Predicting Lifesaving Maternal Commodity Stock outs in Ethiopian Primary Healthcare Facilities Using Supervised Machine Learning Classifiers: Evidence from EDHS 2024/2025

Abstract Background; Obstetric hemorrhage and hypertensive disorders drive Ethiopia’s high maternal mortality ratio (267/100000 live births), requiring continuous availability of lifesaving maternal commodities (LMCs). Conventional statistics offer retrospective insights rather than facility level risk forecasting. This study developed and validated machine learning classifiers to predict LMC stock outs in Ethiopian primary healthcare facilities using EDHS 2024/2025 data. Methods; A secondary analysis evaluated 1245 primary facilities. The binary outcome measured physical absence, expiration or recent stock outs of any 5 WHO tracer commodities. Nineteen features spanning logistics, human resources and infrastructure were analyzed. Five algorithms Logistic Regression, SVC, MLP-ANN, Random Forest and XGBoost were implemented, handling class imbalance via SMOTE and evaluating performance via ROC-AUC, PR-AUC, Decision Curve Analysis (DCA), probability calibration and SHAP interpretability. Results; Overall, 45.8% of facilities experienced stock outs. XGBoost outperformed other models on the test set, achieving a ROC-AUC of 0.894, Accuracy of 88.8%, Sensitivity of 89.5% and Specificity of 88.1%. It exhibited high reliability (Brier Score of 0.084) and superior net clinical benefit via DCA across thresholds from 10% to 85%. SHAP analysis identified non-functional cold chain equipment, lack of IPLS reporting, absence of APTS auditing, irregular bin card reconciliation and lack of professional pharmacy personnel as the top risk drivers. Conclusion; Supervised machine learning algorithms notably XGBoost accurately forecast maternal commodity stock outs. Integrating this architecture into the national eLMIS enables proactive risk flagging, helping the Ministry of Health and EPSS prioritize cold chain upgrades, digital logistics and staffing in vulnerable facilities.

Authors

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11221251/v1
Primary Topic
Artificial Intelligence in Healthcare
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Predicting Lifesaving Maternal Commodity Stock outs in Ethiopian Primary Healthcare Facilities Using Supervised Machine Learning Classifiers: Evidence from EDHS 2024/2025

Ashebir Mamay Gebiru, Yenehun Demelash Gelaw
4 links
Research Square
Artificial Intelligence in Healthcare
preprint

Predicting Lifesaving Maternal Commodity Stock outs in Ethiopian Primary Healthcare Facilities Using Supervised Machine Learning Classifiers: Evidence from EDHS 2024/2025

Ashebir Mamay Gebiru, Yenehun Demelash Gelaw
preprint en

Abstract

Abstract Background; Obstetric hemorrhage and hypertensive disorders drive Ethiopia’s high maternal mortality ratio (267/100000 live births), requiring continuous availability of lifesaving maternal commodities (LMCs). Conventional statistics offer retrospective insights rather than facility level risk forecasting. This study developed and validated machine learning classifiers to predict LMC stock outs in Ethiopian primary healthcare facilities using EDHS 2024/2025 data. Methods; A secondary analysis evaluated 1245 primary facilities. The binary outcome measured physical absence, expiration or recent stock outs of any 5 WHO tracer commodities. Nineteen features spanning logistics, human resources and infrastructure were analyzed. Five algorithms Logistic Regression, SVC, MLP-ANN, Random Forest and XGBoost were implemented, handling class imbalance via SMOTE and evaluating performance via ROC-AUC, PR-AUC, Decision Curve Analysis (DCA), probability calibration and SHAP interpretability. Results; Overall, 45.8% of facilities experienced stock outs. XGBoost outperformed other models on the test set, achieving a ROC-AUC of 0.894, Accuracy of 88.8%, Sensitivity of 89.5% and Specificity of 88.1%. It exhibited high reliability (Brier Score of 0.084) and superior net clinical benefit via DCA across thresholds from 10% to 85%. SHAP analysis identified non-functional cold chain equipment, lack of IPLS reporting, absence of APTS auditing, irregular bin card reconciliation and lack of professional pharmacy personnel as the top risk drivers. Conclusion; Supervised machine learning algorithms notably XGBoost accurately forecast maternal commodity stock outs. Integrating this architecture into the national eLMIS enables proactive risk flagging, helping the Ministry of Health and EPSS prioritize cold chain upgrades, digital logistics and staffing in vulnerable facilities.

Research Square
Good health and well-being, Zero hunger
Artificial Intelligence in Healthcare
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