Linear Rank Tests for Survival Outcomes in Complex Sample Survey Data

• Activity Number: 560
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Stuart Lipsitz*+ and Debajyoti Sinha and Kevin Rader and Sundar Natarajan
• Companies: Harvard Medical School and Florida State University and Harvard University and New York University
• Keywords: logrank test ; partial likelihood ; score test ; Wilcoxon test

Abstract

Linear rank tests (logrank, Wilcoxon, Peto-Peto, etc) are commonly used to compare survival outcomes across two groups. Unfortunately, no simple extension of the linear rank tests have been proposed for complex sample survey data. With a random sample of independent subjects, any of the linear rank test statistics can be shown to equal a Cox partial likelihood score test, where the particular linear rank test is determined by a weight given to each risk set in the partial likelihood. For example, a weight of 1 gives the log-rank test, and a weight equal to the number at risk gives the Wilcoxon test. For complex survey data, we formulate our extension of the linear rank tests as an estimating equations score statistic for no group effect in the partial likelihood. The proposed method is applied to an NHANES study which explores if a dash diet (versus a normal diet) reduces the risk of death.