Identification of Significant Prognostic Factors Among a Large Number of Potential Markers: Applied to Genetic Abnormalities in Head and Neck Squamous Cell Carcinoma (HNSCC)

• Activity Number: 215
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 12:00 PM to 1:50 PM
• Sponsor: Biometrics Section*
• Author(s): Weiji Shi*+ and Bhuvanesh Singh and Howard Thaler
• Affiliation(s): Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center
• Keywords: multiple testing ; cytogenetics ; Monte Carlo simulation ; survival analysis ; permutation test

Abstract

Technology can provide high-dimensional characterizations of biological material. In cytogenetics, comparative genomic hybridization (CGH) profiles assay hundreds of possible chromosomal abnormalities. Evaluating statistical significance of tumor CGH profiles as predictors of disease-specific survival (DSS) in cancer patients presents major multiplicity of testing. Univariate significance tests for all abnormalities could yield very small nominal p-values by chance alone. In 82 patients with HNSCC, we first reduced the CGH profile from 303 to 199 abnormalities, eliminating rare or redundant ones. 35 log-rank statistics were nominally significant at p< 0.05 by chi-square approximation. From 10,000 Monte Carlo permutations of DSS across patients (maintaining co-occurrence patterns of the multiple chromosomal abnormalities), we estimated distributions of the largest log-rank statistics under generic H0: No relation between cytogenetics and prognosis. We then compared the largest original sample statistics with estimated order statistic distributions and applied multiple comparison procedures. Four abnormalities were jointly significant predictors of DSS in stepwise Cox regression.