Software for Survival Analysis of Studies Nested within Cohorts To Estimate Relative, Absolute, and Attributable Risks

• Activity Number: 501
• Type: Contributed
• Date/Time: Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Hormuzd Katki*+ and Steven D. Mark
• Companies: National Cancer Institute and University of Colorado Health Sciences Center
• Keywords: case-cohort ; nested case-control ; two-stage design ; two-phase design ; validation study ; epidemiology

Abstract

In studies nested within cohorts, censored survival outcomes and easy to obtain covariates are observed on everyone, but hard to obtain covariates are observed only on a subsample. Examples include two-stage, case-cohort, and nested case-control designs. Current methodology in these studies focuses on relative risks, but absolute and attributable risks are needed to assess clinical and public health impact. Our software estimates all three risks (standardized for confounders) using Cox models or Kaplan-Meier in studies nested within cohorts. Subsampling must be MAR, so can stratify on any variable observed on everyone, can subsample cases, and can frequency-match. We estimate efficient weights that could exploit surrogates for exposure. Our R software (based on Mark & Katki, JASA, in press) is freely available. We analyze a study of cancer and zinc where only 25% of cohort had zinc measured.