Nonparametric Estimation of the Survival Function for Prevalent and Incident Cases Under Stationary Incidence

• Activity Number: 415
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Pierre-Jerome Bergeron*+ and Masoud Asgharian
• Companies: University of Ottawa and McGill University
• Keywords: nonparametric ; length-biased sampling ; right-censoring ; survival analysis ; EM algorithm

Abstract

In epidemiological studies, subject who experience disease onset after recruitment are called incident cases, and subjects whose onset occured prior to the study are prevalent cases. The latter tend to survive longer than the former due to sampling bias. The Kaplan-Meier (KM) estimator is the best known estimator of the survival function with right-censoring, and a simple modification allows for estimation with general left-truncation, conditionally adjusting for the bias. However, when dealing with prevalent cases under stationary incidence, the data are length-biased. The nonparametric MLE of the survival function that corrects for length-bias is obtained through an EM algorithm which provides efficient estimation of the survival from onset that accounts for dependent censoring and known truncation distribution. The approach is completely different than that of the KM estimator. In large longitudinal studies, one may observe both incident and prevalent cases, and there is no simple way to combine both estimators to use all the data. We present a product-limit approach, obtain its asymptotic properties and illustrate the method with data from Canadian Study of Health and Aging.