Additional file 2 of Epigenetic prediction of complex traits and mortality in a cohort of individuals with oropharyngeal cancer

Additional file 2: Supplementary Figure 1a: Kaplan-Meier survival curves based on demographic and clinical covariates. Comorbidity categories were defined according to the severity or organ decompensation: none (coded 0), mild (coded 1), moderate (coded 2), or severe (coded 3). See text for more details. Supplementary Figure 1b: Kaplan-Meier survival curves based on our phenotypes of interest. Supplementary Figure 2: The association between variance explained by DNAm score and hazard ratio for 4-year mortality (Model 3). Hazard ratios are plotted as absolute log-transformed values for comparability. Supplementary Figure 3: ROC curves detailing the predictive accuracy of DNAm risk scores, self-reported phenotype and a combination of the two, against ~4-year mortality (median 3.9 years) in HN5000. ROC curves are provided for smoking, alcohol consumption, BMI and educational attainment. DNAm AUCs reflect use of the DNAm scores for these phenotypes which explained the greatest phenotypic variance: smoking = Trejo Bayesian model, alcohol consumption = Liu et al. model 4, BMI = Trejo Bayesian model, educational attainment = McCartney LASSO model. Abbreviations: AUC, area under curve; DNAm, DNA methylation; ROC, receiver-operator curve