Individualized estimates of treatment benefit from hormone therapy for men with prostate cancer

Abstract National treatment guidelines, including those from the National Comprehensive Cancer Network (NCCN), for men with localized prostate cancer are based on prostate cancer specific mortality (PCSM) risk and life expectancy which is driven by other cause (non-cancer) mortality (OCM). While OCM is the dominant risk for

many prostate cancer patients, it has received limited attention in the literature. We developed and validated a clinically relevant model for OCM and life expectancy using prospective data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) trial. OCM varied substantially based on patient

demographics and comorbidities and found that social security life expectancy estimations based on age alone generated substantially biased estimates. Our team also developed and validated a clinical prognostic model for PCSM published in JAMA Oncology in 2020 (STAR-CAP). While the importance of competing risks in

prostate cancer is widely recognized, there are no models to date which provide PCSM risk estimates individualized by OCM risk. Many of the common and important treatment decisions for men with localized PC include whether to treat with hormone therapy (ADT) and for how long. Guidelines recommend ADT for many

men who are treated without it due to morbidity and questions as to actual treatment benefit for individual patients. Phase III trials have provided overall estimates of the benefit of ADT in terms of reduction in PCSM over large groups of patients (e.g. intermediate risk). To date however, there are no tools which provide

treatment benefit estimates for individual patients based on PCSM and OCM risk. We propose to address these shortcomings through two aims. In Aim 1 we will integrate our model for OCM risk with models for distant metastases (DM) and PCSM to provide estimates of the absolute risk of DM and PCSM that are

personalized using both PCSM and OCM risk factors. We hypothesize that the integrated model estimates will differ from clinical ad hoc estimates and significantly improve upon current estimates of PCSM that do not consider OCM risk. In Aim 2 we will develop a web app which provides individualized estimates of ADT

treatment effect in three common clinical scenarios: RT vs RT + short term ADT in intermediate risk, RT + short term ADT vs RT + long term ADT in high risk and RT vs RT + ADT in the salvage setting. To do this we will combine the integrated multistate model from Aim 1 with published hazard ratios for outcomes

of PCSM and DM. Our group is uniquely positioned to carry out this work as we have already developed the component models, have access to the patient level data for additional analyses and have a track record of successful development of web apps to aid in clinical decision making. If successful, the tools developed in this

application will provide clinicians and patients the key information they need to make informed treatment decisions.