Comparative Analysis of Different Methods of Disease Progression Data Capture in Patients With Metastatic Castration-Resistant Prostate Cancer

OBJECTIVES: Across existing real-world prostate cancer datasets, the way variables are captured is heterogenous. One key variable is disease progression, a component of progression-free survival. In some datasets all progressions are captured longitudinally, while in others progressions are captured only at specific milestones such as treatment changes. The purpose of this study was to conduct a comparative analysis of the longitudinal vs. milestone-driven method to determine if progressions unconnected to systemic treatment changes provide additional value.

METHODS: This was a retrospective cohort study using ConcertAI electronic health record data and including United States adult men with metastatic castration-resistant prostate cancer (mCRPC) during 01/01/2008–02/06/2023. Three methods of documenting disease progression were compared: 1) the longitudinal approach; 2) only progressions ±14 days of treatment change; 3) only progressions ±28 days of treatment change. An extended Cox model was used to analyze the relationship between progression and mortality. Model fit was evaluated in terms of concordance, log-likelihood, and Akaike information criterion (AIC). To further explore the clinical relevance of each method of capturing progression, through consideration of disease progression (vs. no progression) and death at any time in the follow-up of patients, chi-square test and Pearson correlation were calculated.

RESULTS: The study included 3,855 mCRPC patients. Across the three different methods of capturing progression, no significant difference was observed in any measure of association or discrimination between those who died and those who did not. However, linking progression to treatment changes resulted in better estimates of model fit (log-likelihood and AIC) and higher chi-square and Pearson correlation values.

CONCLUSIONS: This analysis suggests that there is little additional benefit to capturing progressions not linked to treatment changes when modeling the relationship between progression and mortality in mCRPC. Given the lack of standardization, these findings will facilitate comparison of outcomes across different data sources.