Fig. 5: External validation of IMPACT framework on the UKB dataset.
From: Interpretable machine learning prediction of all-cause mortality
a Relative importance of 51 overlapping features in the 5-year mortality prediction models trained on the NHANES (151 features) and UKB (51 features) datasets. For each model, the figure shows the 20 most important features of prediction (ordered by importance). The purple line indicates that the feature is in the top 20 features of both models. Blue and red lines indicate the feature is in the top 20 features of one model but not the other. The p-value of the Fisher's exact test examines the overlap between the top 20 most important overlapping features in the NHANES and UKB models (the contingency table in Supplementary Figure 6F). The Spearman's correlation coefficient is calculated using the feature importance of the overlapping features in NHANES and UKB (n = 51 featurs). ***p-value < 0.001. b–d The main effect of red cell distribution width, urine albumin and serum uric acid on 5-year mortality in the model trained on UKB (51 features) dataset. e–g The relative 5-year mortality risk of gamma glutamyl transferase, lymphocyte percent, and serum albumin in the model trained on the UKB (51 features) dataset.
