Reanalysis of LUNAR progression-free survival using reconstructed individual patient data: time-varying treatment effects.

Abstract

Purpose: To assess whether treatment effects in the LUNAR trial are time-varying and to contextualize a single hazard ratio (HR) using complementary, clinically interpretable summaries of when and for how long separation is supported by available information, in a secondary methodological reassessment.

Methods: Individual patient data were reconstructed by digitalizing published progression-free survival (PFS) Kaplan-Meier (KM) curves. Proportional hazards (PH) were assessed, and treatment effects were summarized using Fleming-Harrington weighted log-rank tests, a piecewise Cox summary, and restricted mean survival time (RMST) at a priori horizons.

Results: The reconstructed and published KM curves closely matched the trajectories, and the reconstructed Cox HR was consistent with the published HR. PH diagnostics suggested non-proportional hazards (global Schoenfeld test p = 0.001). The early-weighted Fleming-Harrington test showed the strongest separation (χ² = 22.0, p < 0.001), whereas the late-weighted test did not provide confirmatory evidence (χ² = 2.51, p = 0.110). RMST favored ¹⁷⁷Lu-PSMA plus stereotactic body radiotherapy (relative difference in RMST (%) at 12 and 24 months, 43.7 [95% confidence interval: 24.9-65.3] and 67.8 [33.5-111], respectively). However, late follow-up estimates became increasingly imprecise as risk sets decreased.

Conclusion: This reconstruction-based, hypothesis-generating reanalysis supported the primary inference of PFS benefit while suggesting time-varying effects. Weighted tests, piecewise summaries, and RMST may provide interpretable complements to a single HR. Late follow-up interpretation remains uncertain because of sparse late risk sets and events.