Online Program

Interval-Censored Survival Analysis to Reduce Detection Bias in a Study of Family History, Race, and Cancer Risk (307805)

Keywords: detection bias, prostate cancer, interval censoring, differential screening

Summarizing associations between risk factors and disease incidence is an important step towards the targeted detection of disease. Relative risks estimated from patterns of disease diagnosis over a follow-up period in prospectively screened cohorts are common summaries of these associations. Such estimates can be biased when screening patterns in a cohort differ due to risk factors of interest. In order to mitigate this detection bias, we propose to model the natural history of patients, using a generalization of interval censored survival models that allow for imperfect screening tests. Tangen et al. (JCO 2016) have recently shown that relative risks estimated using data from the SELECT prostate cancer trial, where screening was left to community standards, were biased relative to the PCPT prostate cancer trial, where the men were screened annually with a specified criterion and an end of study biopsy. Results from our proposed methodology suggest that the relative risk associated with positive family history, but not black race, were subject to detection bias in SELECT, leading us to conclude that population differences played a role in the divergent results of the two studies.