Figure 4

Selection of pathway-specific PRSs using the Least Absolute Shrinkage and Selection Operator (LASSO) Model. Eight clinical and 20 PRS features with p < 0.05 from the univariate CoxPH regression were selected for 3-year mortality. Fit the Regularized (LASSO) Cox Model in the training dataset with fivefold cross-validation for the regression coefficients of PRS and nongenetic variables such as age at index stroke. (A) X-tile analysis of the features associated with 3-year mortality with p value < 0.05, respectively. Y-axis represents partial log likelihood (LL) deviance from a fivefold cross-validation, Error bar indicate 95% CIs. The left vertical line in (A) showed where the CV-error curve hits its minimum. The right vertical line in (A) shows the most regularized model with CV-error within 1 standard deviation of the minimum. We extract such optimal λ’s. (B) Regularization path for the progressively shrinking of the regression coefficients of variables by tuning the λ in the LASSO method with fivefold CV. Variables with bigger absolute regression coefficients were listed. The top ten features with the larger effect size were labeled.