Exploratory Time‐Dependent Patterns in Estimated Treatment Effect of Talazoparib Plus Enzalutamide in HRR Non‐Deficient or Unknown Metastatic Castration‐Resistant Prostate Cancer: A Post Hoc Analysis of the TALAPRO‐2 Trial

ABSTRACT

Objectives

Talazoparib plus enzalutamide improved overall survival (OS) in the biomarker‐unselected TALAPRO‐2 population, but OS benefit was not statistically significant in the homologous recombination repair (HRR) non‐deficient or unknown subgroup, whereas radiographic progression‐free survival (rPFS) was prolonged. We explored time‐dependent patterns in estimated treatment effects in this subgroup.

Methods

Approximate individual‐level data for 636 patients were reconstructed from published Kaplan–Meier curves; original trial‐level patient data were unavailable. The proportional hazards assumption was assessed using Schoenfeld residuals and log–log plots. Interval‐specific hazard ratios (HRs) were estimated using piecewise Cox models. A treatment‐by‐log(time + 1) interaction was evaluated formally. Sensitivity analyses included alternative interval schemes, landmark analyses, restricted mean survival time (RMST), and descriptive application of the same piecewise framework to the HRR‐deficient subgroup.

Results

The proportional hazards assumption was not rejected for OS (p = 0.63) or rPFS (p = 0.37). For OS, the 36–48‐month interval showed a nominally lower hazard with talazoparib (HR 0.59; 95% confidence interval [CI] 0.36–0.97; nominal p = 0.039). For rPFS, the lowest HR occurred at 0–6 months (HR 0.56; 95% CI 0.37–0.86; p = 0.008). Treatment‐by‐log(time + 1) interactions were not statistically significant for OS (p = 0.44) or rPFS (p = 0.12).

Conclusions

This reconstructed‐data post hoc analysis suggested exploratory time‐dependent patterns in effect estimates without formal evidence of a time‐varying treatment effect. Findings are descriptive, hypothesis‐generating, and should not be extrapolated to molecularly confirmed HRR‐proficient disease.

Trial Registration

ClinicalTrials.gov: NCT03395197

1. Introduction

Poly(ADP‐ribose) polymerase (PARP) inhibitors combined with androgen receptor pathway inhibitors (ARPIs) have emerged as an important therapeutic approach in metastatic castration‐resistant prostate cancer (mCRPC). Three phase 3 trials, namely TALAPRO‐2 (talazoparib plus enzalutamide) [1, 2], PROpel (olaparib plus abiraterone) [3, 4], and MAGNITUDE (niraparib plus abiraterone) [5, 6], have evaluated such combinations as first‐line treatment for mCRPC. While significant radiographic progression‐free survival (rPFS) benefits have been demonstrated across these trials, the extent of benefit among patients lacking detectable homologous recombination repair (HRR) gene alterations has remained a matter of ongoing discussion.

In the TALAPRO‐2 trial, the mature overall survival (OS) data demonstrated a statistically significant survival benefit with the combination in the overall population (hazard ratio [HR] 0.80; 95% confidence interval [CI] 0.66–0.96; p = 0.016), with the largest magnitude of effect among patients with confirmed HRR deficiency (HR 0.55; 95% CI 0.36–0.83; p = 0.0035) [1]. However, among the 636 patients lacking confirmed HRR alterations (hereafter the HRR non‐deficient or unknown subgroup; this trial‐defined group is not equivalent to molecularly confirmed HRR‐proficient disease), the OS benefit associated with the combination did not attain statistical significance (HR 0.88; 95% CI 0.71–1.08; p = 0.22) [1]. In contrast, the rPFS endpoint in this same population reached statistical significance in the primary analysis (HR 0.70; 95% CI 0.54–0.89; p = 0.004) [2].

This discrepancy between OS and rPFS outcomes in the HRR non‐deficient or unknown subgroup raises an important question: does the treatment effect vary over time in a manner not fully captured by a single overall HR? Under the Cox proportional hazards (PH) framework, the treatment effect is assumed to remain constant throughout follow‐up; however, it is widely acknowledged that clinically relevant temporal variation may be obscured when the benefit is summarized by a single HR. Interval‐specific analyses and measures such as the restricted mean survival time (RMST) can provide complementary perspectives on how treatment benefits evolve during follow‐up [7].

The purpose of this study was to explore whether the treatment effect of talazoparib plus enzalutamide in the TALAPRO‐2 HRR non‐deficient or unknown subgroup varied over time. We employed piecewise Cox regression and RMST analysis to characterize temporal patterns of treatment effects for both OS and rPFS. Importantly, this trial‐defined subgroup included both HRR non‐deficient and unknown status and should not be considered equivalent to a molecularly confirmed HRR‐proficient population.

2. Methods

2.1. Data Source

This study was a post hoc secondary analysis of published data from the TALAPRO‐2 trial (ClinicalTrials.gov: NCT03395197). In the phase 3 TALAPRO‐2 trial, participants with mCRPC who were unselected for HRR gene alteration status were randomly assigned in double‐blind fashion to talazoparib (0.5 mg daily) combined with enzalutamide (160 mg daily) or to matching placebo combined with enzalutamide [2]. OS data were obtained from the final analysis (Lancet 2025; data cutoff September 3, 2024; median follow‐up 52.5 months) [1], and rPFS data were obtained from the primary analysis (Lancet 2023; data cutoff August 16, 2022) [2]. Because the two endpoints came from reports with different data maturity, cross‐endpoint comparisons were considered descriptive. The analysis focused on the subgroup of patients with non‐deficient or unknown HRR gene alteration status by randomization stratification (N = 636; talazoparib plus enzalutamide, n = 317; placebo plus enzalutamide, n = 319).

2.2. Reconstruction of Individual Patient Data

Approximate individual‐level data were generated from digitized Kaplan–Meier (KM) curves using the iterative algorithm described by Guyot et al. [8] Curve coordinates were extracted from published figures with WebPlotDigitizer (version 5). Number‐at‐risk counts extracted from the original reports served as input for the reconstruction, executed through the preprocess() and getIPD() functions of the IPDfromKM package for R [9]. Reconstructed individual patient data (IPD; hereafter rIPD) were generated separately for each treatment arm. The quality of reconstruction was validated by comparing the overall HR and median survival estimates derived from the rIPD against the published values, with the percent deviation calculated as |HR(rIPD) − HR(published)| / HR(published) × 100. Digitization was performed once by a single investigator (Shugo Yajima) and cross‐checked against number‐at‐risk tables and median survival times before analysis. Censoring was handled internally by the IPDfromKM algorithm, which distributes censoring times across at‐risk intervals to reproduce the published Kaplan–Meier step pattern. Reconstructed datasets and analytic code have been deposited in Zenodo (DOI: 10.5281/zenodo.20370591).

2.3. Assessment of the Proportional Hazards Assumption

The PH assumption was assessed using two complementary strategies. The cox.zph() function in the survival package for R was used to evaluate the correlation between scaled Schoenfeld residuals and time [10]. Log–log plots were also inspected visually for departures from proportionality or temporal variation in hazard ratios.

2.4. Piecewise Cox Proportional Hazards Model

To estimate interval‐specific HRs, follow‐up was divided into intervals specified for this exploratory analysis. For each endpoint, the reconstructed data were expanded into a counting‐process format with start and stop times for each interval, and separate Cox models with treatment as the sole covariate were fitted within each interval among patients at risk at the beginning of that interval. This approach follows the methodology described by Chen et al. [11] The intervals specified for this exploratory analysis were 0–12, 12–24, 24–36, 36–48, and > 48 months for OS, and 0–6, 6–12, 12–18, 18–24, 24–30, and > 30 months for rPFS. These cut points were determined before fitting the interval‐specific models to balance temporal resolution and the number of events per interval.

2.5. Time‐Varying Treatment Effect

A Cox model including treatment and a treatment‐by‐log(time + 1) interaction term was fitted as a formal assessment of time dependence. A non‐significant interaction was interpreted as no formal statistical evidence of a time‐varying treatment effect on that model scale.

2.6. Sensitivity and Subgroup Analyses

Sensitivity to interval choice was examined using uniform 3‐month, uniform 6‐month, and shifted‐boundary schemes (OS cuts at 9, 18, 30, and 42 months; rPFS cuts at 4, 10, 16, and 22 months). For descriptive comparison, the same rIPD and piecewise framework was applied to the HRR‐deficient subgroup (n = 169; talazoparib plus enzalutamide, n = 85; placebo plus enzalutamide, n = 84), using published Kaplan–Meier curves and number‐at‐risk data from the corresponding TALAPRO‐2 reports [1, 2]. Wider intervals were used because of smaller event counts, and no formal interaction testing across subgroups was performed.

2.7. Restricted Mean Survival Time Analysis

RMST differences (talazoparib plus enzalutamide minus placebo plus enzalutamide) were calculated using the rmst2() function in the survRM2 package for R [12]. The RMST difference was computed at monthly truncation times (τ) from 1 month to the maximum follow‐up (54 months for OS; 30 months for rPFS). The resulting RMST difference trajectories were plotted with pointwise 95% CIs to visualize cumulative treatment benefit over time. Because repeated tau‐specific comparisons were exploratory and no multiplicity adjustment was applied, RMST results were interpreted descriptively.

2.8. Landmark Analyses

Landmark analyses were performed as sensitivity analyses. For OS, landmarks were set at 24 and 36 months, and patients alive at each landmark were included. For rPFS, a 12‐month landmark was used, including patients who remained progression‐free at the landmark. In each analysis, subsequent hazards were compared between treatment arms using Cox models.

2.9. Statistical Software

All computations were conducted in R (version 4.5.2); key packages included survival, survRM2, IPDfromKM, and ggplot2. All reported p values are two‐sided. A nominal alpha level of 0.05 was used for descriptive reference only; because interval‐specific and repeated tau‐specific comparisons were exploratory and no multiplicity adjustment was applied, these p values were not interpreted as confirmatory.

3. Results

3.1. Quality of Reconstructed Individual Patient Data

For OS, the reconstructed data yielded an overall HR of 0.88 (95% CI 0.72–1.09; p = 0.24), closely matching the published value of 0.88 (95% CI 0.71–1.08; p = 0.22), with a relative deviation of 0.2%. Median OS was 45.5 months (95% CI 39.0–50.9) in the talazoparib group and 38.3 months (95% CI 35.0–43.5) in the placebo group. The reconstructed KM curves closely replicated the overall pattern of the published survival curves (eFigure S1A). For rPFS, the reconstructed HR was 0.69 (95% CI 0.54–0.88; p = 0.003), compared with the published value of 0.70 (95% CI 0.54–0.89; p = 0.004), representing a relative deviation of 1.8%. Median rPFS was not reached in the talazoparib group and was 22.6 months (95% CI 19.2–30.7) in the placebo group (eFigure S1B).

3.2. Proportional Hazards Assumption Testing

The Schoenfeld residual test did not reject the PH assumption for either endpoint (OS: p = 0.63; rPFS: p = 0.37). These results did not provide statistical evidence against proportional hazards, but they do not establish that the treatment effect was constant over time or that the overall HR fully captured temporal patterns. Visual inspection of the log–log plots revealed subtle non‐parallelism between treatment arms, particularly for OS, where the curves converged around 24–36 months and diverged again thereafter (eFigure S2). Schoenfeld residual plots showed temporal trends in residuals for both endpoints, although these trends did not reach statistical significance in the global test (eFigure S3). To complement the overall HR and explore possible time‐dependent patterns in the estimates, we performed interval‐specific analyses.

3.3. Piecewise Hazard Ratio Estimation

For OS (Table 1), piecewise Cox regression showed varying point estimates across time intervals. The HR was 0.90 (95% CI 0.56–1.46; p = 0.68) during 0–12 months, 0.83 (95% CI 0.56–1.24; p = 0.37) during 12–24 months, 1.08 (95% CI 0.73–1.61; p = 0.70) during 24–36 months, and 0.59 (95% CI 0.36–0.97; nominal p = 0.039) during 36–48 months (Figure 1A). The 36–48‐month interval showed the lowest point estimate among all intervals, and its pointwise 95% CI excluded 1.0; however, given the exploratory nature of this analysis, the absence of multiplicity adjustment, and the dependence on interval specification, this finding should be regarded as nominal and hypothesis‐generating rather than confirmatory. In particular, because the formal treatment‐by‐log(time + 1) interaction was not statistically significant (reported below), this interval‐dependent observation does not constitute confirmatory evidence of a genuine time‐varying treatment effect. In the > 48‐month interval, the HR was 1.29 (95% CI 0.58–2.87; p = 0.54), although this estimate was based on only 25 events and should be interpreted with caution.

TABLE 1.

Piecewise hazard ratios for overall survival and radiographic progression‐free survival in the HRR non‐deficient or unknown subgroup.

Time interval	At‐risk (n)	Events (n)	HR (95% CI)	p
Overall survival (overall HR 0.88; 95% CI 0.72–1.09; p = 0.24)
0–12 months	636	68	0.90 (0.56–1.46)	0.68
12–24 months	535	99	0.83 (0.56–1.24)	0.37
24–36 months	428	97	1.08 (0.73–1.61)	0.70
36–48 months	318	64	0.59 (0.36–0.97)	0.039
48+ months	231	25	1.29 (0.58–2.87)	0.54
Radiographic progression‐free survival (overall HR 0.69; 95% CI 0.54–0.88; p = 0.003)
0–6 months	636	93	0.56 (0.37–0.86)	0.008
6–12 months	474	73	0.85 (0.54–1.34)	0.48
12–18 months	365	43	0.61 (0.33–1.12)	0.11
18–24 months	296	32	0.80 (0.40–1.59)	0.52
24–30 months	185	10	1.11 (0.31–3.94)	0.87
30+ months	82	3	N/A	N/A

Note: The 30+ month rPFS interval had only 3 events among 82 patients at risk, which was insufficient to fit a Cox proportional hazards model. The Schoenfeld residual test for the proportional hazards assumption yielded p = 0.63 for overall survival and p = 0.37 for radiographic progression‐free survival; neither result rejected the proportional hazards assumption at the 0.05 level.

Abbreviations: CI, confidence interval; HR, hazard ratio; HRR, homologous recombination repair; N/A, not available.

FIGURE 1.

Piecewise hazard ratios (forest plots) for the HRR non‐deficient or unknown subgroup. (A) Overall survival, with intervals of 0–12, 12–24, 24–36, 36–48, and > 48 months. (B) Radiographic progression‐free survival, with intervals of 0–6, 6–12, 12–18, 18–24, and 24–30 months. The > 30‐month rPFS interval was excluded owing to insufficient events. The blue dotted vertical line indicates the overall HR calculated from the reconstructed individual patient data (0.88 for overall survival; 0.69 for radiographic progression‐free survival), which closely matched the published overall HRs from the TALAPRO‐2 reports (0.88 and 0.70, respectively). The gray dashed vertical line indicates HR = 1.0. Error bars represent 95% CIs. Hazard ratios < 1 favor talazoparib plus enzalutamide. CI, confidence interval; HR, hazard ratio; HRR, homologous recombination repair.

For rPFS (Table 1), the lowest HR was observed in the first 6 months (HR 0.56; 95% CI 0.37–0.86; p = 0.008), after which the interval‐specific estimates attenuated. HRs for subsequent intervals were 0.85 (6–12 months), 0.61 (12–18 months), 0.80 (18–24 months), and 1.11 (24–30 months), none of which reached statistical significance (Figure 1B). In the > 30‐month interval, only 3 events occurred among 82 patients at risk, which was insufficient to fit a Cox model; this interval was therefore excluded from the piecewise analysis. These findings suggest that the overall rPFS benefit may have been influenced primarily by early separation of the curves, with less precise estimates thereafter. As with overall survival, these interval‐specific rPFS estimates are exploratory and dependent on the chosen interval boundaries and should not be interpreted as confirmatory evidence of a time‐varying treatment effect.

3.4. Time‐Varying Treatment Effect, HRR‐Deficient Subgroup, and Interval Sensitivity

Treatment‐by‐log(time + 1) interactions were not statistically significant for OS (p = 0.44) or rPFS (p = 0.12), indicating no formal statistical evidence for time‐varying treatment effects. In the HRR‐deficient subgroup, the strongest HR reductions in the piecewise analysis occurred in the earliest intervals (eFigure S4): OS 0–12 months HR 0.06 (95% CI 0.01–0.44; p = 0.006) and rPFS 0–6 months HR 0.23 (95% CI 0.08–0.62; p = 0.004). Interval‐specific numbers at risk and event counts for this subgroup are provided in eTable S1. The very low 0–12‐month OS hazard ratio rested on only one event in the talazoparib arm versus 15 in the placebo arm and is therefore highly unstable and should be interpreted with particular caution. Given the smaller subgroup size, these estimates were interpreted qualitatively, and because no formal interaction test was performed between the HRR‐deficient and HRR non‐deficient or unknown subgroups, any apparent differences between subgroups are descriptive only and must not be interpreted as a formal between‐subgroup comparison. Across alternative interval schemes, the direction of the interval‐specific estimates was broadly consistent, but their magnitude varied substantially—particularly in sparse tail intervals—so this pattern should be regarded as suggestive rather than a robust feature of the data (eFigure S5). Reconstruction validation for the HRR‐deficient subgroup is shown in eFigure S6.

3.5. Restricted Mean Survival Time Analysis

RMST analysis provided a complementary perspective on cumulative treatment benefit (Table 2). For OS, the RMST difference remained imprecise throughout follow‐up, gradually increasing from −0.03 months (95% CI −0.27 to 0.21) at 12 months to +1.03 months (95% CI −1.35 to 3.40) at 48 months (Figure 2A). The OS RMST difference became increasingly positive after approximately 36 months, though the wide CIs throughout indicate substantial uncertainty. For rPFS, the RMST difference increased progressively in favor of talazoparib (Figure 2B). Because the repeated pointwise comparisons across truncation times were exploratory and unadjusted for multiplicity, we focus on the overall trajectory and a small number of clinically relevant truncation times rather than on individual pointwise p values. At a 12‐month truncation the difference was +0.81 months (95% CI 0.30–1.32), and at the maximum evaluable truncation of 30 months it reached +2.67 months (95% CI 0.91–4.43); the trajectory rose steadily over follow‐up.

TABLE 2.

Restricted mean survival time differences at key truncation times in the HRR non‐deficient or unknown subgroup.

Truncation time (τ)	RMST difference, months (95% CI)	p
Overall survival
12 months	−0.03 (−0.27 to +0.21)	0.83
24 months	0.00 (−0.87 to +0.86)	1.00
36 months	+0.19 (−1.41 to +1.80)	0.81
48 months	+1.03 (−1.35 to +3.40)	0.40
Radiographic progression‐free survival
6 months	+0.34 (+0.16 to +0.51)	< 0.001
12 months	+0.81 (+0.30 to +1.32)	0.002
18 months	+1.30 (+0.40 to +2.21)	0.005
24 months	+1.97 (+0.65 to +3.29)	0.004
30 months	+2.67 (+0.91 to +4.43)	0.003

Note: RMST difference = talazoparib plus enzalutamide minus placebo plus enzalutamide. Positive values favor the talazoparib plus enzalutamide group. p values are nominal, derived from pointwise comparisons at each truncation time, and were not adjusted for multiplicity; they are provided for descriptive reference only and should not be interpreted as confirmatory.

Abbreviations: CI, confidence interval; HRR, homologous recombination repair; RMST, restricted mean survival time.

FIGURE 2.

Restricted mean survival time difference curves for the HRR non‐deficient or unknown subgroup. (A) Overall survival. (B) Radiographic progression‐free survival up to 30 months. The solid line represents RMST difference (talazoparib plus enzalutamide minus placebo plus enzalutamide) at each truncation time. Positive values favor talazoparib plus enzalutamide. Shaded areas represent pointwise 95% CIs. The horizontal dashed line at zero indicates no difference. CI, confidence interval; HRR, homologous recombination repair; RMST, restricted mean survival time; rPFS, radiographic progression‐free survival; τ, truncation time.

3.6. Landmark Analyses

Landmark analyses were performed as sensitivity analyses to complement the piecewise findings. For OS, the HR among patients surviving beyond 24 months was 0.90 (95% CI 0.68–1.20; p = 0.47; n = 428), and among those surviving beyond 36 months was 0.74 (95% CI 0.49–1.12; p = 0.15; n = 318). Although the point estimate of the HR decreased from the 24‐month to the 36‐month landmark, neither estimate reached statistical significance and the 95% CIs included 1.0, precluding definitive conclusions regarding a delayed OS benefit. For rPFS, the HR among patients progression‐free beyond 12 months was 0.71 (95% CI 0.46–1.07; p = 0.10; n = 365).

4. Discussion

The present analysis applied piecewise Cox regression and RMST methods to investigate exploratory time‐dependent patterns in estimated treatment effects of talazoparib plus enzalutamide within the HRR non‐deficient or unknown subgroup of TALAPRO‐2. Although the PH assumption was not formally rejected, interval‐specific analyses showed different patterns by endpoint. Formal treatment‐by‐log(time + 1) interactions were not statistically significant; therefore, the piecewise and RMST findings should be interpreted as descriptive rather than inferential evidence of time‐varying treatment effects.

The OS findings merit careful consideration. The overall HR of 0.88 was not statistically significant, and only the 36–48‐month interval showed a nominally lower hazard. This observation was obtained without multiplicity adjustment and was sensitive to interval specification, although late‐follow‐up intervals overlapping this period showed qualitatively similar favorable point estimates in sensitivity analyses. The finding should not be viewed as clinically actionable. Given the non‐significant formal interaction test, it does not establish a genuine time‐varying treatment effect. For rPFS, the most favorable HR occurred in the first 6 months, and the widening RMST difference likely reflected the cumulative impact of early curve separation rather than continuing benefit at every later interval.

The early rPFS signal observed in the HRR non‐deficient or unknown subgroup may warrant clinical and biological interpretation. Combined androgen receptor pathway inhibition and PARP inhibition may rapidly suppress tumor growth and exploit DNA repair vulnerability in some patients without detectable HRR alterations; under‐detection due to limited panel scope or intratumoral heterogeneity may also contribute [13]. However, these observations do not justify a change in routine practice for HRR non‐deficient or unknown patients. The potential early‐rPFS signal must also be weighed against the established toxicity of talazoparib‐based combination therapy: in TALAPRO‐2, the addition of talazoparib increased hematologic toxicity—most notably anemia, the most frequent grade 3–4 adverse event—and commonly required dose interruption, dose reduction, or transfusion, with treatment discontinuation in a subset of patients [1, 2]. In a subgroup that lacks confirmed HRR deficiency and in which the OS benefit did not reach statistical significance, this added burden is particularly relevant to the benefit–risk balance, and an early rPFS signal alone does not justify combination therapy in routine practice.

The HRR‐deficient subgroup provides descriptive context only: in that subgroup, the most pronounced hazard reductions appeared in the earliest intervals for both OS and rPFS, directionally compatible with greater sensitivity to PARP inhibition, although this cannot by itself establish a subgroup difference. Because no formal interaction test was performed across subgroups, any apparent differences between the HRR‐deficient and HRR non‐deficient or unknown subgroups should be regarded as descriptive and not as a formal between‐subgroup comparison. These results should also be interpreted alongside PROpel and MAGNITUDE, but differences in biomarker assays, population definitions, post‐progression treatment, and available subgroup Kaplan–Meier data limit direct cross‐trial reconstruction [3, 4, 5, 6]. The HRR non‐deficient or unknown subgroup also combines biologically distinct categories; unknown status should not be assumed to represent molecular HRR proficiency.

Several limitations warrant emphasis. The analysis relied on reconstructed rather than original trial‐level patient data and could not recover covariates, censoring mechanisms, treatment switching, adverse‐event timing, or imaging‐assessment details. Interval‐specific and RMST analyses were exploratory and unadjusted for multiplicity; late intervals had few events and wide CIs. OS and rPFS were reconstructed from reports with different data cutoffs, follow‐up maturity, and post‐progression treatment exposure, limiting cross‐endpoint interpretation. Finally, landmark and interval‐specific estimates are conditional on remaining alive or progression‐free at interval start and may partly reflect changes in the risk set rather than biological time variation. Landmark estimates are therefore supportive but heavily conditional and should not be viewed as independent confirmation of delayed benefit.

In summary, piecewise and RMST analyses suggested exploratory time‐dependent patterns in OS and rPFS effect estimates for talazoparib plus enzalutamide in the TALAPRO‐2 HRR non‐deficient or unknown subgroup. Because formal time‐varying tests were non‐significant and all analyses were reconstructed and descriptive, these findings should be considered hypothesis‐generating and require validation in datasets with longer follow‐up and molecularly defined populations.

Author Contributions

Shugo Yajima: conceptualization, formal analysis, data curation, visualization, writing – original draft. Hitoshi Masuda: writing – review and editing. Soichiro Yoshida: conceptualization, supervision, writing – review and editing. Hiroshi Fukushima: investigation, writing – review and editing. Hiroyuki Sato: writing – review and editing. Wei Chen: methodology, formal analysis, writing – review and editing. Akihiro Hirakawa: writing – review and editing. Hajime Tanaka: investigation, writing – review and editing. Yasuhisa Fujii: supervision, writing – review and editing.

Ethics Statement

This investigation constitutes a secondary post hoc analysis relying entirely on aggregate summary data from previously published TALAPRO‐2 reports; no collection or handling of new patient‐level data was undertaken, and institutional ethical approval was accordingly not required.

Consent

The authors have nothing to report.

Conflicts of Interest

Yasuhisa Fujii is an Editorial Board Member of International Journal of Urology and a co‐author of this article. The other authors declare no conflicts of interest.

Supporting information

Figure S1:e Reconstructed Kaplan–Meier curves for the HRR non‐deficient or unknown subgroup (N = 636). (A) Overall survival. (B) Radiographic progression‐free survival. Overall HRs were derived from Cox proportional hazards models fitted to the reconstructed individual patient data.

Figure S2:e Log–log plots for the proportional hazards assumption. (A) Overall survival. (B) Radiographic progression‐free survival. Approximate parallelism of the curves is broadly consistent with the PH assumption but does not prove proportional hazards.

Abbreviations: PH, proportional hazards.

Figure S3:e Schoenfeld residual plots for the proportional hazards assumption. (A) Overall survival (cox.zph test, p = 0.63). (B) Radiographic progression‐free survival (cox.zph test, p = 0.37). The dashed line represents the smoothed trend. A relatively flat trend is broadly consistent with the PH assumption but does not establish it.

Figure S4:e Piecewise hazard ratios (forest plots) in the HRR‐deficient subgroup of the TALAPRO‐2 trial, from reconstructed individual patient data. (A) Overall survival, primary interval scheme (0–12, 12–24, 24–36, and > 36 months). (B) Radiographic progression‐free survival, primary interval scheme (0–6, 6–12, 12–18, and > 18 months). Reference: placebo plus enzalutamide; hazard ratios < 1 favor talazoparib plus enzalutamide. The gray dashed vertical line indicates a hazard ratio of 1.0. Error bars represent 95% CIs.

Figure S5:e Sensitivity of piecewise hazard ratios to interval specification in the HRR non‐deficient or unknown subgroup. (A) Overall survival. (B) Radiographic progression‐free survival. Each color represents one interval scheme: the primary scheme used in the main analysis, alternative uniform 3‐ and 6‐month grids, and a shifted‐boundary scheme (see Methods). Points are plotted at the midpoint of each interval; error bars represent 95% CIs. The horizontal dashed line indicates a hazard ratio of 1.0. Across schemes, the direction of the interval‐specific estimates was broadly consistent, but their magnitude varied substantially, particularly in the tail intervals with sparse events; the pattern should therefore be regarded as suggestive rather than as a robust feature of the data.

Figure S6:e Kaplan–Meier curves derived from the reconstructed individual patient data for the HRR‐deficient subgroup of the TALAPRO‐2 trial. (A) Overall survival. (B) Radiographic progression‐free survival. Red = talazoparib plus enzalutamide; blue = placebo plus enzalutamide. The hazard ratio annotation in each panel shows the value computed from the reconstructed individual patient data; the corresponding published hazard ratios are 0.55 (95% CI 0.36–0.83; p = 0.0035) for overall survival (Agarwal et al. Lancet [1], final overall survival analysis) and 0.45 (95% CI 0.33–0.61; p < 0.0001) for radiographic progression‐free survival (Agarwal et al. Lancet [2], primary analysis). The close agreement between reconstructed and published estimates supports the fidelity of the reconstruction for the descriptive comparisons reported in this study.

Table S1:e Interval‐specific numbers at risk, event counts, and piecewise hazard ratios in the HRR‐deficient subgroup of the TALAPRO‐2 trial (reconstructed individual patient data).

Data Availability Statement

Reconstructed individual‐level data generated through this analysis are available from a public Zenodo repository (DOI: 10.5281/zenodo.20370591; https://doi.org/10.5281/zenodo.20370591), together with the analytic code, under a CC‐BY 4.0 license for the data and an MIT license for the code. Underlying source data were obtained from Kaplan–Meier curves published in the primary TALAPRO‐2 reports.