Individualized treatment decisions for prostate cancer using prostate cancer specific and other-cause mortality predictors
Although prostate cancer is the most common cancer in American men, the majority of men with prostate cancer do not die of their cancer: instead, they die of something else. As a result, treatment guidelines encourage clinicians to consider comorbidity-adjusted life expectancy when making prostate cancer treatment decisions. However, research suggests that clinicians struggle to estimate comorbidity-adjusted life expectancy and may overtreat younger men with substantial comorbidity burden. We want to build a prostate cancer treatment decision aid that combines PCSM risk predictors (e.g. Gleason, PSA) and OCM risk predictors (e.g. cholesterol,
smoking, comorbidities) to estimate absolute risk of PCSM, OCM, and all-cause mortality in American men with prostate cancer following a particular treatment regimen. We want to validate this risk prediction model in the PLCO prostate cancer data.
Accessible data sources with information on PCSM and OCM predictors in American men with prostate cancer are in short supply. To estimate the OCM component of the model, we used data from the National Health and Nutrition Examination Survey. To estimate the PCSM model, we used a model developed by collaborators of ours (currently in preparation); however, we plan to consider multiple PCSM models and give clinicians the option of using their preferred PCSM model. To estimate the effects of particular treatments, we pulled hazard ratios from published randomized clinical trials, although again, we will allow clinicians to input their own hazard ratios if desired.
Because of this patchwork model-building approach, we want to validate our model in data that were collected on a single population of men with prostate cancer, to ensure that the model makes sound recommendations. Our ideal validation data will have the following variables on American prostate cancer patients around the time of diagnosis: age, race, education, marital status, smoking status, BMI, comorbidities, cancer stage, T stage, Gleason score, PSA level at diagnosis, initial treatment regimen, and mortality follow-up. From study documentation, we believe that the PLCO prostate cancer data meet these criteria.
If our model validates satisfactorily in the PLCO data, we will publish our findings and create a user-friendly risk prediction app for clinical use.
1. To build a prostate cancer treatment decision aid that estimates absolute risk of death using a mixture of prostate cancer specific and general comorbidity predictors in American men with prostate cancer.
2. To validate this treatment rule using the PLCO prostate cancer data.
3. To share this treatment rule with clinicians via a user-friendly prediction app.
Elizabeth Chase, MS, University of Michigan
Daniel Spratt, MD, University of Michigan