FIGURE 3:

Variables characterizing DNAJB11-PKD patients. LASSO for variable selection via logistic regression for discriminating between ADPKD and DNAJB11-PKD. The plot shows standardized coefficient estimates (y-axis) as a function of the ‘L1-norm’ of the standardized coefficients (i.e. the maximum allowed sum of the absolute values of the coefficients) (x-axis). The tuning parameter (not shown in the plot) ‘shrinks’ the coefficient toward zero as its value gets larger. By setting some coefficient to zero, the tuning parameter determines which variables the LASSO will eventually exclude. The final value of the L1-norm (vertical dash line) resulted from the selection among the potential candidates of the tuning parameters that were estimated by cross-validation (see text) of the one value that minimized the Bayesian information criterion. Age at last follow-up, valvular defects (negative association with DNAJB11-PKD), kidney stones and type 2 diabetes were the variables with the largest standardized coefficients and that were eventually selected by LASSO.