Flatiron EHR data again favor enzalutamide over abiraterone in first-line mCRPC
A large electronic health record analysis drawn from the Flatiron Health database found shorter overall survival with first-line abiraterone acetate than with enzalutamide in chemotherapy-naïve, androgen receptor pathway inhibitor-naïve metastatic castration-resistant prostate cancer. The finding is clinically consequential because this is the same direction of effect seen in prior claims-based work, now confirmed with more granular chart abstraction rather than billing codes alone.
The cohort included 2,731 patients treated between 2014 and 2023. Of those, 1,316 started abiraterone and 1,415 started enzalutamide. Median follow-up was 21.8 months in the abiraterone group and 23.2 months in the enzalutamide group. After stabilized inverse probability treatment weighting adjusted for baseline differences, the median overall survival was 24.6 months with abiraterone and 25.8 months with enzalutamide, with a hazard ratio of 1.13 and a 95% confidence interval of 1.03 to 1.23.
The survival gap was modest in absolute terms but consistent across endpoints. Treatment duration and time to subsequent therapy both ran shorter in the abiraterone arm, and the pattern persisted in the subgroup of patients who never received another line of treatment.
Shorter treatment duration tracked with the survival signal
Treatment duration reached a median of 7.9 months with abiraterone versus 9.5 months with enzalutamide after weighting, corresponding to a hazard ratio of 1.21 with a 95% confidence interval of 1.11 to 1.31. Time to subsequent therapy also favored enzalutamide: 14.7 months versus 17.9 months, with a hazard ratio of 1.20 and a 95% confidence interval of 1.08 to 1.32.
This alignment matters clinically because it suggests the overall survival difference was not driven by a single isolated endpoint. Patients on abiraterone moved off initial therapy sooner and reached the next treatment line earlier. In metastatic castration-resistant prostate cancer, where disease biology often accelerates once resistance emerges, shorter disease control on first-line androgen receptor pathway inhibition can translate into less time before clinical deterioration, more frequent therapy transitions, and narrower room for salvage sequencing.
The mechanistic contrast between the two agents remains biologically plausible. Abiraterone suppresses androgen synthesis through CYP17 inhibition and requires concomitant corticosteroid use, while enzalutamide directly antagonizes the androgen receptor signaling axis. Even within the same therapeutic class, differences in endocrine suppression, receptor blockade, adherence, and tolerability can shape how long disease remains controlled in the real world.
A no-subsequent-treatment subgroup sharpened the difference
A prespecified subgroup analysis focused on the 1,102 patients, representing about 40% of the cohort, who received first-line ARPI only and no subsequent treatment. In that population, the survival separation widened. Weighted median overall survival was 14.0 months with abiraterone and 17.7 months with enzalutamide, with a hazard ratio of 1.32 and a 95% confidence interval of 1.14 to 1.53.
Treatment duration in the same subgroup also remained shorter with abiraterone: 7.5 months versus 10.6 months, with a hazard ratio of 1.30 and a 95% confidence interval of 1.12 to 1.50. The sharper divergence in this subset argues against the survival pattern being entirely explained by downstream treatment choices. When no later therapy enters the equation, the initial drug’s depth and durability of disease control become even more central to the observed outcome.
One sentence captures the bedside meaning: first-line choice in this setting can alter how long a patient remains on therapy before progression forces a change.
Why EHR data matter in metastatic castration-resistant prostate cancer
Prior analyses using claims databases had already pointed toward worse overall survival with abiraterone than with enzalutamide in chemotherapy- and ARPI-naïve mCRPC, but claims datasets often miss the clinical texture that shapes treatment sequencing. Electronic health records offer a fuller view of real-world practice, including the ability to define eligibility with greater precision and to capture treatment duration and subsequent therapy more directly.
Here, that added granularity did not erase the earlier signal. It reinforced it.
The study design also helps explain why the result carries practical weight. Investigators used Kaplan-Meier methods and Cox proportional hazards models for time-to-event endpoints, then applied stabilized inverse probability treatment weighting to reduce imbalance at baseline. In a disease setting where treatment selection can be influenced by performance status, comorbidity, lab trends, and physician preference, that statistical adjustment is essential. It does not eliminate residual confounding, but it strengthens confidence that the observed differences were not simply an artifact of obvious baseline imbalances.
What oncologists can take from the survival gap
For clinicians treating chemotherapy-naïve mCRPC, this analysis pushes first-line ARPI selection from a theoretical debate into a more immediate sequencing issue. Abiraterone remains widely used, but these data suggest enzalutamide may offer more durable disease control and longer survival in routine practice for patients who have not yet received chemotherapy or another ARPI.
The absolute survival difference between the two agents was small in the full cohort, yet the consistency across overall survival, treatment duration, and time to subsequent therapy makes the pattern hard to ignore. In oncology, a few months can matter when the disease is advanced and treatment options narrow with each progression event.
The trial registration number, NCT05968599, signals formal prospective registration of the research program, but the analysis itself remains observational. Selection bias, unmeasured disease burden, differences in prescribing behavior, and factors not fully captured in EHR abstraction can still shape outcomes. Those limitations do not negate the finding; they define how cautiously it should be interpreted.
The main unresolved question is not whether the two drugs differ in this dataset. They do. The remaining issue is how much of the gap reflects underlying patient selection and how much reflects a true treatment effect in everyday oncology practice. Additional comparative effectiveness work, especially with richer clinical covariates and treatment sequencing data, will be needed to settle that boundary more firmly.