Machine Learning-Based Prediction of RBC Recovery in Type 2 Diabetes

Recovery time, defined as the duration required for a red blood cell (RBC) membrane to return to its original shape following deformation, represents an emergent biophysical property with clinical significance in type 2 diabetes. This study aimed to investigate obesity-associated determinants of RBC recovery by integrating micro-electro-fluidic manipulation with conventional statistical methods and machine learning. We analyzed six predictors of RBC recovery in type 2 diabetics based on obesity status (obese vs. non-obese): strain, stretch factor, glycated hemoglobin (HbA1c), serum creatinine (SCr), age, and body mass index (BMI) values. Using standard statistical analysis, we found that SCr was the most significant predictor of recovery time in the obese group (R² = 0.24, r = 0.49), whereas stretch factor and age were the strongest in the non-obese group (R² = 0.535 and 0.503; r = -0.73 and -0.71, respectively). To investigate multivariate contributions, we applied Random Forest Regression (RFR) method, which revealed a distinct shift in predictor importance between groups. In obese individuals, stretch factor (importance = 0.31) and HbA1c (0.25) were the most dominant, while strain had negligible importance (0.02). In contrast, recovery time in non-obese individuals was primarily influenced by age (0.22), strain (0.18), and stretch factor (0.15), with HbA1c contributing minimally (0.02). Together, these findings highlight the importance of combining statistical and machine learning approaches to identify phenotype-specific biochemical, biomechanical, and demographic determinants of RBC recovery in type 2 diabetes.