EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Development and Validation of Machine Learning Models for Prediction of Type 2 Diabetes Mellitus in Asymptomatic Healthy Adults in Bangladesh

Anindita Das Barshan, Md Mahafuj Hossain

Asia Pacific Journal of Medical Innovations · 31.03.2026 · 3 (1) · 14-28

DOI: 10.70818/apjmi.v03i01.042 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Asia Pacific Journal of Medical Innovations, online erschienen am 31.03.2026. Verfasst von Anindita Das Barshan und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Type 2 diabetes mellitus (T2DM) represents a major public health challenge in Bangladesh, here a substantial proportion of cases remain undiagnosed at time of clinical presentation. Machine learning (ML) offers enhanced discriminative capacity over conventional screening tools; however, validated ML prediction models contextualised to the Bangladeshi adult population are lacking.

Methods

A cross-sectional study enrolled 743 asymptomatic healthy adults (January 2025 to December 2025). Seven ML algorithms and a Soft Voting Ensemble were developed using 23 routinely collectable clinical variables. The dataset was stratified into 70% training (n = 520) and 30% independent holdout test (n = 223) sets. Nested cross-validation with within-fold SMOTE was employed. Post-training Platt Scaling was applied to calibrate probabilities. Model performance was evaluated by AUROC, sensitivity, specificity, PPV, NPV, F1-score, and MCC. SHAP values, decision curve analysis, and three-tier risk stratification were reported.

Results

The newly diagnosed DM prevalence was 17.1% (n = 127/743). XGBoost was the best-performing individual model: AUC = 0.918 (95% CI: 0.882–0.954), sensitivity = 84.2%, specificity = 88.6%, NPV = 96.5%, Brier score = 0.081, and Hosmer–Lemeshow p = 0.614. XGBoost demonstrated superior net reclassification improvement (NRI = 0.328; p < 0.001) and IDI = 0.138 (p < 0.001) over logistic regression. Waist-to-hip ratio, BMI, and age were the three most influential predictors by SHAP analysis. Risk stratification allocated 51.3% of participants to a low-risk tier (DM prevalence = 3.4%) and 23.7% to a high-risk tier (prevalence = 42.1%).

Conclusion

A well-calibrated XGBoost model using 23 routine variables demonstrates high discriminative accuracy and clinical utility for T2DM screening in asymptomatic Bangladeshi adults, with a practical three-tier risk stratification pathway suitable for integration into primary care.

Zitieren

APA

Barshan, A. D., & Hossain, M. M. (2026). Development and Validation of Machine Learning Models for Prediction of Type 2 Diabetes Mellitus in Asymptomatic Healthy Adults in Bangladesh. Asia Pacific Journal of Medical Innovations, 3(1), 14-28. https://doi.org/10.70818/apjmi.v03i01.042

Vancouver

Barshan AD, Hossain MM. Development and Validation of Machine Learning Models for Prediction of Type 2 Diabetes Mellitus in Asymptomatic Healthy Adults in Bangladesh. Asia Pacific Journal of Medical Innovations. 2026;3(1):14-28. doi:10.70818/apjmi.v03i01.042

BibTeX
@article{Barshan2026,
  author = {Barshan, Anindita Das and Hossain, Md Mahafuj},
  title = {Development and Validation of Machine Learning Models for Prediction of Type 2 Diabetes Mellitus in Asymptomatic Healthy Adults in Bangladesh},
  journal = {Asia Pacific Journal of Medical Innovations},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {14--28},
  doi = {10.70818/apjmi.v03i01.042}
}
Kostenlos starten →