Predicting lifetime cervical cancer screening among sexually active women in low- and middle-income countries using machine learning models: evidence from Demographic and Health Surveys

Résumé fourni par la source

INTRODUCTION: Cervical cancer poses a major health risk in low- and middle-income countries (LMICs), with high morbidity and mortality among women. It is largely preventable through screening and early treatment, but underused programs lead to a heavy burden. The goal is to identify key factors influencing screening, reveal inequities, and improve policies in resource-limited areas. METHODS: Using recent Demographic and Health Surveys (DHS) data from 19 LMICs, this study estimated lifetime cervical cancer screening among women aged 15 to 49 years. Missing data were removed, and class balancing was performed with synthetic minority oversampling technique. GridSearch optimized hyperparameters before dividing data into training and validation sets. Multiple machine learning algorithms predicted lifetime screening, including decision trees, random forests, gradient boosting, extra trees, XGBoost, LightGBM, and catBoost classifiers. Model performance was evaluated using accuracy, recall, precision, F1 score, and receiver operating characteristics area under the curve. RESULTS: The analysis of 261,371 women aged 15 to 49 from the DHS found that random forest achieved 80% accuracy, 87% precision, 80% recall, and an 80% F1 score, while extra trees achieved 79%, 87%, 79%, and 83%. Key predictors included household size, wealth, women's age, and education. Ensemble methods such as random forests and extra trees demonstrated strong generalization, making them useful for targeted interventions. CONCLUSION: Ensemble algorithms such as random forests and extra trees predict lifetime cervical cancer screening with ∼80% accuracy in LMICS. The study emphasizes targeted awareness, education, and age-sensitive screening. Findings will be tested in other populations to support clinical decision systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.