Fig. 4: Assessing the importance of various disease risk factors using SHAP value from FCNN.
From: UKB-MDRMF: a multi-disease risk and multimorbidity framework based on UK biobank data
a Normalized proportion of the top 30 significant risk factors among six categories of independent variables for each of the 21 disease types. b Frequencies for the top 5, top 10, and top 20 important variables in each of the 21 types of diseases. c Distribution of variable importance for all Phecodes, with colors ranging from blue (negative effect) to red (positive effect). d Average importance values for each variable category and disease type, with blue indicating negative effects and red indicating positive effects. e Comparison of risk factor importance for disease prediction (left, from FCNN) and risk assessment (right, from DeepSurv). Diseases were aggregated into nine major types derived from the 21 disease types. Thicker lines indicate greater importance. Icons provided by Icons8 (https://icons8.com).
