Predicting the Absolute Risk of Undetected Uterine Cancer in a Matched Case-Control Study

• Activity Number: 418 - SPEED: Biostatistical Methods, Application, and Education, Part 2
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 2:00 PM to 2:45 PM
• Sponsor: Section on Risk Analysis
• Author(s): Catherine Lee* and Scott E. Lentz and Eve Zaritsky and Lue-Yen Tucker and Tina Raine-Bennett
• Companies: Kaiser Permanente Division of Research and , The Southern California Permanente Medical Group, Los Angeles and The Permanente Medical Group, Oakland California and The Division of Research, Kaiser Permanente Northern California and Oakland California and The Division of Research, Kaiser Permanente Northern California
• Keywords: matched case control; absolute risk; prediction; weighted logistic regression

Abstract

For women undergoing hysterectomy, the ability to identify hidden cancers ahead of surgery would guide the method of surgery that minimizes the possible spread of cancer. To this end, we aimed to estimate the absolute risk of hidden cancer based on preoperative clinical features using retrospective data from a previous matched case-control study. Whereas methods for estimation of absolute risk in standard case-control studies are well-known, such methods for matched case-control studies in the literature are relatively unknown and uncited. One such method, proposed by Rose, van der Laan (2008), uses weighted logistic regression, where the weights are based on the outcome prevalence in the population and within matching strata. We applied this method to our study data using prevalence estimates from a prior study. The method was easy to implement using standard statistical software, performed well in simulations based on our study sample, and the resulting risk prediction model had good discrimination in our dataset. Although further work is needed to validate the model in an independent study population, it may ultimately aid in guiding method of surgery based on predicted risk.