Uncertainty-aware longitudinal prediction of multiple adverse pregnancy outcomes

Risk of pregnancy complications is reassessed at each antenatal visit, but whether earlier predictions add information beyond the most recent one is unclear. Classifying every prediction as low or high risk can leave uncertainty unexpressed. Whether reliability per visit holds across visits is also unclear. This secondary analysis included 9289 nulliparous women from the nuMoM2b cohort (eight US centres, 2010-2013). Logistic regression models predicted gestational diabetes, pre-eclampsia, preterm birth, suspected fetal growth restriction and large for gestational age at 6-13, 16-21 and 22-29 weeks (Visits 3). Mondrian conformal prediction reported 'no' (low risk), 'yes' (high risk), or both (review). To test whether prediction history adds information beyond the Visit 3 prediction, we compared, among women classified as high risk at Visit 3, those first classified as high risk at Visit 2 or 3 (emerging high) with those classified as high risk at all three visits (persistent high), adjusting for Visit 3 predicted probability. The adjusted proportion with the outcome was higher in the emerging-high than the persistent-high group, by 6.1 percentage points for large for gestational age (95% CI -0.4 to 12.2) and 3.8 percentage points for suspected fetal growth restriction (-0.9 to 8.1). Other outcomes showed adjusted differences close to zero. With conformal prediction calibrated to a 90% level at each visit, the reported answers included the actual outcome in 89.6-90.6% of women at each visit (coverage). Coverage decreased when each woman's three visits were considered together (79.0-85.4%). For fetal growth outcomes, how a woman's predicted risk evolved during pregnancy may add information beyond her Visit 3 prediction. For repeated clinical use, the target coverage level should reflect the acceptable risk of incorrect low- or high-risk classifications over a woman's follow-up and the clinical decisions supported at each visit.

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-17
DOI
https://doi.org/10.64898/2026.11.15.26368140
Primary Topic
Gestational Diabetes Research and Management
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Uncertainty-aware longitudinal prediction of multiple adverse pregnancy outcomes

S. Guid Oei, Myrthe van der Ven, M. Beatrijs van der Hout‐van der Jagt, Elisabetta Peri et al.
1 links
medRxiv
Gestational Diabetes Research and Management
preprint

Uncertainty-aware longitudinal prediction of multiple adverse pregnancy outcomes

S. Guid Oei, Myrthe van der Ven, M. Beatrijs van der Hout‐van der Jagt, Elisabetta Peri, Massimo Mischi, Xi Long, Sima Asvadi, Mengzhu Xu, Yanqi Wu, Judith OEH van Laar, Qiong Luo, Chen Chen
preprint en

Abstract

Risk of pregnancy complications is reassessed at each antenatal visit, but whether earlier predictions add information beyond the most recent one is unclear. Classifying every prediction as low or high risk can leave uncertainty unexpressed. Whether reliability per visit holds across visits is also unclear. This secondary analysis included 9289 nulliparous women from the nuMoM2b cohort (eight US centres, 2010-2013). Logistic regression models predicted gestational diabetes, pre-eclampsia, preterm birth, suspected fetal growth restriction and large for gestational age at 6-13, 16-21 and 22-29 weeks (Visits 3). Mondrian conformal prediction reported 'no' (low risk), 'yes' (high risk), or both (review). To test whether prediction history adds information beyond the Visit 3 prediction, we compared, among women classified as high risk at Visit 3, those first classified as high risk at Visit 2 or 3 (emerging high) with those classified as high risk at all three visits (persistent high), adjusting for Visit 3 predicted probability. The adjusted proportion with the outcome was higher in the emerging-high than the persistent-high group, by 6.1 percentage points for large for gestational age (95% CI -0.4 to 12.2) and 3.8 percentage points for suspected fetal growth restriction (-0.9 to 8.1). Other outcomes showed adjusted differences close to zero. With conformal prediction calibrated to a 90% level at each visit, the reported answers included the actual outcome in 89.6-90.6% of women at each visit (coverage). Coverage decreased when each woman's three visits were considered together (79.0-85.4%). For fetal growth outcomes, how a woman's predicted risk evolved during pregnancy may add information beyond her Visit 3 prediction. For repeated clinical use, the target coverage level should reflect the acceptable risk of incorrect low- or high-risk classifications over a woman's follow-up and the clinical decisions supported at each visit.

medRxiv
Fairleigh Dickinson University (US), Zhejiang University-University of Edinburgh Institute (CN), Zhejiang University (CN), Eindhoven University of Technology (NL)
Good health and well-being
Gestational Diabetes Research and Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon