Using Regularized Linear Regression to Identify Drivers of Diabetes Progression

Objective: Determine the crucial patient factors that predict diabetes progression in a real-world data set (Efron, 2004). Approach: We applied standard and regularized linear regression (lasso) to model disease progression using 10 explanatory variables (age, sex, bmi, average blood pressure, and 6 blood sera measurements). Key insight: Standard regression leads to overfitting and generalizes poorly to new data. Regularized regression inherently favors “simpler” models that successfully predict diabetes progression using fewer explanatory variables. Results: Five patient factors (including sex, bmi, blood pressure, s3 and s5) emerged as statistically significant drivers of diabetes disease progression.