Landmark Prediction of Survival

• Activity Number: 184
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Layla Parast*+ and Tianxi Cai
• Companies: Harvard University and Harvard School of Public Health
• Keywords: survival analysis ; biomarkers ; ROC curves

Abstract

Although new biological and genetic markers promise better disease prognosis, the accuracy in identifying short term and long term survivors remains unsatisfactory for most complex diseases. It has often been argued that short term clinical outcomes may have potential in predicting long term outcomes. We propose to develop and evaluate conditional prognostic rules for the prediction of long term outcomes based on baseline marker information along with short term outcome status at an earlier landmark time. When there are multiple markers available, we construct a robust composite score by fitting a proportional hazards working model for the conditional survival distribution. The accuracy of the score was evaluated non-parametrically based on inverse probability weighting. Resampling procedures were proposed to derive estimation procedures for the accuracy measures.