Figure 2
From: The necessity of incorporating non-genetic risk factors into polygenic risk score models
Comparison of models including only PGS or additional variables that can be attained through a questionnaire. Risk predicted with questionnaire-based variables performs similar or better at identifying individuals at high risk (10th decile), compared to PGS. Adding PGS to a questionnaire-based model can, however, further improve the identification of high-risk individuals, but requires a large dataset to be detectable due to its limited effect. For a comparison of risk in different risk strata at different ages, we refer to Supplementary Fig. 2. (A) Absolute incidence and prevalence per decile based on PGS alone or combine with additional variables. Performance improves if additional variables are added beyond PGS alone. Risk is increasing exponentially in higher risk categories. (B) Odds and incidence ratios of individuals in the top decile according to different models. Model including questionnaire-based risk variables performs significantly better at identifying individuals that will get the respective outcome than a model based on PGS alone. (C) C-indexes of the different models. Added value of PGS on top of variables that can be derived from a questionnaire is limited. PGS Polygenic risk score, BMI Body mass index, C-index Harrell’s C-index, PA Physical activity (based on number of days moderate and days of vigorous activity), Parent Parental T2D status, Variables not included as predictors in the model were included as covariates. Additionally, in the UK Biobank, data the first 4 PCs and genotyping batch were included as covariate. Bars indicate 95% confidence interval. For numerical representation we refer to the Supplementary Table 1.
