Background
Neonatal morbidity attributable to prematurity predominantly occurs among early preterm births (<32 weeks) rather than late preterm births (32 to <37 weeks). Methods to distinguish early and late preterm births are lacking given the heterogeneity in pathophysiology and risk factors, including maternal obesity. Although preterm births are often characterized by clinical presentation (spontaneous or clinically indicated), classifying deliveries by placental features detected on histopathology reports may help identify subgroups of preterm births with similar etiology and risk factors. Latent class analysis is an empirical approach to characterize preterm births on the basis of observed combinations of placental features.
Objective
To identify histopathologic markers that can distinguish early (<32 weeks) and late preterm births (32 to <37 weeks) that are also associated with maternal obesity and neonatal outcomes.
Study Design
Women with a singleton preterm birth at University of Pittsburgh Medical Center Magee-Womens Hospital (Pittsburgh, PA) from 2008 to 2012 and a placental evaluation (89% of preterm births) were stratified into early (n=900, 61% spontaneous) and late preterm births (n=3362, 57% spontaneous). Prepregnancy body mass index was self-reported at first prenatal visit and 16 abstracted placental features were analyzed. Placental subgroups (ie, latent classes) of early and late preterm births were determined separately by latent class analysis of placental features. The optimal number of latent classes was selected by comparing fit statistics. The probability of latent class membership across prepregnancy body mass indexes was estimated in early preterm births and in late preterm births by an extension of multinomial regression called pseudo-class regression, adjusting for race, smoking, education, and parity. The frequencies of severe neonatal morbidity (composite outcome: respiratory distress, bronchopulmonary dysplasia, intraventricular hemorrhage, necrotizing enterocolitis, periventricular leukomalacia, patent ductus arteriosus, and retinopathy of prematurity), small-for-gestational-age, and length of neonatal intensive care unit stay were compared across latent classes by chi-square and Kruskal-Wallis tests.
Results
Early preterm births were grouped into 4 latent classes based on placental histopathologic features: acute inflammation (38% of cases), maternal vascular malperfusion with inflammation (29%), maternal vascular malperfusion (25%), and fetal vascular thrombosis with hemorrhage (8%). As body mass index increased from 20 to 50kg/m 2 , the probability of maternal vascular malperfusion and fetal vascular thrombosis with hemorrhage increased, whereas the probability of maternal vascular malperfusion with inflammation decreased. There was minimal change in the probability of acute inflammation with increasing body mass index. Late preterm births also had 4 latent classes: maternal vascular malperfusion (22%), acute inflammation (12%), fetal vascular thrombosis with hemorrhage (9%), and low-risk pathology (58%). Body mass index was not associated with major changes in likelihood of the latent classes in late preterm births. Associations between body mass index and likelihood of the latent classes were not modified by type of delivery (spontaneous or indicated) in early or late preterm births. Maternal malperfusion and fetal vascular thrombosis with hemorrhage were associated with greater neonatal morbidity than the other latent classes in early and late preterm births.
Conclusion
Obesity may predispose women to early but not late preterm birth through placental vascular impairment. Latent class analysis of placental histopathologic data provides an evidence-based approach to group preterm births with shared underlying etiology and risk factors.
Introduction
In the United States, approximately 80% of preterm births (PTBs) are late PTBs (32 to <37 weeks of gestation), but 75% of neonatal deaths among PTBs occur in early PTBs (<32 weeks). Early and late PTB may have distinct risk factors (eg, maternal obesity) with variable pathophysiology. In addition, PTBs are classified as spontaneous or clinically indicated PTBs. Yet, this classification is not informative of PTB etiology and findings between clinical classification and neonatal morbidity risk are conflicting. Placental histopathologic evaluations are routinely conducted to inform clinicians’ assessments of PTB causes. These evaluations are an underutilized resource for understanding the pathophysiology and neonatal sequelae of early and late PTBs.
Why was this study conducted?
To classify early and late preterm births by pathophysiology using latent class analysis of placental histopathologic data and to determine if latent classes were associated with prepregnancy obesity and neonatal morbidity.
Key findings
Early preterm births grouped into 2 known placental phenotypes: acute inflammation and maternal vascular malperfusion, and 2 novel phenotypes: maternal malperfusion with chorioamnionitis and fetal vascular thrombosis with hemorrhage. Half of late preterm births had no evident predominant placental pathology. From 20 to 50 kg/m 2 , prepregnancy body mass index doubled the probability of maternal vascular malperfusion in early but not late preterm births. Small-for-gestational-age and severe neonatal morbidity were most prevalent among deliveries with placental malperfusion.
What does this add to what is known?
A novel approach to classify preterm births into subgroups with distinct placental etiologies and risk factors like obesity.
Interpreting placental histopathology is challenging because findings can be incidental or related to physiological processes like labor. A high proportion (28%–78%) of uncomplicated term pregnancies are reported to have at least 1 histopathologic feature. , In the absence of an approach to distinguish healthy from complicated placentas, histopathology offers limited insight for pediatric follow-up. Clusters of histopathologic features may be more reflective than single measures of true pathology and prognostic of neonatal outcomes.
Various approaches have previously classified PTBs by patterns of histopathologic features. Previous approaches have been based on anticipated groupings of histopathology (eg, inflammatory or vascular impairment) using correlative measures, factor analysis, and predefined groups based on expert opinion. These methods may presuppose placental pathology groups, miss key histopathologic features and/or lack strong associations with clinical outcomes. Empirical classification, as applied here, may identify novel placental histopathologic patterns in PTB.
Latent class analysis (LCA) is a statistical method that classifies individuals into groups based on different patterns of observed data. Groups are called latent classes because they are estimated, not directly measured. For example, grouping individuals into personality types based on survey response patterns. LCA is appealing over other clustering methods (eg, hierarchical or k-means clustering) because standard model fit statistics inform the appropriate number of latent classes, LCA allows missing data, and latent classes are easy to interpret. , Covariates can be added to LCA models to test if risk factors like obesity predict latent class membership.
We aimed to: (1) classify early and late PTBs into placental latent classes based on observed patterns of placental histopathologic features by LCA and (2) determine if prepregnancy body mass index (BMI) was associated with specific latent classes. We focused on prepregnancy BMI because our approach may help understand inconsistent findings on obesity and PTB. Evidence suggests obesity is associated with an increased risk of early PTB (<32 weeks), but not late PTB. , Obesity may alter the risk of early and late PTB by different placental mechanisms. For clinical relevance of our classification approach, we compared the prevalence of severe neonatal morbidity (composite outcome: respiratory distress syndrome, bronchopulmonary dysplasia, intraventricular hemorrhage, necrotizing enterocolitis, patent ductus arteriosus, periventricular leukomalacia, retinopathy of prematurity), length of neonatal intensive care unit (NICU) stay, and small-for-gestational-age (SGA) across latent classes.
Materials and Methods
Study participants
Delivery data were collected from the Magee Obstetric Maternal & Infant (MOMI) database. We included live singleton early (20 to <32 weeks) and late PTBs (32 to <37 weeks) delivered between 2008 and 2012 with available placental pathology data (94% of early PTBs and 88% of late PTBs). We excluded stillbirths (0.5% of PTBs) because our automated placental report abstraction approach did not reliably distinguish placental findings. Women with multifetal gestations were excluded because placental findings in multifetal pregnancies are different from those of singleton pregnancies, and placental findings were unable to be linked to each fetus. , The University of Pittsburgh Institutional Review Board approved this project (STUDY20050303), and no consent was needed because data were deidentified.
Placental data
Placental histopathologic features considered for LCA were extracted from pathology reports conducted by 2 placental pathologists following a standardized protocol and linked to the MOMI database by an automated process previously described. Histopathology definitions were adapted from the 2014 Amsterdam criteria ( Supplemental Table 1 ). , Pathologic features included markers of inflammation (acute chorioamnionitis, vasculitis, funisitis, deciduitis, villitis, intervillitis), maternal vascular malperfusion (MVM: villous infarct, intraparenchymal hemorrhage, subchorionic hemorrhage, advanced villous maturation, decidual vasculopathy, villous agglutination, intervillous thrombus), fetal vascular malperfusion (avascular villi, stromal-vascular karyorrhexis, fetal vascular thrombosis, chorangiosis), and other markers (placental growth, chorioangioma, chorangiomatosis, delayed villous maturation, dysmaturity). Stromal-vascular karyorrhexis and avascular villi were combined given that stromal-vascular karyorrhexis progresses to avascular villi. There was excellent agreement for review of placental slides between clinical pathology reports and a pathologist blinded to all clinical information except gestational age (W.T.P.) for features of inflammation (82%) and MVM (kappa=0.78).
Anthropometry
Prepregnancy BMI was calculated as a ratio of weight (kg) to height (m) squared (kg/m 2 ) using self-reported data at first prenatal visit. Self-reported prepregnancy weight was highly correlated with first measured weight in pregnancy (r=0.99) at UPMC Magee-Womens Hospital.
Pregnancy characteristics
Gestational age was determined using best obstetrical estimates based on first/second trimester ultrasound in conjunction with last menstrual period. Pregnancy and neonatal outcomes were extracted from medical records based on International Classification of Diseases, Ninth Revision codes provided in the Supplemental Methods sections. Pregnancy complications included gestational diabetes mellitus, gestational hypertension, preeclampsia/eclampsia, cervical shortening, clinical chorioamnionitis, and preterm premature rupture of membranes (PPROM). PTBs were classified as early PTB for deliveries <32 weeks of gestation and late PTB for deliveries at 32 to <37 weeks. Early PTBs were not further classified because there were few PTBs at <28 weeks (n=374, 7.8% of PTBs), and PTBs at <32 weeks are postulated to have similar etiologies. , A spontaneous PTB was defined as a pregnancy with spontaneous onset of contractions or premature rupture of fetal membranes before 37 weeks (irrespective of induction or cesarean delivery after labor). A clinically indicated PTB was defined as an induced pregnancy or cesarean delivery before 37 weeks. Adverse neonatal outcomes included SGA (birthweight <10th percentile using the Alexander birthweight curve), length of NICU stay (infant hospital discharge date minus date of delivery), and a composite score for severe neonatal morbidity: respiratory distress syndrome, bronchopulmonary dysplasia, intraventricular hemorrhage, necrotizing enterocolitis, periventricular leukomalacia, patent ductus arteriosus, and retinopathy of prematurity.
Statistical analyses
Frequencies of individual placental features were compared between early and late PTBs by chi-square and Fisher exact tests. Early (<32 weeks) and late (32 to <37weeks) PTBs were analyzed separately for: (1) grouping deliveries into classes reflective of distinct placental pathology, (2) predicting class membership across prepregnancy BMI, and (3) comparing clinical outcomes across classes.
Deliveries were classified on the basis of observed patterns of 16 placental histopathologic features using LCA. We call these groups placental latent classes because they are empirically derived from observed histopathology patterns, and not directly measured by the pathologist. Rare placental features (< 1% of PTBs) were excluded because of limited numbers. The optimal number of placental latent classes in early and late PTBs was determined by comparing 6 models ranging from 1 to 6 latent classes using model fit statistics. We excluded models that generated rare latent classes (<5%) and models with poor entropy (<0.70), a measure of how accurately pregnancies are classified. , Latent classes were labeled on the basis of the combination of placental features with probabilities of occurring >25% within that class. Among late PTBs, we excluded deliveries at 36 weeks as a sensitivity analysis to assess the effect of PTB misclassification attributable to potential inaccurate dating. Few women (4.5%) had more than 1 pregnancy in the cohort. Restricting to the first pregnancy did not change the number or composition of placental features within latent classes. Therefore, we included all pregnancies in the analysis.
We evaluated if prepregnancy BMI was associated with likelihood of the placental latent classes by 4 regression methods (pseudo-class, most-likely class, probability-weighted, and single-step latent class regression) described in the Supplemental Methods section. , Briefly, latent classes are statistically estimated; therefore, pregnancies may be misclassified. Misclassification adds variability that is accounted for by these regression methods. Assessing consistency in findings across methods helps identify if errors were introduced by the estimation method. Pseudo-class regression is the preferred method because it adequately accounts for variability from potential latent class misclassification and is flexible in handling missing data. Associations were visualized by predicted probability plots. Models were adjusted for maternal race, education, parity, and smoking. We examined whether associations between BMI and likelihood of latent classes were modified by interaction with clinical presentation of PTB (spontaneous vs indicated), race, and fetal sex. , For any variable found to be an effect modifier ( P <.10), stratified results are presented.
We compared the proportions of pregnancy complications and neonatal morbidities across placental latent classes (deliveries assigned to most-likely latent class) by chi-square and Fisher exact tests for categorical outcomes and Kruskal-Wallis tests for continuous outcomes. An alpha level of 0.05 was assumed for nominal significance and a Bonferroni-corrected alpha was used for significance after multiple comparisons of pregnancy (0.05/8=0.006) and neonatal outcomes (0.05/10=0.005).
Prepregnancy BMIs were missing in 42% of PTBs. Women with and without a reported prepregnancy BMI had comparable maternal and pathology characteristics ( Supplemental Tables 2 and 3 ). Missing data were imputed by multiple imputation with chained equations. We compared regression estimates by complete-case analysis and by imputation to assess sensitivity to missingness. Additional details are provided in the Supplemental Methods section. Analyses were conducted in RStudio (RStudio, Boston, MA).
Results
There were 4262 liveborn singleton PTBs with placental histopathology at UPMC Magee-Womens Hospital between 2008 and 2012; 20% were early PTBs (<32 weeks) and 80% were late PTBs (32 to <37 weeks) ( Figure 1 ). Women with early PTBs were younger, more likely to be Black, less likely to have a college education, and less likely to have diabetes mellitus (preexisting or gestational) than women with late PTB ( Table 1 ). Women with early PTB were more likely to have PPROM and spontaneous PTB, and less likely to have a SGA baby than women with late PTB. There were 797 (89%) early PTBs and 2392 (71%) late PTBs with at least 1 placental histopathologic feature. Early PTBs had higher frequencies of histopathologic features of acute inflammation and MVM than late PTBs ( Figure 2 ).
| Maternal characteristics | Missing (n) | Early PTB (20 to <32 wk) | Missing (n) | Late PTB (32 to <37 wk) | P value |
|---|---|---|---|---|---|
| Maternal age (y) | 0 | 27.3±6.3 | 0 | 28.4±6.2 | <.001 |
| Race, n (%) | 5 | 14 | .003 | ||
| White | 613 (68.5) | 2434 (72.7) | |||
| Black | 255 (28.5) | 780 (23.3) | |||
| Other | 27 (3.0) | 134 (4.0) | |||
| Education, n (%) | 20 | 33 | <.001 | ||
| High school/GED or less | 432 (49.1) | 1263 (37.9) | |||
| Some college | 448 (50.9) | 2066 (62.1) | |||
| Prepregnancy BMI, kg/m 2 | 445 | 26.9±7.5 | 1365 | 26.3±6.8 | .274 |
| Weight status, n (%) | 445 | 1365 | .162 | ||
| Underweight (<18.5 kg/m 2 ) | 30 (6.6) | 109 (5.5) | |||
| Lean (18.5 to <25 kg/m 2 ) | 203 (44.6) | 939 (47.0) | |||
| Overweight (25 to <30 kg/m 2 ) | 93 (20.4) | 465 (23.3) | |||
| Obese (>30 kg/m 2 ) | 129 (28.4) | 484 (24.2) | |||
| Smoking in pregnancy, n (%) | 12 | 214 (24.1) | 27 | 774 (23.2) | .608 |
| Nulliparous at enrollment, n (%) | 0 | 463 (51.4) | 5 | 1593 (47.5) | .037 |
| Multiple abortion history, n (%) | 0 | 140 (15.6) | 5 | 487 (14.5) | .462 |
| Diabetes mellitus, n (%) | 0 | 0 | <.001 | ||
| None | 821 (91.2) | 2851 (84.8) | |||
| Preexisting | 47 (5.2) | 314 (9.3) | |||
| Gestational | 32 (3.6) | 197 (5.9) | |||
| Chronic hypertension, n (%) | 0 | 98 (10.9) | 0 | 280 (8.3) | .020 |
| Hypertensive disorders, n (%) | 2 | 7 | .207 | ||
| No hypertensive disorders | 626 (69.7) | 2390 (71.2) | |||
| Gestational hypertension | 26 (2.9) | 125 (3.7) | |||
| Preeclampsia, HELLP syndrome, eclampsia | 246 (27.4) | 840 (25.0) | |||
| Delivery characteristics | |||||
| PPROM, n (%) | 0 | 345 (38.3) | 0 | 988 (29.4) | <.001 |
| Clinical presentation, n (%) | 3 | 2 | .034 | ||
| Indicated | 350 (39.0) | 1446 (43.0) | |||
| Spontaneous | 547 (61.0) | 1914 (57.0) | |||
| Gestational weight gain | |||||
| Weight gain mean (kg) | 494 | 8.9±6.6 | 1476 | 12.8±6.8 | <.001 |
| Weight gain Z-score | 505 | −0.21±1.22 | 1505 | −0.10±1.10 | .032 |
| Infant birthweight (g) | 26 | 1138±459 | 14 | 2493±560 | <.001 |
| Small-for-gestational-age, n (%) | 21 | 113 (12.9) | 13 | 581 (17.3) | .002 |
| Male fetal sex, n (%) | 0 | 504 (56.0) | 0 | 1821 (54.2) | .345 |
a Continuous variables are represented as mean±standard deviation
b Continuous variables were compared by t tests for normally distributed variables and Mann-Whitney U tests for skewed data. Categorical variables were compared by chi-square tests.
On the basis of fit statistics, class size, and classification accuracy, early PTBs were grouped into 4 placental latent classes by LCA ( Supplemental Figures 1 and 2 ). Among early PTBs, 38% had acute inflammation, 29% had MVM with chorioamnionitis, 25% had MVM, and 8% had fetal vascular thrombosis (FVT) with hemorrhage ( Figure 3 , A). Late PTBs were also grouped into 4 latent classes ( Supplemental Figures 1 and 3 ). Among late PTBs, 58% had low-risk pathology (no pathologic features with >25% probability of occurring for the latent class), 22% had MVM, 12% had acute inflammation, and 9% had FVT with hemorrhage ( Figure 3 , B). Excluding late PTBs at 36 weeks did not affect the optimal number of latent classes ( Supplemental Figure 4 ).
Among early PTBs, increasing prepregnancy BMI from 20 to 50kg/m 2 was associated with a higher probability of MVM (20.2% to 37.0%) and FVT with hemorrhage (6.6% to 14.9%), and a lower probability of MVM with chorioamnionitis (35.0% to 13.2%) after adjusting for race, education, parity, and smoking ( Figure 4 ). There was minimal change in the probability of acute inflammation (38.2% to 34.8%) with increasing BMI in early PTBs. In late PTBs, increasing BMI from 20 to 50 kg/m 2 was associated with an increased probability of FVT with hemorrhage (7.4% to 14.9%) and a decreased probability of low-risk pathology (60.1% to 53.0%). There was minimal change in the probabilities of MVM (21.4% to 21.7%) and acute inflammation (11.2% to 10.4%) with increasing BMI in late PTBs. There were no interactions between BMI and fetal sex, race, or clinical presentation of PTB (spontaneous vs indicated) for early or late PTB. Associations were consistent across early and late PTBs irrespective of regression method or imputation ( Supplemental Table 4 ).
Among early PTBs, women with acute inflammation were most likely to have a spontaneous delivery (87.5%), PPROM (60.6%), and clinical chorioamnionitis (38.8%) relative to the other latent classes ( Table 2 ). Early PTBs with MVM had the highest frequencies of preeclampsia, eclampsia, and hemolysis, elevated liver enzymes, and low platelet count (HELLP) syndrome (79.7%); severe neonatal morbidity (82.0%); and SGA (31.8%) relative to the other latent classes. Among late PTBs, women with acute inflammation had the highest frequencies of spontaneous delivery (75.0%), PPROM (43.6%), and clinical chorioamnionitis (8.7%) compared with the other latent classes ( Table 3 ). Late PTBs with MVM had the highest prevalence of preeclampsia, eclampsia, and HELLP syndrome (53.1%); SGA (42.1%); and longer median NICU stay (9 days; interquartile range, 18 days), whereas women with FVT with hemorrhage had the highest frequency of severe neonatal morbidity (19.9%) relative to the other latent classes.
| Clinical outcomes | Acute inflammation (n=343) | FVT with hemorrhage (n=71) | MVM (n=202) | MVM with chorioamnionitis (n=284) | P value |
|---|---|---|---|---|---|
| Pregnancy outcomes | |||||
| Diabetes mellitus, n (%) | .115 | ||||
| None | 318 (92.7) | 65 (91.5) | 179 (88.6) | 259 (91.2) | |
| Preexisting | 19 (5.5) | 2 (2.8) | 15 (7.4) | 11 (3.9) | |
| Gestational | 6 (1.8) | 4 (5.6) | 8 (4.0) | 14 (4.9) | |
| Hypertensive disorders, n (%) | <.001 c | ||||
| No hypertensive disorders | 318 (93.0) | 39 (54.9) | 39 (19.3) | 230 (81.3) | |
| Gestational hypertension | 13 (3.8) | 5 (7.0) | 2 (1.0) | 6 (2.1) | |
| Preeclampsia, HELLP syndrome, eclampsia | 11 (3.2) | 27 (38.0) | 161 (79.7) | 47 (16.6) | |
| Cervical shortening, n (%) | 11 (3.2) | 0 (0.0) | 0 (0.0) | 9 (3.2) | .014 b |
| Clinical chorioamnionitis, n (%) | 134 (38.8) | 8 (11.3) | 5 (2.5) | 25 (8.8) | <.001 c |
| PPROM, n (%) | 208 (60.6) | 22 (31.0) | 20 (9.9) | 95 (33.5) | <.001 c |
| Delivery method, n (%) | <.001 c | ||||
| Indicated | 43 (12.5) | 39 (56.5) | 169 (83.7) | 99 (35.0) | |
| Spontaneous | 300 (87.5) | 30 (43.5) | 33 (16.3) | 184 (65.0) | |
| Cesarean delivery, n (%) | 118 (35.9) | 48 (68.6) | 166 (82.2) | 152 (55.9) | <.001 c |
| Male fetal sex, n (%) | 188 (54.8) | 41 (57.7) | 104 (51.5) | 171 (60.2) | .262 |
| Neonatal outcomes | |||||
| Small-for-gestational-age, n (%) | 15 (4.5) | 14 (20.3) | 63 (31.8) | 21 (7.6) | <.001 c |
| Severe neonatal morbidity, n (%) e | 228 (68.1) | 49 (71.0) | 159 (82.0) | 214 (76.2) | .004 c |
| Respiratory distress, n (%) | 200 (59.7) | 43 (62.3) | 140 (72.2) | 191 (68.0) | .020 b |
| Bronchopulmonary dysplasia, n (%) | 74 (22.1) | 12 (17.4) | 32 (16.5) | 47 (16.7) | .269 |
| Intraventricular hemorrhage, n (%) | 90 (26.2) | 16 (22.5) | 39 (19.3) | 60 (21.1) | .245 |
| Necrotizing enterocolitis, n (%) | 32 (9.3) | 2 (2.8) | 23 (11.4) | 24 (8.5) | .162 |
| Patent ductus arteriosus, n (%) | 58 (16.9) | 15 (21.1) | 46 (22.8) | 57 (20.1) | .389 |
| Periventricular leukomalacia, n (%) | 12 (3.5) | 0 (0.0) | 3 (1.5) | 7 (2.5) | .299 |
| Retinopathy of prematurity, n (%) | 42 (12.2) | 10 (14.1) | 20 (9.9) | 26 (9.2) | .479 |
| Median days in NICU (IQR) | 30.0 (52.8) | 33.0 (35.5) | 36.0 (28.0) | 31.0 (38.0) | .130 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
