Empirical-Null Calibration Challenges Comparative CNS Safety Signals for Androgen-Receptor Pathway Inhibitors: A FAERS Re-Analysis

Background/Objectives: The four androgen-receptor pathway inhibitors (ARPIs) for advanced prostate cancer are broadly comparable in efficacy, so the choice between them depends on tolerability and, in elderly patients, on central nervous system (CNS) safety. CNS signals from the FDA Adverse Event Reporting System (FAERS) now inform that choice, yet they rest on uncalibrated disproportionality, which cannot separate pharmacology from reporting. Methods: Across the full FAERS database (20,328,575 reports), we computed reporting odds ratios (ROR) for enzalutamide, apalutamide, darolutamide, and abiraterone over 32 drug–event pairs, recalibrated every estimate against a drug-specific empirical null from 18 negative controls, and corrected for multiplicity. The null was fitted two ways and validated against twelve label-established positive controls. Results: Uncalibrated analysis flagged 22 of 32 pairs, and also 41 of 72 control pairs on events judged to carry no true effect, so its null does not hold. Calibration removes that bias, but the verdict then depends on how the null is fitted: with its dispersion taken as the sample standard deviation of the control estimates, no pair reached q < 0.05, while under the maximum-likelihood fit the framework prescribes six did, three of them being CNS pairs. Positive controls confirmed that calibration retains genuine agent-specific effects, and the detection floor lay above every CNS estimate for the same agent. Conclusions: Comparative ARPI CNS signals in FAERS are not identifiable; the verdict is set by the calibration design rather than by the drugs, so such signals should not stand alone in agent selection. The firmer safety axis remains fall and fracture risk, the class’s dominant serious harm in trials and guidelines.