The investigators evaluated two complementary strategies. The first relied on the aggregate FINDRISC score as a single predictor, analyzed with conventional logistic regression. The second used the questionnaire’s individual items as features for four machine learning algorithms: decision tree, random forest, support vector machine, and artificial neural network. Model discrimination was assessed with the AUC-ROC, and calibration was examined using decile-based calibration plots and the Hosmer–Lemeshow test. To address class imbalance, the researchers applied SMOTE strictly within training folds during cross-validation.

The aggregate FINDRISC score achieved an AUC of 0.76, with a 95 percent confidence interval of 0.74 to 0.79, which the authors describe as acceptable discrimination for predicting six-year incident type 2 diabetes. When the questionnaire’s components were modeled individually using continuous anthropometric features, random forest achieved the highest observed discriminative performance, with an AUC of 0.81 and a 95 percent confidence interval of 0.79 to 0.83.

The study also examined screening thresholds for the FINDRISC score. A threshold greater than 12 yielded the highest Youden index, at 0.41, with sensitivity of 73.0 percent and specificity of 68.4 percent. A lower threshold, greater than 11, offered higher sensitivity of 78.9 percent but lower specificity of 61.1 percent.

The authors conclude that although the component-level machine learning models yielded higher observed AUCs in internal validation, the clinical significance of these modest differences remains uncertain. They note that further assessment of calibration, clinical utility, and external validation is needed, and that the standard FINDRISC therefore remains the practical tool for primary care screening. As a prospective population-based cohort analysis with internal validation, the study’s findings apply to the population studied, and the reported performance figures carry the uncertainty indicated by their confidence intervals.

Subject of Research: Comparative prediction of six-year type 2 diabetes incidence using the FINDRISC score and machine learning algorithms in a population-based cohort

Article Title: Predicting the incidence of type 2 diabetes using FINDRISC and machine learning approaches: evidence from a population-based cohort

Article References: Hamedi, A., Seif, M., Sharifi, M. H., Rezaianzadeh, A., Lashkardoost, H., & Hassanzadeh, J. (2026). Predicting the incidence of type 2 diabetes using FINDRISC and machine learning approaches: evidence from a population-based cohort. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02683-3

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02683-3

Keywords: type 2 diabetes, FINDRISC, machine learning, random forest, risk prediction, population-based cohort, screening, AUC-ROC, logistic regression, incidence, Predicting, type