) SGA, very low birthweight, IUGR, and/or an Apgar score of 4 or less and 2) SGA, very low birthweight, IUGR, an Apgar score of 4 or less, and/or neonatal death (p=0.002 and p=0.031, respectively). There was not a significant difference between the exposures for the following adverse neonatal events: 1) neonatal death (p=0.18), 2) IUGR and/or an Apgar of 4 or less (p=0.80), and 3) neonatal death, IUGR and/or an Apgar of 4 or less (p=0.46). Conclusion: The results of our exploratory study support the use of GA as a predictor of adverse neonatal outcomes over the diagnosis of HELLP Syndrome vs. sPE. Specifically, we observed that models developed with GA as a predictor were equally as good as or better at predicting adverse neonatal outcomes compared to models developed with diagnosis of HELLP and sPE.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

sFLT-1 gene polymorphisms and risk of severe-spectrum hypertensive disorders of pregnancy

PDF

HIF-1α gene polymorphisms and risk of severe-spectrum hypertensive disorders of pregnancy: a pilot triad-based case-control study

PDF

Preeclampsia and occurrence of neurological outcomes in the child: a meta-analysis

PDF

Effect of biomass fuel exposure on infant respiratory health outcomes in Bangladesh

PDF

Assessing the impact of air pollution on adverse birth outcomes in a low resource setting

PDF

Predictive factors of breast cancer survival: a population-based study

PDF

Predicting hospital length of stay (LOS) using the National Inpatient Sample

PDF

Surgical aortic arch intervention at the time of extended ascending aortic replacement is associated with increased mortality

PDF

Association of maternal and environmental factors with infant feeding behaviors in a birth cohort study

PDF

Comparison of participant and study partner predictions of cognitive impairment in the Alzheimer's disease neuroimaging initiative 3 study

PDF

Spatial modeling of non-tailpipe emissions and its association with children's lung function

PDF

Survival of children and adolescents with low-risk non-rhabdomyosarcoma soft tissue sarcomas (NRSTS) treated with surgery only: an analysis of 234 patients from the Children’s Oncology Group stud...

PDF

Disparities in gallbladder, intra-hepatic bile duct, and other biliary cancers among multi-ethnic populations: a California Cancer Registry study

PDF

Association of single nucleotide polymorphisms in GCK, GCKR and PNPLA3 with type 2 diabetes related quantitative traits in Mexican-American population

PDF

Effects of post-menopausal hormone therapy on arterial stiffness in the ELITE trial

PDF

Carcinogenic exposures in racial/ethnic groups

PDF

Developmental trajectories of sensory patterns in young children with and without autism spectrum disorder: a longitudinal population-based study from infancy to school age

PDF

Selected papers on the evaluation of healthcare costs of prematurity and necrotizing enterocolitis using large retrospective databases

PDF

Investigating a physiological pathway for the effect of guided imagery on insulin resistance

Asset Metadata

Creator Hauptman, Isabella Sarah (author)

Core Title Predicting neonatal outcomes among women diagnosed with severe preeclampsia and HELLP syndrome: a comparison of models

School Keck School of Medicine

Degree Master of Science

Degree Program Applied Biostatistics and Epidemiology

Publication Date 04/29/2021

Defense Date 04/28/2021

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

Tag ACOG,adverse neonatal outcomes,American College of Obstetricians and Gynecologists,gestational age,gestational hypertensive disorders,HELLP syndrome,hemolysis, elevated liver enzymes, and low platelet count syndrome,neonatal outcomes,OAI-PMH Harvest,predicting neonatal outcomes,preeclampsia,preeclampsia with severe feature,severe preeclampsia

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wilson, Melissa (committee chair), Mack, Wendy (committee member), Mullin, Patrick (committee member)

Creator Email ihauptma@usc.edu,Isabella.hauptman@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-455251

Unique identifier UC11668241

Identifier etd-HauptmanIs-9561.pdf (filename),usctheses-c89-455251 (legacy record id)

Legacy Identifier etd-HauptmanIs-9561.pdf

Dmrecord 455251

Document Type Thesis

Rights Hauptman, Isabella Sarah

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 a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA