194 – Contributed Oral Poster Presentations: ENAR
Risk Factors and Outcomes in a Multiple Tumor Marker Setting: The Issues of Correlated and Missing Tumor Markers
Kathryn C. Fitzgerald
Harvard TH Chan School of Public Health
Robert J. Glynn
Harvard TH Chan School of Public Health
Rulla M. Tamimi
Harvard TH Chan School of Public Health
Wendy Y. Chen
Brigham and Women's Hospital and Harvard Medical School
Graham C. Colditz
Harvard TH Chan School of Public Health
Susan Hankinson
Harvard TH Chan School of Public Health
Bernard Rosner
Brigham and Women's Hospital and Harvard Medical School
Risk profiles for cancer outcomes often vary by the presence of different tumor markers or subtypes. Increasing the number of markers adds an additional layer of complexity as markers are often correlated, and as a result, leads to difficulty in assessing subtype-specific effects of particular risk factors without considering other markers. Scientifically, it's also of interest to identify which marker or combination of markers is most relevant for disease. Finally, with a larger number of markers, the likelihood of missing marker information also increases and raises the question as to how to properly address this issue. In this paper, we apply methodology originally introduced by Rosner et al. to compute adjusted hazard-ratios that account for multiple correlated markers while evaluating four candidate approaches for missing tumor subtype. We consider the complete case, missing indicator, inverse probability weighting and multiple imputation approaches for missing tumor markers. We evaluate these four approaches using simulation studies and apply each to a real study of breast cancer risk factors considering multiple subtypes.