ABSTRACT
Absolute risk estimates based on cause-specific models rely on survival analysis methods that were first widely applied for estimating pure risks following disease diagnosis. In Chapter 4 we discussed how standard inference for the Cox proportional hazards model (Cox, 1972) could be used to estimate relative risks for cause-specific hazards as well as cumulative and instantaneous contributions to the cause-specific hazards. Thus, a vast literature on survival regression modeling also applies to modeling cause-specific hazards and absolute risk estimation based on cause-specific hazard modeling. Excellent discussions of topics such as how to code covariates, flexible representations of dose-response for quantitative covariates (e.g., splines), handling missing values and model checking by examining residuals and other tests of goodness-of-fit, are found in classic books on survival and risk modeling, including Andersen et al. (1993), Harrell (2001), Kalbfleisch and Prentice (2002), Steyerberg (2009), Therneau and Grambsch (2000), and van Houwelingen and Putter (2012). We do not attempt to discuss these topics comprehensively, but we briefly touch on three important issues for modeling absolute risk: covariate selection, missing covariate data, and updating previously well-established risk models by adding new covariates.
