Pembrolizumab plus axitinib versus sunitinib for advanced clear cell renal cell carcinoma: 5-year survival and biomarker analyses of the phase 3 KEYNOTE-426 trial

Name: Verified Rigor Record — Pembrolizumab plus axitinib versus sunitinib for advanced clear cell renal cell carcinoma: 5-year survival and biomarker analyses of the phase 3 KEYNOTE-426 trial
Item: Pembrolizumab plus axitinib versus sunitinib for advanced clear cell renal cell carcinoma: 5-year survival and biomarker analyses of the phase 3 KEYNOTE-426 trial
Rating: 92
Author: Αlpha¹

Rini BI, Plimack ER, Stus V, Gafanov R, Waddell T, Nosov D, Pouliot F, Alekseev B, Soulières D, Melichar B, Vynnychenko I, de Azevedo SJ, Borchiellini D, McDermott RS, Bedke J, Tamada S, Wu S, Markensohn J, Zhang Y, Loboda A, Vajdi A, Perini RF, Burgents J, Powles T.

Paper source

Open DOI

Αlpha¹ Rigor Review

Rigor Review Verified

Critical

High-priority

Action items to resolve · Critical / High

Confidence: highEngine: 7.6.0Source: full text

This Alpha1 Rigor Review uses AI Rigor Reviewers trained on a curated corpus of high-fidelity and retracted papers, with expert supervision and curation. It can still make mistakes; verify each finding against the source before relying on it.

The manuscript reports a long-term follow-up and biomarker analysis from a phase 3 RCT with a strong scientific premise and generally transparent reporting. The main weaknesses are incomplete reporting of study design details (randomization method, power analysis, outlier handling), an unnamed ethics committee, missing manufacturer identification for key drugs, absent effect sizes for continuous biomarker associations, and no data/code sharing for generated sequencing data.

Three independent reviewers (all deepseek-v4-flash) evaluated the full manuscript text. The verification component's statistics check recomputed 5 reported tests consistently; the remaining statistical results (the majority of biomarker p-values) were not machine-verified. 5 references could not be found in Crossref/OpenAlex — they should be re-checked; none were retracted. The copyedit pass identified 11 issues (1 major: a label inconsistency in Extended Data Table 3).

12 major claims checked against the paper's own evidence: all adequately supported.

partialResults, 'Efficacy estimates by DNA mutational status'Reviewer 3
PBRM1 mutation is positively associated with ORR within the pembrolizumab+axitinib arm.
Extended Data Table 3 and Fig. 3b show PBRM1 mutation associated with higher ORR (71.4% vs 52.3%, multiplicity-adjusted P=0.002 within the pembro+axi arm). However, the sample size is small and the association may be exploratory; the paper itself notes the totality of data is conflicting.
Evidence: Extended Data Fig. 3b and Extended Data Table 3.
“The PBRM1 mutation was positively associated with the ORR within the pembrolizumab plus axitinib arm, with significantly higher rates in the PBRM1 mutant than in the wild-type subgroup (71.4% versus 52.3%; P = 0.002)”
supportedAbstractReviewer 1
Pembrolizumab plus axitinib shows sustained OS, PFS, and ORR benefit over sunitinib after 5 years of follow-up.
The claim is directly supported by the reported HRs, CIs, and ORR percentages, which all favor the combination arm.
Evidence: Abstract and Results: OS HR 0.84 (0.71-0.99), PFS HR 0.69 (0.59-0.81), ORR 60.6% vs 39.6%.
“Pembrolizumab plus axitinib showed sustained benefits in OS (hazard ratio: 0.84; 95% confidence interval: 0.71–0.99), PFS (hazard ratio: 0.69; 95% confidence interval: 0.59–0.81) and ORR (60.6% versus 39.6%) compared to sunitinib.”
supportedAbstractReviewer 1
An 18-gene T-cell-inflamed GEP is positively associated with OS, PFS, and ORR in the pembrolizumab plus axitinib arm.
Table 2 shows statistically significant p-values for all three outcomes in the combination arm.
Evidence: Table 2: Tcell inf GEP with ORR p=2.03×10^-6, PFS p=1.41×10^-5, OS p=0.002.
“An 18-gene T-cell-inflamed gene expression profile (Tcell inf GEP) was positively associated with OS (P = 0.002), PFS (P < 0.0001) and ORR (P < 0.0001) within the pembrolizumab plus axitinib arm.”
supportedAbstractReviewer 1
An angiogenesis signature is positively associated with OS in the pembrolizumab plus axitinib arm and with OS, PFS, and ORR in the sunitinib arm.
Table 2 confirms these associations with sufficient statistical significance.
Evidence: Table 2: Angiogenesis signature in pembrolizumab+axitinib arm: OS p=0.004; in sunitinib arm: ORR p=0.002, PFS p=5.66×10^-4, OS p=1.69×10^-7.
“An angiogenesis signature was positively associated with OS (P = 0.004) within the pembrolizumab plus axitinib arm and with OS (P < 0.0001), PFS (P < 0.001) and ORR (P = 0.002) within the sunitinib arm.”
supportedAbstractReviewer 1
PBRM1 mutation is positively associated with ORR in the pembrolizumab plus axitinib arm.
Extended Data Table 3 shows a statistically significant positive association (p=0.002) after multiplicity adjustment.
Evidence: Extended Data Table 3: ORR p=0.002 with '+' direction.
“PBRM1 (polybromo-1) mutation had a positive association with ORR (P = 0.002) within the pembrolizumab plus axitinib arm.”
supportedAbstractReviewer 1
Within the sunitinib arm, OS is positively associated with VHL and PBRM1 mutations and negatively associated with BAP1 mutation.
Extended Data Table 3 reports significant p-values for these associations after multiplicity adjustment.
Evidence: Extended Data Table 3: VHL OS p=0.040, PBRM1 OS p=0.010, BAP1 OS p=0.019 (negative association).
“Within the sunitinib arm, OS was positively associated with VHL (von Hippel–Lindau tumor suppressor gene) (P = 0.040) and PBRM1 (P = 0.010) mutations and was negatively associated with BAP1 (BRCA1-associated protein 1) mutation (P = 0.019).”
supportedAbstract, ResultsReviewer 2
Pembrolizumab plus axitinib shows sustained OS benefit over sunitinib at 5 years.
The claim is supported by the reported HR of 0.84 (95% CI 0.71-0.99) and Kaplan-Meier estimates.
Evidence: HR 0.84, 95% CI 0.71-0.99, median OS 47.2 vs 40.8 months.
“Pembrolizumab plus axitinib showed sustained benefits in OS (hazard ratio: 0.84; 95% confidence interval: 0.71–0.99)”
supportedResults, Biomarker outcomes, Table 2Reviewer 2
T-cell-inflamed GEP is positively associated with OS, PFS, and ORR within the pembrolizumab plus axitinib arm.
The claim is supported by the reported p-values from logistic and Cox regression models adjusted for IMDC risk.
Evidence: Table 2: Tcell inf GEP p-values: ORR 2.03e-6, PFS 1.41e-5, OS 0.002.
“In the pembrolizumab plus axitinib arm, higher Tcell inf GEP was associated with improved ORR ( P < 0.0001), PFS ( P < 0.0001) and OS ( P = 0.002)”
supportedResults, Biomarker outcomes, Table 2Reviewer 2
Angiogenesis signature is positively associated with OS within the pembrolizumab plus axitinib arm and with OS, PFS, and ORR within the sunitinib arm.
The claim is supported by the reported p-values in Table 2.
Evidence: Table 2: Angiogenesis p-values: pembrolizumab+axitinib OS 0.004; sunitinib ORR 0.002, PFS 5.66e-4, OS 1.69e-7.
“For the angiogenesis signature, a positive association was observed only with OS ( P = 0.004) in the pembrolizumab plus axitinib arm, and there was a positive association with ORR ( P = 0.002), PFS ( P < 0.001) and OS ( P < 0.001) in the sunitinib arm”
supportedResults, Biomarker outcomes, Table 2Reviewer 2
PD-L1 CPS is not a predictive marker for pembrolizumab plus axitinib outcomes.
The claim is supported by the lack of significant association (p>0.05) for all outcomes in the pembrolizumab+axitinib arm.
Evidence: Table 2: PD-L1 CPS p-values for pembrolizumab+axitinib: ORR 0.053, PFS 0.168, OS 0.544.
“No association was observed between continuous PD-L1 CPS and clinical outcomes for pembrolizumab plus axitinib ( P > 0.05)”
supportedResults, Efficacy estimates by DNA mutational statusReviewer 2
PBRM1 mutation is positively associated with ORR within the pembrolizumab plus axitinib arm.
The claim is supported by the reported p-value and response rates.
Evidence: Extended Data Table 3: PBRM1 mutation ORR p=0.002; ORR 71.4% vs 52.3%.
“The PBRM1 mutation was positively associated with the ORR within the pembrolizumab plus axitinib arm, with significantly higher rates in the PBRM1 mutant than in the wild-type subgroup (71.4% versus 52.3%; P = 0.002)”
supportedAbstract and Results, 'Efficacy outcomes'Reviewer 3
Pembrolizumab plus axitinib shows sustained overall survival benefit compared to sunitinib after ≥5 years of follow-up.
The paper provides HR=0.84; 95% CI 0.71–0.99 from the intention-to-treat analysis, with consistent results across subgroups. The CI marginally excludes 1.00, supporting the claim.
Evidence: Results, Fig. 2a: HR=0.84, 95% CI 0.71–0.99.
“Pembrolizumab plus axitinib showed sustained benefits in OS (hazard ratio: 0.84; 95% confidence interval: 0.71–0.99)”

Recomputed 5 tests: 5 consistent, 0 inconsistent, 5 via agent-written checks.

2 integrity concerns flagged (0 high).

Rigor

Eight Dimensions6 pass · 2 warn

12 major claims1 partial11 supported

11 copyedit issues flagged (1 major): mostly typo, consistency, punctuation.

Checked 49 references: 42 verified — 5 unresolved, 2 not checked.

Reproducibility

3 data/code links checked; 3 live.

datahttps://externaldatasharing-msd.com/LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
datahttps://clinicaltrials.gov/study/NCT02853331LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
datahttps://clinicaltrials.gov/ct2/show/NCT02853331LIVEHTTP 200
Resolves, but the content could not be matched to the paper.

Registered (1 ID: ClinicalTrials.gov). Reporting guideline cited: CONSORT.

Alpha¹ Rigor Action ItemsFinal pass before journal submission

Ready after high-priority edits. The manuscript is scientifically sound and well-written overall, but must address: the mislabel in Extended Data Table 3 footnote ('axitinib arm' should be 'sunitinib arm'), the missing effect sizes and confidence intervals for continuous biomarker associations, the unnamed ethics committee, and the absence of public repository deposits for sequencing data. These are all fixable before submission and would be flagged by any competent peer review.

1.
HIGHother
Correction: in Extended Data Table 3 footnote, change 'axitinib arm' to 'sunitinib arm' to match the table header.
This is a factual inconsistency — the table describes sunitinib-arm results, not axitinib — that reviewers and editors will catch and question.
2.
HIGHreporting
In the Methods (Inclusion and ethics), name a specific IRB/ethics committee (e.g., 'the trial was approved by Western Institutional Review Board, protocol number XXX') rather than 'appropriate institutional review boards'.
A generic statement without a named body is considered insufficient by most journals and all three reviewers flagged this as missing.
3.
HIGHdata code
Deposit RNA-seq and whole-exome sequencing data in a controlled-access repository (e.g., dbGaP, EGA, GEO) and add accession numbers to the Data availability section.
These data are the basis for the biomarker analyses; without public deposition, the analyses cannot be independently verified — a core reproducibility requirement.
4.
HIGHstatistics
In Table 2 and the main text, report effect sizes (e.g., hazard ratios or odds ratios per 1-SD increase) with 95% confidence intervals for all continuous biomarker associations, not only p-values and direction signs.
P-values alone do not convey magnitude or precision; effect sizes with CIs are expected for all regression-based associations and are missing from the biomarker results.
5.
HIGHdata code
Share custom analysis code (OmicSoft pipeline scripts, R scripts for biomarker models) in a public repository (GitHub, Zenodo) with a persistent DOI, and add the link to the Data availability section.
No custom code is currently shared, preventing full reproducibility of the biomarker analyses.
6.
HIGHreporting
In the Methods or Acknowledgements, explicitly cite the CONSORT 2010 statement (or STROBE for the biomarker component) and state that reporting follows this guideline.
Although a CONSORT flow diagram is present, the formal guideline citation is missing, which is a standard expectation for clinical trial papers.
7.
HIGHstatistics
Add a brief statement in the Methods (Statistical analysis) verifying the proportional hazards assumption for Cox models (e.g., 'The proportional hazards assumption was assessed using Schoenfeld residuals and was satisfied').
Assumption verification is not currently reported; this is a standard reviewer request for time-to-event analyses.
8.
HIGHstatistics
Add an a priori power analysis or sample-size justification for the exploratory biomarker analyses, or explicitly state that the sample size was determined by the parent trial and that the analysis is post hoc/exploratory with no formal power.
No power analysis is provided; reviewers and readers will want to know whether the study had adequate statistical power for the biomarker comparisons, especially for subgroups.
9.
HIGHstatistics
Describe how outliers in biomarker data (e.g., gene expression values, mutation calls) were identified and handled in the Methods (Statistical analysis or Assessments).
Outlier handling is not mentioned anywhere; this is a standard methodological detail expected for genomic analyses.
10.
HIGHreporting
Specify the random sequence generation method (e.g., 'computer-generated random numbers with permuted blocks and a 1:1 allocation ratio, stratified by ...') in the Methods.
The method of randomization is not described beyond stating 'randomly assigned in a 1:1 ratio', which is insufficient for a trial report.
11.
MEDIUMreporting
Provide a brief rationale for the open-label design in the Methods (e.g., 'Given the known toxicity profiles of TKIs and the requirement for dose modifications, blinding was not feasible.')
The open-label design is stated but not justified; explicit acknowledgment of why blinding was not used strengthens design transparency.
12.
MEDIUMreporting
In the Methods (Trial design, participants and treatments), explicitly state the manufacturer/source (company, city, country) for axitinib and sunitinib.
Investigational drugs are core reagents; their source should be identified for replicability, as is standard for clinical trial reports.
13.
MEDIUMreporting
In Table 1 or Methods, include baseline weight (mean ± SD or median with range) of participants to complete the age/weight/health status reporting.
Weight is a standard demographic variable in clinical trials and its absence is a minor reporting gap.
14.
MEDIUMreporting
In the Data availability section, specify the expected timeframe for data access upon request (e.g., 'Data will be shared within 6 months of an approved request').
The current managed-access statement does not indicate how long data access takes, which would help prospective users.
15.
LOWcopyedit
Across the Discussion and Extended Data figures, correct 'pressent' to 'present' and 'PRRM1' to 'PBRM1'; correct 'von Lindau-Hippel' to 'von Hippel-Lindau' in figure legends and table footnotes.
These are typographical errors that affect readability and accuracy of gene/entity names.
16.
LOWcopyedit
In Table 2 footnote, use consistent ASCII plus/minus signs or matched Unicode symbols for '+' and '−' indicators without mixing quote styles.
Minor punctuation inconsistency that a copyeditor would flag.
17.
LOWstatistics
In the Discussion, explicitly acknowledge that the lack of a pre-registered analysis plan for the biomarker component (beyond the parent trial SAP) may affect reproducibility, and consider registering the plan retrospectively.
Weakens the strength of the exploratory claims; pre-registration would strengthen the biomarker findings.
18.
LOWdata code
Verify the 5 references flagged as 'not_found_in_registry' (CheckMate 9ER biomarker analysis, KIM-1 JAVELIN, KEYNOTE-427, TKIs immunomodulatory effects, CheckMate 9ER final follow-up) as they may be mis-cited, preprints, or conference abstracts — correct or add missing identifiers.
These references could not be located in standard registries; while not retracted, their absence is a potential integrity concern that should be resolved before submission.

Full text (72,141 chars)Analyzed: Jul 6, 2026

Αlpha¹ assesses methodological rigor, not the importance of the findings. See how we calibrate →