ABSTRACT
This chapter discusses estimating absolute risk for both approaches. For the cause-specific hazard models, it also discusses estimation from sub-samples of a cohort, including nested case-control and case-cohort studies, and from cohorts obtained by complex sampling from a general population. The chapter shows how to combine estimates of relative and attributable risk from observational studies with overall population incidence rates of the primary and competing events to estimate absolute risk. It presents some methods for estimating absolute risks, with examples and describes variance calculations and confidence interval construction. The case-cohort design, introduced for time-to-event data by Prentice and further studied by Self and Prentice, requires collecting covariate data for all cases and for a random sample of the entire cohort (the "subcohort"), which can include cases. The size of the subcohort, n, is usually much smaller than the size of the cohort, n.
