Development of a machine learning model to predict venous thromboembolism risk in women using combined oral contraceptives
4 affiliations
- 1Saudi Food and Drug Authority, Riyadh, Saudi Arabia
- 2King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia
- 3Department of Obstetrics and Gynecology, King Abdulaziz Medical City, Ministry of the National Guard-Health Affairs, Riyadh, Saudi Arabia
- 4King Abdullah International Medical Research Center, Riyadh, Saudi Arabia
Abstract
Objective: Venous thromboembolism (VTE) is a major cause of morbidity and mortality, with combined oral contraceptives (COCPs) being a modifiable risk factor. Existing clinical tools have limited ability to capture complex predictor interactions. This study aimed to develop and evaluate machine learning (ML) models to predict VTE risk among COCP users.
Methods: This study was conducted using data from the Saudi Real-World Evidence Network (2016–2023). The primary outcome was the first VTE event within 2 years of COCP initiation. Following data partitioning, feature selection was performed on the training set. Elastic-net logistic regression, random forest (RF), extreme gradient boosting (XGBoost), one-class support vector machine and Isolation-Forest were trained on original and resampled datasets. Performance was evaluated using the area under the receiver operating characteristic curve (AUROC) and F1 score.
Results: The cohort included 24 518 treatment episodes (n=17 512, VTE events=52). After handling class imbalance, RF and XGBoost were the best-performing models (F1: 84.3% and 81.6%; AUROC: 59.6% and 69.2%, respectively), compared with other models (F1: 2.6%–55.3%, AUROC: 54.2%–68.4%). Key predictors included age, glucocorticoid use and rheumatic, renal and thyroid diseases.
Discussion: ML shows potential for VTE risk detection despite low event rate and class imbalance. However, findings suggest synthetic sample overfitting, and external validation is warranted before clinical deployment. Identified predictors reinforce the role of metabolic and inflammatory factors in VTE susceptibility among COCP users.
Conclusion: ML models, particularly XGBoost and RF, demonstrated moderate predictive ability for VTE among COCP users, supporting their use in individualised risk assessments, though external validation is warranted.
Related
- Predicting women’s intention to use contraceptives in East Africa: a machine learning analysis of predictors
- Signal of venous thromboembolism with ethinyl estradiol vs natural estrogens in oral contraceptives: analysis of the EudraVigilance and Food and Drug Administration Adverse Events Reporting System databases
- COMPREHENSIVE LIPIDOMIC PROFILING COMBINED WITH MACHINE LEARNING REVEALS BASELINE SIGNATURES PREDICTIVE OF FIRST LEG ULCER DEVELOPMENT IN SICKLE CELL DISEASE
- Basic leucine zipper ATF-like transcription factor (BATF) as a candidate marker associated with venous thromboembolism risk factors in sepsis: an integrated bioinformatics and machine learning analysis
- Beyond Padua and IMPROVE: Machine Learning Outperforms Guideline Risk Scores for Prediction of Radiologically Confirmed Hospital-Acquired Venous Thromboembolism
BibTeX
@article{alfakhri2026development,
title = {{Development of a machine learning model to predict venous thromboembolism risk in women using combined oral contraceptives}},
author = {Almaha Alfakhri and Ohoud Almadani and Hanan Alsehli and Ibrahim Asiri},
journal = {BMJ Health \& Care Informatics},
year = {2026},
month = oct,
doi = {10.1136/bmjhci-2025-102020},
url = {https://doi.org/10.1136/bmjhci-2025-102020},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Development of a machine learning model to predict venous thromboembolism risk in women using combined oral contraceptives
AU - Almaha Alfakhri
AU - Ohoud Almadani
AU - Hanan Alsehli
AU - Ibrahim Asiri
PY - 2026
DA - 2026/10/28
JO - BMJ Health & Care Informatics
DO - 10.1136/bmjhci-2025-102020
UR - https://doi.org/10.1136/bmjhci-2025-102020
AB - Objective: Venous thromboembolism (VTE) is a major cause of morbidity and mortality, with combined oral contraceptives (COCPs) being a modifiable risk factor. Existing clinical tools have limited ability to capture complex predictor interactions. This study aimed to develop and evaluate machine learning (ML) models to predict VTE risk among COCP users. Methods: This study was conducted using data from the Saudi Real-World Evidence Network (2016–2023). The primary outcome was the first VTE event within 2 years of COCP initiation. Following data partitioning, feature selection was performed on the training set. Elastic-net logistic regression, random forest (RF), extreme gradient boosting (XGBoost), one-class support vector machine and Isolation-Forest were trained on original and resampled datasets. Performance was evaluated using the area under the receiver operating characteristic curve (AUROC) and F1 score. Results: The cohort included 24 518 treatment episodes (n=17 512, VTE events=52). After handling class imbalance, RF and XGBoost were the best-performing models (F1: 84.3% and 81.6%; AUROC: 59.6% and 69.2%, respectively), compared with other models (F1: 2.6%–55.3%, AUROC: 54.2%–68.4%). Key predictors included age, glucocorticoid use and rheumatic, renal and thyroid diseases. Discussion: ML shows potential for VTE risk detection despite low event rate and class imbalance. However, findings suggest synthetic sample overfitting, and external validation is warranted before clinical deployment. Identified predictors reinforce the role of metabolic and inflammatory factors in VTE susceptibility among COCP users. Conclusion: ML models, particularly XGBoost and RF, demonstrated moderate predictive ability for VTE among COCP users, supporting their use in individualised risk assessments, though external validation is warranted.
SN - 2632-1009
ER -