Development and temporal validation of a model for identifying depressive symptoms in postpartum women

Abstract Background Identifying early postpartum depressive symptoms is important for perinatal mental healthcare. The incremental value of psychosocial scales beyond routine clinical information remains uncertain. We developed and temporally validated models identifying Edinburgh Postnatal Depression Scale (EPDS) screen positivity at a tertiary hospital in Xinjiang, China. Methods This cross-sectional study included 3595 women recruited between February 2023 and May 2026. Clinical information and scores from EPDS, the Connor-Davidson Resilience Scale, Perceived Social Support Scale, and Satisfaction With Life Scale were collected within six days postpartum. EPDS ≥ 10 defined screen positivity. Analyses used Mann–Whitney U and χ² tests. Admission dates determined development (2023–2024; n = 2089) and temporal validation (2025–2026; n = 1402) sets; 104 women with missing or invalid dates were excluded from modelling. Elastic net logistic regression, Explainable Boosting Machine (EBM), CatBoost, and XGBoost were compared using strict nested cross-validation, Platt calibration of out-of-fold predictions, and prespecified high-sensitivity threshold selection. Discrimination, calibration, decision curves, and review workload were evaluated. Results Overall, 892 women (24.8%) screened positive. Temporal validation areas under the receiver operating characteristic curve were 0.837, 0.832, 0.832, and 0.821, respectively. Corresponding precision–recall areas were 0.623, 0.619, 0.612, and 0.599. EBM had a Brier score of 0.129, calibration slope of 1.162, and grouped expected calibration error of 0.031. Its locked threshold of 0.146 yielded 91.4% sensitivity, 50.9% specificity, 36.0% positive predictive value, and 95.1% negative predictive value, requiring approximately 589 reviews per 1000 women. The psychosocial model improved discrimination over the basic clinical model by 0.250 (95% confidence interval 0.210–0.290). Adding clinical predictors to psychosocial scores offered no additional gain (difference − 0.006; 95% confidence interval − 0.016 to 0.004). Resilience, life satisfaction, and perceived social support ranked highest in permutation importance. Conclusions Psychosocial scales provided most information for identifying concurrent EPDS screen positivity; routine clinical variables offered no additional discrimination. The model cannot replace EPDS screening or psychiatric diagnosis. Multicentre prospective evaluation is needed.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11318045/v1
Primary Topic
Maternal Mental Health During Pregnancy and Postpartum
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Development and temporal validation of a model for identifying depressive symptoms in postpartum women

Ying Zhang, Ting Jiang, Wenjing Yong, Xiaofan Ma et al.
Research Square
Maternal Mental Health During Pregnancy and Postpartum
preprint

Development and temporal validation of a model for identifying depressive symptoms in postpartum women

Ying Zhang, Ting Jiang, Wenjing Yong, Xiaofan Ma, Xiang Li, Xin He, Haoying Wang, Shengfeng Chang, Shuqi Zhang
preprint en

Abstract

Abstract Background Identifying early postpartum depressive symptoms is important for perinatal mental healthcare. The incremental value of psychosocial scales beyond routine clinical information remains uncertain. We developed and temporally validated models identifying Edinburgh Postnatal Depression Scale (EPDS) screen positivity at a tertiary hospital in Xinjiang, China. Methods This cross-sectional study included 3595 women recruited between February 2023 and May 2026. Clinical information and scores from EPDS, the Connor-Davidson Resilience Scale, Perceived Social Support Scale, and Satisfaction With Life Scale were collected within six days postpartum. EPDS ≥ 10 defined screen positivity. Analyses used Mann–Whitney U and χ² tests. Admission dates determined development (2023–2024; n = 2089) and temporal validation (2025–2026; n = 1402) sets; 104 women with missing or invalid dates were excluded from modelling. Elastic net logistic regression, Explainable Boosting Machine (EBM), CatBoost, and XGBoost were compared using strict nested cross-validation, Platt calibration of out-of-fold predictions, and prespecified high-sensitivity threshold selection. Discrimination, calibration, decision curves, and review workload were evaluated. Results Overall, 892 women (24.8%) screened positive. Temporal validation areas under the receiver operating characteristic curve were 0.837, 0.832, 0.832, and 0.821, respectively. Corresponding precision–recall areas were 0.623, 0.619, 0.612, and 0.599. EBM had a Brier score of 0.129, calibration slope of 1.162, and grouped expected calibration error of 0.031. Its locked threshold of 0.146 yielded 91.4% sensitivity, 50.9% specificity, 36.0% positive predictive value, and 95.1% negative predictive value, requiring approximately 589 reviews per 1000 women. The psychosocial model improved discrimination over the basic clinical model by 0.250 (95% confidence interval 0.210–0.290). Adding clinical predictors to psychosocial scores offered no additional gain (difference − 0.006; 95% confidence interval − 0.016 to 0.004). Resilience, life satisfaction, and perceived social support ranked highest in permutation importance. Conclusions Psychosocial scales provided most information for identifying concurrent EPDS screen positivity; routine clinical variables offered no additional discrimination. The model cannot replace EPDS screening or psychiatric diagnosis. Multicentre prospective evaluation is needed.

Research Square
Xinjiang Medical University (CN), Xinjiang Production and Construction Corps (CN)
National Natural Science Foundation of China
Good health and well-being
Maternal Mental Health During Pregnancy and Postpartum
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