Figure 3

Comparative effect sizes for selected significant features differentiating top 10% or bottom 10% subgroup cases compared to all other cases, respectively, for each disease. CAD coronary artery disease, BC breast cancer, SZ schizophrenia, PRS polygenic risk score, LDL low density lipoprotein, Sibling Hx sibling with a history of heart disease, BMI body mass index, SBP systolic blood pressure, HDL high density lipoprotein, Sibling Hx. of Dep. sibling with a history of depression, Hx history. Note that CAD_6M, CAD_1.7M, CAD_46k, and CAD_202 are polygenic risk scores for heart disease and BC_313, BC_5k, and BC_77 are polygenic risk scores for breast cancer. Note: The x-axis label, Cohen’s value, represents effect size measured by Cohen’s d for numerical features and Cohen’s h for binary features. (a) For CAD, the absolute value effect sizes across all non-genetic significant features for top 10% (Left) and bottom 10% (Right) case comparisons range from ~ 0.15 up to ~ 0.5, representing overall low-to-moderate effect sizes. The top 10% comparison displays the effect sizes for comparisons between the top 10% subgroup cases and all other cases in the population (i.e., the remaining 90% that benefit less from PRSs). The bottom 10% subgroup comparison visualizes the effect sizes for the features that significantly differentiate the bottom 10% subgroup cases from all other cases in the population (i.e., the remaining 90% that benefit more from PRSs, a superset of the top 10% subgroup). (b) For BC, the absolute value effect sizes across all non-genetic significant features are low, ranging from 0.23 to 0.24. The comparisons are shown for the top 10% (Left) and bottom 10% (Right) against the respective rest of the population. (c) For SZ, the absolute value effect sizes across all significant features for cases in the top (Left) and bottom 10% (Right) case comparisons range ~ 0.3 up to ~ 0.8, representing overall moderate-to-high effect sizes.