Progression‐free survival and objective response rate as a surrogate end point for overall survival in metastatic renal cell carcinoma: A systematic review and meta‐analysis

ABSTRACT

The objective of this systematic review and meta‐analysis was to evaluate the validity of progression‐free survival (PFS) and the objective response rate (ORR) as surrogate end points for overall survival (OS) in randomized controlled trials (RCTs) of metastatic renal cell carcinoma. The MEDLINE, Embase, and Cochrane CENTRAL databases were searched from inception to June 10, 2025. Associations between treatment effects on OS, PFS, and ORR were assessed using Pearson correlation coefficients (r). Surrogate threshold effect (STE) analyses determined the minimum PFS hazard ratio (HR) or ORR odds ratio (OR) required to predict OS benefit. Subgroup analyses were conducted by treatment class, line of therapy, and histology. Sixty‐two randomized controlled trials comprising 24,518 patients were included. PFS‐OS demonstrated a moderate correlation (r = 0.52; 95% confidence interval [CI], 0.38–0.66), with an STE (HR, 0.92; 95% CI, 0.79–1.09). ORR‐OS demonstrated a moderate correlation (r = −0.59; 95% CI, −0.72, −0.45), with an STE (OR, 1.40; 95% CI, 0.99–1.89). Subgroup estimates varied across treatment settings and were frequently imprecise, particularly for ORR and in smaller treatment‐class subgroups; limited representation of non–clear cell renal cell carcinoma precluded reliable histology‐specific conclusions. Overall, PFS and ORR had moderate trial‐level associations with OS in metastatic renal cell carcinoma, but the overall analyses did not demonstrate sufficiently consistent surrogacy to reliably predict OS benefit. These findings should be interpreted cautiously given the substantial heterogeneity of the included trials. Surrogate validity appears to be context‐dependent, supporting continued reliance on OS and the need for more robust surrogate end points.

Short abstract

Progression‐free survival and the objective response rate demonstrate moderate trial‐level associations with overall survival in patients with metastatic renal cell carcinoma but lack clinically meaningful surrogacy when evaluated using surrogate threshold effects. Surrogate validity is highly context‐dependent, varying by treatment class, line of therapy, and histology.

INTRODUCTION

Metastatic renal cell carcinoma (mRCC) accounts for approximately 30% of initial RCC presentations, and an additional 30% of patients with localized disease will ultimately develop metastatic progression. ¹ The management of advanced RCC has evolved substantially over the past 2 decades, transitioning from cytokine‐based immunotherapies, such as high‐dose interleukin‐2 and interferon‐alpha, to newer therapies that target the vascular endothelial growth factor (VEGF) and mammalian target of rapamycin (mTOR) receptors, and immune checkpoint inhibitors (ICIs). ² Combination ICI therapy, such as ipilimumab plus nivolumab, and combination therapies, including pembrolizumab plus lenvatinib, nivolumab plus cabozantinib, and pembrolizumab plus axitinib, have now become the standard of care based on clinical trials demonstrating long‐term treatment efficacy. ³

Overall survival (OS) remains the gold standard for the assessment of treatment efficacy in clinical trials. However, its use as a primary end point requires prolonged follow‐up and large sample sizes to detect statistically significant differences, resulting in increased costs and longer trial follow‐up durations needed to prove durable treatment benefits. Surrogate end points, such as progression‐free survival (PFS) and the objective response rate (ORR), offer practical advantages, including shorter follow‐up duration and reduced confounding from subsequent lines of therapy. Consequently, these end points are frequently used in clinical trials and have been accepted in certain contexts to support regulatory approval. Nevertheless, the validity of PFS and ORR as surrogate end points for OS in mRCC remains uncertain.

A key limitation of many prior evaluations of surrogate end points is their reliance on measures of statistical association, such as correlation coefficients, without assessing whether observed effects can predict OS at the trial level. The surrogate threshold effect (STE) provides a more rigorous framework for surrogate validation by estimating the minimum treatment effect on a surrogate end point required to reliably predict a significant benefit in OS. ⁴ Unlike correlation alone, the STE enables assessment of whether a given magnitude of improvement in PFS or ORR can reliably suggest survival benefit, thereby offering greater clinical and regulatory interpretability.

This distinction is particularly relevant in the context of drug approval. Regulatory agencies such as the US Food and Drug Administration have increasingly relied on surrogate end points to support accelerated approval pathways for newer therapies in RCC, especially when OS data are immature. ⁵ However, the strength and consistency of the relation between surrogate end points and OS directly influence the certainty of clinical benefit and the risk of approving therapies that may not ultimately improve survival. As such, robust validation of surrogate end points, particularly through methods incorporating STE, is essential to inform both clinical trial design and regulatory decision‐making.

In this context, we conducted a systematic review and meta‐analysis to evaluate the strength of PFS and ORR as surrogate end points for OS in randomized controlled trials (RCTs) of mRCC. By incorporating both traditional measures of association and STE‐based analyses, the objective of this study was to provide a more comprehensive assessment of surrogate validity and to inform the appropriate use of these end points in future trials and regulatory frameworks.

MATERIALS AND METHODS

Search strategy

We searched the MEDLINE, Embase, and the Cochrane CENTRAL databases for publications from inception to June 10, 2025. We used keywords to identify RCTs involving neoplasms, cancers, tumors, and carcinomas of the kidney and renal system. The search was limited to only include studies published in English. The full search strategy is provided in Table S1.

Selection criteria

Studies were included if they were RCTs that included participants with mRCC and evaluated two different therapeutic interventions, including systemic and locoregional therapies. Studies that involved both treatment‐naive and previously treated cohorts were included. Abstracts were included provided there were sufficient outcome data. Studies were excluded if they were not RCTs or only included participants with primary localized RCC or non‐RCC kidney tumors. Studies that only involved a single arm, assessed the same intervention in both treatment arms, or lacked relevant outcome data were also excluded. Data screening was done in duplicate (by R.L., L.A., N.L., and K.X.), and conflicts were resolved by a third reviewer (D.C.).

Data extraction

The following information was extracted from each included trial: first author’s name, year of publication, trial phase, year of first and last enrollment, number of enrolled patients per arm, treatment arms, median follow‐up, and outcome data, including PFS, OS, and ORR (e.g., hazard ratio [HR], odds ratio [OR], and 95% confidence interval [CI] for each estimate). If data were not reported, OS or PFS rates were extracted from the Kaplan–Meier curves or prior meta‐analyses identifying these numbers. Data extraction was done in duplicate (by R.L., L.A., N.L., and K.X.) using a standardized form, and conflicts were resolved by a third reviewer (D.C.). Studies were then de‐duplicated to include the publication with the longest median follow‐up that included all relevant outcome data.

Definition of end points

PFS was defined as the length of time between randomization or start of the intervention and the first evidence of clinical disease progression or death from any cause. OS was defined as the length of time between randomization or start of the intervention and death from any cause. The ORR was defined according to the response criteria used in each trial and generally represented the proportion of participants achieving a complete or partial response.

Overall surrogacy of PFS and ORR to OS

To assess the surrogacy of PFS to OS, we used the Pearson correlation coefficient (r) to quantify the association between the HRs for PFS and OS. STE analysis was conducted by transforming HRs and their CIs to the natural log scale, then fitting a weighted least‐squares linear regression model using OS log‐HRs as the dependent variable, PFS log‐HRs as the independent variable, and weighted based on the inverse variance of the OS log‐HR to assign higher weight to trials with more precise estimates. STE was estimated based on the PFS HR in which the upper bound of the 95% CI for the OS HR crossed 1.0.

To assess the surrogacy of ORR to OS, we used r to quantify the association between the ORR ORs and OS HRs. STE analysis was conducted by transforming HRs and their CIs to the natural log scale, then fitting a weighted least‐squares linear regression model using OS log‐HRs as the dependent variable, ORR log‐ORs as the independent variable, and weighted based on the inverse variance of the OS log‐HR to assign higher weight to trials with more precise estimates. STE was estimated based on the ORR OR in which the upper bound of the 95% CI for the OS HR crossed 1.0.

Risk of bias and publication bias

Risk of bias was assessed using the Cochrane RoB 2.0 tool (The Cochrane Collaboration) across five domains: bias arising from the randomization process, bias because of deviations from intended interventions, bias because of missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. An overall risk‐of‐bias judgment was also assigned for each study. To assess for publication biases, funnel plots were visually inspected and assessed using the Egger test for asymmetry.

Subgroup analyses

Subgroup analyses were performed using the same weighted least‐squares framework as the primary analysis stratified by the following subdomains: treatment class (VEGF inhibitors, mTOR inhibitors, immunotherapy, or combination therapy), line of treatment (first line or second line), and RCC histology (clear cell, non‐clear cell, or mixed histology). In some subgroups, particularly those with small sample sizes or unstable subgroup‐specific regression fits, no finite solution to this threshold equation could be identified. In those cases, the STE was reported as not estimable, with no STE CI shown.

Protocol and registration

This systematic review was registered with PROSPERO (registration number CRD420251076709).

RESULTS

Search results

We identified 4870 records through database searches, of which 3574 abstracts were screened, and 420 full‐text articles were assessed for eligibility (Figure 1). We included 62 RCTs comprising 66 treatment–control comparisons involving 24,518 participants with a median follow‐up of 29 months. ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷

FIGURE 1.

PRISMA diagram. PRISMA indicates Preferred Reporting Items for Systematic Reviews and Meta‐Analyses; RCC, renal cell carcinoma; RCT, randomized controlled trial.

Study characteristics

Among the 62 RCTs, 56 trials comprising 58 treatment–control comparisons reported both OS and PFS end points, and 59 trials comprising 63 treatment–control comparisons reported both OS and ORR end points. There were 35 phase 3 trials, 23 phase 2 trials, one phase 1 trial, and three that were unspecified. Two studies were published in 1990–1999, 11 were published in 2000–2009, 23 were published in 2010–2019, and 26 were published in 2020–2025. Individual study sample sizes ranged from 10 to 1110 participants.

Most studies compared the experimental arm versus the standard of care, with only three studies comparing the experimental arm versus placebo. Of note, five studies involved locoregional therapy, such as nephrectomy, stereotactic radiotherapy, and cryoablation. Of the 66 treatment–control comparisons, 25 assessed VEGF inhibitors, 11 assessed mTOR inhibitors, eight assessed contemporary immunotherapy, nine assessed cytokine therapy, and 13 assessed combination therapy involving immunotherapy and targeted agents. Common interventions included VEGF inhibitors, such as sunitinib, axitinib, bevacizumab, and cabozantinib; mTOR inhibitors, such as everolimus and temsirolimus; contemporary immunotherapy agents, such as nivolumab and pembrolizumab; and cytokine therapy agents, such as interferon alpha, and interleukin‐2.

Thirty‐eight trials assessed an intervention in the first‐line setting, while 24 trials assessed an intervention in the second‐line setting. Forty‐four trials assessed participants with clear cell RCC, six trials assessed participants with non‐clear cell RCC, eight trials assessed participants with both, while four trials did not specify histologic subtypes. Additional study characteristics are provided found in Tables 1 and 2.

TABLE 1.

Study characteristics of included trials for the surrogacy analysis of progression‐free survival and overall survival.

Abbreviations: 5‐FU, 5 fluorouracil; CI, confidence interval; ECOG PS, Eastern Cooperative Oncology Group performance status; EORTC, European Organization for the Research and Treatment of Cancer; EudraCT, European Union Clinical Trials Register; HR, hazard ratio; IFN‐α, interferon alpha; IMDC, International Metastatic Renal Cell Carcinoma Database Consortium; IQR, interquartile range; ISRCTN, International Standard Randomized Controlled Trial Number (UK); MRC, Medical Research Council; MSKCC, Memorial Sloan Kettering Cancer Center; mTOR, mammalian target of rapamycin; NCT, ClinicalTrials.gov identifier (US); NE, not estimable; NR, not reached; NS, not specified; SWOG, Southwest Oncology Group; UMIN, University Hospital Medical Information Network (Japan); VEGF, vascular endothelial growth factor.

TABLE 2.

Study characteristics of included trials for the surrogacy analysis of the objective response rate and overall survival.

Abbreviations: 5‐FU, 5 fluorouracil; CI, confidence interval; CR, complete response; ECOG PS, Eastern Cooperative Oncology Group performance status; EORTC, European Organization for the Research and Treatment of Cancer; EudraCT, European Union Clinical Trials Register; HR, hazard ratio; IFN‐α, interferon alpha; IMDC, International Metastatic Renal Cell Carcinoma Database Consortium; IQR, interquartile range; ISRCTN, International Standard Randomized Controlled Trial Number (UK); MRC, Medical Research Council; MSKCC, Memorial Sloan Kettering Cancer Center; mTOR, mammalian target of rapamycin; NCT, ClinicalTrials.gov identifier (US); NE, not estimable; NR, not reached; NS, not specified; ORR, objective response rate; OS, overall survival; PD, progressive disease; PR, partial response; SWOG, Southwest Oncology Group; UMIN, University Hospital Medical Information Network (Japan); VEGF, vascular endothelial growth factor.

Overall surrogacy of PFS and ORR to OS

The correlation between PFS and OS demonstrated a moderate association (r = 0.52; 95% CI, 0.38–0.66), and the STE was estimated at an HR of 0.92 that crossed unity (95% CI, 0.79–1.09; Figure 2). The correlation between ORR and OS demonstrated a moderate association (r = −0.59; 95% CI, −0.72, −0.45), and the STE was estimated at an OR of 1.40 that crossed unity (95% CI, 0.99–1.89; Figure 3).

FIGURE 2.

Scatterplot of log‐transformed HR of PFS and log‐transformed HR of OS. HR indicates hazard ratio; OR, odds ratio; ORR, objective response rate; PFS, progression‐free survival; r, correlation coefficient; STE, surrogate threshold effect.

FIGURE 3.

Scatterplot of log‐transformed OR of ORR and log‐transformed HR of OS. HR indicates hazard ratio; OR, odds ratio; ORR, objective response rate; OS, overall survival; r, correlation coefficient; STE, surrogate threshold effect.

A sensitivity analysis was performed to exclude the five trials that involved locoregional therapies. In doing so, the correlation between PFS and OS similarly suggested a moderate association (r = 0.51; 95% CI, 0.35–0.67), and the STE was estimated at an HR of 0.92 that crossed unity (95% CI, 0.77–1.11). The correlation between ORR and OS similarly suggested a moderate association (r = −0.62; 95% CI, −0.74, −0.48), and the STE was estimated at an OR of 1.40 that crossed unity (95% CI, 0.99–1.88).

Risk of bias

Risk of bias concerns were common among the included trial results. Twenty‐two studies were rated as having a high risk of bias, 37 were rated as having some concerns, and three were rated as having low risk of bias (Figure 4).

FIGURE 4.

Risk‐of‐bias assessment.

Publication bias

For OS, funnel plots did not demonstrate asymmetry, and the Egger test was not significant (p = .06; see Figure S1). For PFS, funnel plots did not demonstrate asymmetry, and the Egger test was not significant (p = .14; see Figure S2). For ORR, funnel plots did not demonstrate asymmetry, and the Egger test was not significant (p = .29; see Figure S3). Overall, funnel plots and Egger tests did not identify clear asymmetry.

Surrogacy by treatment class

In the VEGF inhibitor subgroup, the correlation between PFS and OS demonstrated a moderate association (r = 0.59; 95% CI, 0.29–0.78), and the STE was estimated at an HR of 0.83 that crossed unity (95% CI, 0.68–1.03). The correlation between ORR and OS suggested a strong association (r = −0.68; 95% CI, −0.83, −0.51), and the STE was estimated at an OR of 1.76 with wide confidence intervals (95% CI, 1.12–2.73).

For mTOR inhibitors, the correlation between PFS and OS suggested an insignificant association (r = 0.81; 95% CI, −0.31, 0.94), and the STE was estimated at an HR of 0.19 (95% CI, 0.05–0.74). The correlation between ORR and OS demonstrated an insignificant association (r = −0.75; 95% CI, −0.96, 0.02), and the STE was not estimable.

Among studies that assessed immunotherapy agents, the correlation between PFS and OS demonstrated a strong association (r = 0.70; 95% CI, 0.54–0.91), and the STE was estimated at an HR of 1.01 that crossed unity (95% CI, 0.82–1.37). The correlation between ORR and OS suggested an insignificant association (r = −0.36; 95% CI, −0.74, 0.20), and the STE was not estimable. Among immunotherapy trials that assessed contemporary agents such as ICIs rather than cytokine‐based therapies, the correlation between PFS and OS similarly suggested a strong association (r = 0.82; 95% CI, 0.63–1.00), and the STE was estimated at an HR of 1.07 that crossed unity (95% CI, 0.80–1.48). The correlation between ORR and OS similarly demonstrated an insignificant association (r = −0.41; 95% CI, −0.86, 0.48), and the STE was not estimable.

For combination regimens, the correlation between PFS and OS suggested a moderate association (r = 0.52; 95% CI, 0.12–0.77), and the STE was estimated at an HR of 0.67 (95% CI, 0.44–0.85). The correlation between ORR and OS demonstrated a moderate association (r = −0.56; 95% CI, −0.82, −0.30), and the STE was estimated at an OR of 2.53 with wide CIs (95% CI, 1.44–4.61).

Surrogacy by line of treatment

In the first‐line treatment subgroup, the correlation between PFS and OS demonstrated a moderate association (r = 0.50; 95% CI, 0.31–0.66), and the STE was estimated at an HR of 0.94 that crossed unity (95% CI, 0.80–1.16). The correlation between ORR and OS suggested a moderate association (r = −0.46; 95% CI, −0.64, −0.28), and the STE was estimated at an OR of 1.32 that crossed unity (95% CI, 0.83–2.01).

For second‐line treatments, the correlation between PFS and OS suggested a moderate association (r = 0.60; 95% CI, 0.29–0.83), and the STE was estimated at an HR of 0.67 (95% CI, 0.47–0.89). The correlation between ORR and OS demonstrated a strong association (r = −0.80; 95% CI, −0.89, −0.69), and the STE was estimated at an OR of 2.23 with wide CIs (95% CI, 1.57–3.33).

Surrogacy by histology

In the clear cell RCC subgroup, the correlation between PFS and OS demonstrated a moderate association (r = 0.46; 95% CI, 0.28–0.64), and the STE was estimated at an HR of 0.93 that crossed unity (95% CI, 0.78–1.15). The correlation between ORR and OS suggested a moderate association (r = −0.53; 95% CI, −0.69, −0.34), and the STE was estimated at an OR of 1.40 that crossed unity (95% CI, 0.85–2.10).

For non–clear cell RCC, the correlation between PFS and OS suggested an insignificant association (r = 0.68; 95% CI, −1.00, 1.00), and the STE was not estimable. The correlation between ORR and OS demonstrated a moderate association (r = −0.58; 95% CI, −0.94, −0.02), and the STE was not estimable.

Among studies that assessed mixed histology, the correlation between PFS and OS demonstrated an insignificant association (r = 0.33; 95% CI, −0.94, 0.97), and the STE was not estimable. The correlation between ORR and OS suggested an insignificant association (r = −0.48; 95% CI, −0.98, 0.61), and the STE was not estimable.

DISCUSSION

Overall surrogacy of PFS and ORR to OS

Across all included studies, PFS and ORR demonstrated moderate correlations with OS. These findings are consistent with prior meta‐analyses in mRCC, including works by Delea et al. and Heng et al., who reported similar moderate associations between treatment effects on PFS and OS in the targeted therapy era. ⁶⁸ , ⁶⁹ Likewise, analyses by Bria et al. and Johnson et al. demonstrated that, although PFS correlates with OS at the trial level, the strength of this relation is insufficient to establish it as a universally reliable surrogate end point. ⁷⁰ , ⁷¹

Crucially, the STEs for PFS and ORR both crossed unity, indicating that no consistent magnitude of improvement in either end point reliably predicted OS benefit. This aligns with broader oncology literature emphasizing that correlation alone is insufficient to validate surrogacy, and that clinically actionable thresholds are often lacking. ⁷² These findings are also consistent with analyses in other diseases, such as those in non–small cell lung cancer and melanoma, in which surrogate end points have shown variable and context‐dependent performance. ⁷³ , ⁷⁴

Surrogacy by treatment class

Surrogate end point performance varied meaningfully by therapeutic mechanism. In VEGF‐directed and mTOR‐directed therapies, PFS appeared to retain some pragmatic value as an intermediate end point, likely reflecting more direct and temporally aligned effects on tumor growth kinetics. ⁶⁸ , ⁷⁰ However, the absence of consistent STE‐defined thresholds indicates that these relations remain imperfect and should not be interpreted as reliable surrogates for OS.

In contrast, although trial‐level PFS and OS effects were strongly correlated in immunotherapy‐containing trials, threshold estimates were imprecise and did not identify a stable, clinically useful PFS effect capable of reliably predicting OS benefit. The ORR showed a weaker and less precise association with OS. This is consistent with established patterns of delayed response and durable disease control, which are not adequately captured by conventional end points such as PFS and ORR. ⁷⁴ , ⁷⁵ These findings suggest that surrogate end points may have some limitations in immunotherapy settings. Combination regimens demonstrated intermediate performance, reflecting the integration of cytoreductive and immunomodulatory mechanisms, although without establishing consistent or robust surrogacy.

Surrogacy by line of treatment

Numerically stronger associations were observed among second‐line comparisons, although these estimates cannot be attributed to the line of therapy because treatment class, trial era, and subsequent therapy patterns differed substantially between subgroups. Prior studies have suggested that surrogate end points do not consistently perform better in later lines of therapy and may demonstrate weaker or less reliable associations with OS. ⁷⁰

Surrogacy by histology

Histologic subtype had limited influence on surrogate performance, largely because of the predominance of clear cell histology and the relative paucity of data in non–clear cell populations. Although moderate associations were observed in clear cell cohorts, the lack of consistent STE‐defined thresholds limits their clinical interpretability. In non–clear cell analyses, estimates were unstable and frequently nonestimable, reflecting small sample sizes and heterogeneity, consistent with prior reviews emphasizing the challenges of studying this population. ⁷⁶ These findings highlight the need for histology‐specific validation, particularly in underrepresented subgroups.

Overall implications

These findings are consistent with prior meta‐analyses in mRCC, which have demonstrated that, at best, PFS is a moderate surrogate for OS and that ORR is generally less reliable. ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ It is noteworthy that these findings highlight the limitation of relying on correlation alone to suggest surrogate validity. Although moderate associations between surrogate end points and OS were observed, STE analysis demonstrated that clinically meaningful and reliable prediction of survival benefit is often not achieved. This underscores the critical distinction between statistical association and true clinical surrogacy.

From a clinical and regulatory perspective, these results support a more cautious and context‐specific approach to end point interpretation. Improvements in PFS or ORR should not be assumed to translate into survival benefit, particularly in the absence of robust, threshold‐based validation. As surrogate end points continue to inform trial design and drug approval pathways, greater emphasis on rigorous validation frameworks will be essential to ensure that observed treatment effects reflect meaningful clinical benefit.

LIMITATIONS

Several limitations should be considered. First, this analysis was conducted at the trial‐level without individual patient‐level validation, which may introduce ecological bias and limit the ability to account for patient‐level heterogeneity. Second, crossover between treatment groups in some included trials may have attenuated differences in OS by allowing patients to receive subsequent effective therapies, potentially diluting the survival benefit associated with the initial intervention. This may have weakened the observed association between PFS or ORR and OS and thus could have led to an underestimation of their validity as surrogate end points. Third, the weighted regression accounted for precision in the OS treatment effect but did not explicitly model estimation error in the surrogate treatment effect or the covariance between end point estimates, which may have influenced the estimated surrogacy relation. Fourth, the inclusion of trials evaluating locoregional therapies may have introduced heterogeneity in the biologic pathways linking treatment effects to OS because interventions such as cytoreductive nephrectomy or stereotactic radiotherapy may influence survival through mechanisms not fully captured by RECIST (Response Evaluation Criteria in Solid Tumor)‐based end points. Therefore, an OS benefit could plausibly occur without a proportionate effect on PFS or ORR (or vice versa), which could alter the observed surrogate relation. Fifth, it should be noted that many second‐line mRCC trials have historically failed to demonstrate statistically significant OS benefits despite improvements in PFS and/or ORR. Although this was not reflected in our analysis, we acknowledge that our findings may reflect the greater representation of VEGF and mTOR inhibitors in later lines, as opposed to immunotherapy‐based regimens. Sixth, some subgroup analyses—particularly those involving mTOR inhibitors, immunotherapy regimens, and non–clear cell histology—were limited by small sample sizes, wide CIs, and instability in STE estimation, reducing precision and interpretability. Finally, the potential for publication bias and selective reporting cannot be excluded, particularly in analyses demonstrating stronger surrogate associations.

CONCLUSION

PFS and ORR demonstrate moderate trial‐level associations with OS in patients with mRCC but lack consistent, clinically meaningful surrogacy when evaluated using STEs. Surrogate validity is highly context‐dependent, varying by treatment class, line of therapy, and histology. These findings underscore the need for cautious interpretation of surrogate end points. Furthermore, given the substantial heterogeneity of the included trials and treatment strategies, these findings should be interpreted cautiously and may not be fully generalizable to contemporary mRCC practice. OS remains the most reliable measure of clinical benefit, and continued efforts are needed to identify and validate more robust surrogate measures in mRCC.

AUTHOR CONTRIBUTIONS

Russell Leong: Conceptualization; methodology; data curation; investigation; formal analysis; writing—original draft. David Chen: Conceptualization; methodology; software; investigation; visualization; project administration; writing—review and editing. Laith Almasri: Data curation; writing—review and editing. Nicholas Lum: Data curation; writing—review and editing. Karren Xiao: Data curation; writing—review and editing. Maryam Soleimani: Writing—review and editing. Christian K. Kollmannsberger: Writing—review and editing. Srinivas Raman: Conceptualization; methodology; investigation; supervision; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

Maryam Soleimani reports personal/consulting fees from Ipsen, Merck, and Novartis; support for other professional activities from Bayer and Ferring; and travel support from Pfizer outside the submitted work. Christian K. Kollmannsberger reports personal/consulting fees from Novartis Pharmaceuticals Canada and support for other professional activities from Advanced Accelerator Applications, Astellas Pharma Canada, Bayer Healthcare, BioNTech US Inc., Bristol Myers Squibb Canada, Eisai, Ipsen Biopharmaceuticals Inc., Janssen Pharmaceuticals, Merck, Pfizer, PFIZER CANADA INC., and Seagen Inc. outside the submitted work. Srinivas Raman reports grants/contracts from AstraZeneca and Varian Medical Systems outside the submitted work. The remaining authors disclosed no conflicts of interest.

Supporting information

Supporting Information S1

Figure S1

Figure S2

Figure S3

Table S1