AI-Based Estimation and Segmentation of Biological Age Using Clinical Data

Background: Biological age (BA) has gained increasing attention as a more accurate representation of an individual's physiological condition than chronological age (CA). Unlike CA, BA may better reflect the cumulative impact of genetics, lifestyle, and environmental factors on aging. However, many existing models rely on complex or inaccessible biomarkers, limiting their practical use in routine care. Methods: This retrospective cohort study aimed to estimate BA using routine clinical parameters and to identify aging patterns via unsupervised clustering. Data from 412 adults aged 18–64 were extracted from electronic health records. An Explainable Boosting Machine (EBM) model was employed to predict BA based on 15 selected features, including anthropometric, biochemical, and hematological variables. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R². The ΔAge (BA–CA) metric was used for clustering, and biomarker contributions were assessed using SHAP-style analysis. Results: The model achieved an MAE of 7.07 years, RMSE of 8.52 years, and R² of 0.57. Three distinct clusters were identified based on ΔAge. Despite being the oldest, Cluster 3 showed the lowest biological age, suggesting a favorable aging profile. Key predictors of BA included height, waist circumference, glycated hemoglobin (HbA1c), fasting blood glucose, Total cholesterol, and ferritin. Higher HbA1c and glucose levels increased predicted age, while high HDL levels showed a protective effect. ALT and RDW-SD demonstrated significant impact only at extreme values. Conclusion: The study demonstrates that BA can be effectively estimated using accessible clinical data. The use of interpretable artificial intelligence models such as EBM allows for both accurate prediction and meaningful insight into the physiological factors driving aging. This approach may support personalized health assessments and more targeted preventive strategies.