Machine learning‐based placental clusters and their associations with adverse pregnancy outcomes

Document Type

Article

Publication Date

11-28-2022

Publication Title

Paediatric and Perinatal Epidemiology, Volume37, Issue4 Special Issue:Big Data in Perinatal Epidemiology Methodologic Considerations May 2023

Keywords

Muskie, Public Health, Staff Scholarship, Machine Learning, placental clusters

Abstract

Background

Placental abnormalities have been described in clinical convenience samples, with predominately adverse outcomes. Few studies have described placental patterns in unselected samples.

Objective

We aimed to investigate associations between co-occurring placental features and adverse pregnancy outcomes in a prospective cohort of singletons.

Methods

Data were from the Safe Passage study (U.S. and South Africa, 2007–2015). Before 24 weeks' gestation, participants were randomly invited to donate placental tissue at delivery for blinded, standardised pathological examination. We used hierarchical clustering to construct statistically derived groups using 60 placental features. We estimated associations between the placental clusters and select adverse pregnancy outcomes, expressed as unadjusted and adjusted risk ratios (RRs) and robust 95% confidence intervals (CI).

Results

We selected a 7-cluster model. After collapsing 2 clusters to form the reference group, we labelled the resulting 6 analytic clusters according to the overarching category of their most predominant feature(s): severe maternal vascular malperfusion (n = 117), fetal vascular malperfusion (n = 222), other vascular malperfusion (n = 516), inflammation 1 (n = 269), inflammation 2 (n = 175), and normal (n = 706). Risks for all outcomes were elevated in the severe maternal vascular malperfusion cluster. For instance, in unadjusted analyses, this cluster had 12 times the risk of stillbirth (RR 12.07, 95% CI 4.20, 34.68) and an almost doubling in the risk of preterm delivery (RR 1.93, 95% CI 1.27, 2.93) compared with the normal cluster. Small infant size was more common among the abnormal clusters, with the highest unadjusted RRs observed in the fetal vascular malperfusion cluster (small for gestational age birth RR 2.99, 95% CI 2.24, 3.98, head circumference <10th percentile RR 2.86, 95% CI 1.60, 5.12). Upon adjustment for known risk factors, most RRs attenuated but remained >1.

Conclusion

Our study adds to the growing body of epidemiologic research, finding adverse pregnancy outcomes may occur through etiologic mechanisms involving co-occurring placental abnormalities.

Link to Full Text

Share

COinS