Effect of adjuvant carboplatin intensified chemotherapy versus standard chemotherapy on survival in women with high risk, early stage, triple negative breast cancer (CITRINE): randomised, open label phase 3 trial.
Liu Y, Gong Y, Zhu XZ, Liu GY, Yu KD, Yang F, Chen L, He M, Hu Z, Chen CM, Cao AY, Li JJ, Hou YF, Di GH, Wu J, Jiang YZ, Fan L, Wang ZH, Shao ZM
- DOI
- 10.1136/bmj-2025-085457
- Record issued
- 2026-08-15
- Engine
- 7.39.0
- Exported
- 2026-09-20
Prepared by Alpha1. This document is confidential: it is intended for the recipient it was shared with and must not be redistributed. The live record at alpha1science.com/verify/aba13e07-b54c-4a39-b4bf-3103322b1b03 is authoritative.
How this rating was calculated
- IntegrityIntegrity concern ×2−1★
This Kaimen Rigor review uses Kaimen 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.
This is a well-conducted and transparently reported phase 3 randomized trial. The methodology is rigorous, with proper randomization, blinding of assessors, pre-specified power analysis, and appropriate statistical handling of non-proportional hazards. Minor reporting gaps (e.g., missing CONSORT flow diagram details, exact p-values for piecewise analyses) and a few copyedit issues do not undermine the core findings.
Both reviewers classified the study as interventional, and this was adopted. The evaluation covered the full text, including methods, results, and supplementary materials. Non-applicable sub-criteria (e.g., animal housing, cell line authentication) were excluded. The statistics verification covered only a subset of reported tests; the absence of detected errors does not confirm overall statistical correctness.
Numerical inconsistencies
1 finding · worst lowValues that contradict each other or are impossible for the stated sample: recomputed p-values and test statistics, GRIM/GRIMMER checks on summary numbers, percentages against their own counts, totals against their parts, and estimates against their own confidence intervals.
- Internal contradictions in the reported numbersAssessed
Recomputed 7 tests: 7 consistent, 0 inconsistent; 2 recomputed directly from the reported test statistics, 5 via agent-written checks.
- CONSISTENTreported p = .030 · recomputed p = .027Recomputed hazard ratio 0.64 (95% CI 0.43–0.95), reported p=0.03
“hazard ratio 0.64, 95% confidence interval (CI) 0.43 to 0.95; P=0.03”
Taken as given: 0.43–0.95 is a two-sided 95% confidence interval for the hazard ratio of 0.64, not a range, an IQR, or a different interval level; the hazard ratio is a RATIO measure, so the interval is symmetric on the log scale; p=0.03 is the p for THIS estimate, not for another comparison in the same sentenceMethod: back the two-tailed p out of the log-scale CI width and compare it against the printed pHow we recomputed it: pCI(0.64, 0.43, 0.95, 1) - CONSISTENTreported p = .020 · recomputed p = .025Recomputed hazard ratio 0.59 (95% CI 0.37–0.93), reported p=0.02
“hazard ratio 0.59, 95% CI 0.37 to 0.93; P=0.02”
Taken as given: 0.37–0.93 is a two-sided 95% confidence interval for the hazard ratio of 0.59, not a range, an IQR, or a different interval level; the hazard ratio is a RATIO measure, so the interval is symmetric on the log scale; p=0.02 is the p for THIS estimate, not for another comparison in the same sentenceMethod: back the two-tailed p out of the log-scale CI width and compare it against the printed pHow we recomputed it: pCI(0.59, 0.37, 0.93, 1) - CONSISTENTreported p = .040 · recomputed p = .047Reviewers 1, 2Secondary endpoint distant disease-free survival hazard ratio p-value from CI
“hazard ratio 0.61, 0.37 to 0.98; P=0.04”
Taken as given: The hazard ratio is 0.61 with 95% CI 0.37 to 0.98.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed two-sided p-value from the hazard ratio and its 95% CI using the normal approximation for the log hazard ratio.How we recomputed it: pCI(0.61, 0.37, 0.98, 1) - CONSISTENTreported p = .010 · recomputed p = .014Reviewers 1, 2Secondary endpoint overall survival hazard ratio p-value from CI
“hazard ratio 0.41, 0.20 to 0.83; P=0.01”
Taken as given: The hazard ratio is 0.41 with 95% CI 0.20 to 0.83.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed two-sided p-value from the hazard ratio and its 95% CI using the normal approximation for the log hazard ratio.How we recomputed it: pCI(0.41, 0.20, 0.83, 1) - CONSISTENTreported p = .030 · recomputed p = .030Reviewer 1Adjusted primary endpoint hazard ratio p-value from CI
“adjusted hazard ratio of 0.65 (95% CI 0.44 to 0.96; P=0.03)”
Taken as given: The adjusted hazard ratio is 0.65 with 95% CI 0.44 to 0.96.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed two-sided p-value from the hazard ratio and its 95% CI using the normal approximation for the log hazard ratio.How we recomputed it: pCI(0.65, 0.44, 0.96, 1) - CONSISTENTreported p = .006 · recomputed p = .006Reviewer 1Restricted mean survival time difference p-value from CI
“The difference in restricted mean survival time was 2.4 (95% CI 0.7 to 4.1; P=0.006) months”
Taken as given: The difference in restricted mean survival time is 2.4 months with 95% CI 0.7 to 4.1.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed two-sided p-value from the difference and its 95% CI using the normal approximation.How we recomputed it: pCI(2.4, 0.7, 4.1, 0) - CONSISTENTreported p = .006 · recomputed p = .007Reviewer 1Restricted mean time lost ratio p-value from CI
“the restricted mean time lost ratio was 0.59 (95% CI 0.40 to 0.86; P=0.006)”
Taken as given: The restricted mean time lost ratio is 0.59 with 95% CI 0.40 to 0.86.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed two-sided p-value from the ratio and its 95% CI using the normal approximation for the log ratio.How we recomputed it: pCI(0.59, 0.40, 0.86, 1)
- lowinternal contradictionThe abstract reports 808 enrolled and 807 received treatment, but the safety analysis uses 403 in the carboplatin arm and 404 in the control arm, which sums to 807. This is consistent, but the text says 'All 807 patients in the safety population experienced at least one treatment related adverse event' while Table 3 shows 404 and 403, which is consistent. No contradiction found.
Of the 808 enrolled patients, 807 received study treatment. ... All 807 patients in the safety population experienced at least one treatment related adverse event.
Abstractreviewer’s wording
Overstated conclusions
None foundConclusions that reach past what the paper's own results support — including a significance claim that no longer holds when the statistic is recomputed, and efficacy resting on an unvalidated surrogate endpoint.
Checked — nothing surfaced.
5 major claims checked against the paper's own evidence: all adequately supported.
- supportedReviewers 1, 2The addition of carboplatin to adjuvant anthracycline/taxane based chemotherapy significantly improved survival outcomes in patients with high risk, early stage, triple negative breast cancer.The primary endpoint (disease-free survival) showed a statistically significant improvement (HR 0.64, 95% CI 0.43-0.95, P=0.03), and secondary endpoints also showed significant improvements. The claim is supported by the presented data.Evidence: Primary endpoint disease-free survival HR 0.64 (95% CI 0.43-0.95, P=0.03); secondary endpoints RFS, DDFS, OS all significant.
“The addition of carboplatin to adjuvant anthracycline/taxane based chemotherapy significantly improved survival outcomes in patients with high risk, early stage, triple negative breast cancer, driven by reduction of early recurrence risk without new safety concerns.”
ConclusionFind in source - supportedReviewers 1, 2The benefit is driven by reduction of early recurrence risk.The piecewise hazard ratios show a strong early benefit (HR 0.31 for 0-12 months) that attenuates over time, supporting the claim that the benefit is driven by early recurrence reduction.Evidence: Piecewise hazard ratios: 0.31 (0-12 months), 0.65 (12-36 months), 1.98 (>36 months).
“driven by reduction of early recurrence risk”
ConclusionFind in source - supportedReviewers 1, 2No new safety concerns were identified.The safety profile showed expected higher hematologic toxicity with carboplatin, but no treatment-related deaths and no significant difference in serious adverse events. The claim is supported.Evidence: No treatment-related deaths; grade 3-4 AEs 66.7% vs 55.0%; SAEs 4.2% vs 3.2% (not significantly different).
“without new safety concerns”
ConclusionFind in source - supportedReviewer 1This study is the first to add carboplatin to anthracycline/taxane based adjuvant chemotherapy regimen for patients with high risk, early stage, triple negative breast cancer.The paper claims novelty, and the discussion compares with other trials (e.g., NRG-BR003) that have similar designs but were not yet reported. The claim is plausible and supported by the literature review.Evidence: Discussion states 'The CITRINE trial was designed to explore, for the first time, whether the addition of carboplatin to an anthracycline/taxane based adjuvant chemotherapy regimen could improve the prognosis...'
“The CITRINE trial was designed to explore, for the first time, whether the addition of carboplatin to an anthracycline/taxane based adjuvant chemotherapy regimen could improve the prognosis of patients with high risk, early stage, triple negative breast cancer.”
Discussion ¶1Find in source - supportedReviewers 1, 2The treatment effect was consistent across all subgroups.The interaction tests were not significant for any subgroup, supporting the claim of consistency, though the power to detect interactions is limited.Evidence: Subgroup analyses showed no significant interaction (e.g., stage interaction P=0.63, Ki-67 interaction P=0.75).
“the interaction test did not indicate significant heterogeneity, suggesting that the treatment effect was consistent across all subgroups”
ResultsFind in source
Efficacy claim is anchored to an adequate endpoint and a meaningful effect.
- ADEQUATESurrogate endpointThe primary endpoint is disease-free survival, a hard clinical outcome (time to recurrence, second primary, or death). No surrogate or biomarker is used as the primary basis for the efficacy claim.
“The primary endpoint was disease-free survival in the intention-to-treat population.”
- ADEQUATEEffect sizeThe effect is reported as a 6.5% absolute improvement in 3-year disease-free survival (92.3% vs 85.8%) with a hazard ratio of 0.64 (95% CI 0.43 to 0.95; P=0.03). The magnitude is anchored to a clinically meaningful outcome and supported by statistical significance.
“Estimated percentages of patients who would be disease-free at three years were 92.3% in the carboplatin arm and 85.8% in the control arm (hazard ratio 0.64, 95% confidence interval (CI) 0.43 to 0.95; P=0.03).”
Data authenticity concerns
1 finding · worst lowAn adversarial read for patterns associated with data that may not be genuine: results that look too clean, implausibly large effects, duplicated data or images, and methods that do not match the results reported.
- Other integrity concernAssessed
2 integrity concerns flagged (0 high).
- lowotherThe paper reports a statistically significant primary endpoint (P=0.03) but the proportional hazards assumption is violated, and the late hazard ratio (1.98) is in the opposite direction. This is appropriately discussed, but the headline conclusion of 'significantly improved survival' may be somewhat overstated given the non-proportional hazards.
“the hazard ratio was 0.31 (95% CI 0.13 to 0.73) for 0-12 months, 0.65 (0.39 to 1.09) for 12-36 months, and 1.98 (0.69 to 5.69) for >36 months.”
AbstractFind in source
Reporting gaps
None foundRequired detail the manuscript never states — study design, biological variables, ethics approval and consent, key resources, statistical reporting, data and code availability, and overall transparency.
Checked — nothing surfaced.
The paper cites multiple prior studies (ECOG E1199, CALGB 9741, GIM-2, EBCTCG meta-analysis) to establish the standard of care and the rationale for dose-dense regimens. It acknowledges the controversy regarding platinum agents' long-term survival benefit and the limited evidence in the adjuvant setting, directly motivating the trial. The hypothesis follows logically from the cited evidence.
“The Eastern Cooperative Oncology Group (ECOG) E1199 study optimised the administration of taxanes and suggested that weekly paclitaxel, after anthracyclines, improves disease-free survival and overall survival in patients with triple negative breast cancer.”
“Nevertheless, significant controversy remains regarding whether platinum agents can improve patients’ long term survival and the association between the efficacy of platinum agents and the germline BRCA mutation status.”
“In this context, we did a phase 3 trial (carboplatin intensified chemotherapy for triple negative breast cancer or CITRINE) to evaluate the efficacy and safety of epirubicin and cyclophosphamide, followed by weekly paclitaxel, with or without carboplatin, as adjuvant therapy for patients with high risk, early stage, triple negative breast cancer.”
“In this context, we did a phase 3 trial (carboplatin intensified chemotherapy for triple negative breast cancer or CITRINE) to evaluate the efficacy and safety of epirubicin and cyclophosphamide, followed by weekly paclitaxel, with or without carboplatin, as adjuvant therapy for patients with high risk, early stage, triple negative breast cancer.”
“Nevertheless, significant controversy remains regarding whether platinum agents can improve patients’ long term survival and the association between the efficacy of platinum agents and the germline BRCA mutation status.”
Randomization used a computer-generated block method with block size eight, and allocation was concealed through a central office. The trial was open-label, but outcome assessors, data collectors, and analysts were blinded. A power analysis was pre-specified, assuming 83% vs 89% three-year disease-free survival, requiring 144 events for 80% power. Inclusion/exclusion criteria were detailed in the protocol. Outlier handling is addressed through the pre-specified analysis population (ITT for efficacy, safety population for toxicity).
“Randomisation was done using a block randomisation method with a block size of eight, generated through a computer based allocation sequence created by an independent statistician.”
“Neither patients nor investigators were blinded to the treatment assignments, whereas all outcome assessors, data collectors, and analysts were blinded to the allocation.”
“A total of 144 disease-free survival events would provide 80% power to detect the difference between the two arms at a two sided significance level of 5%, on the basis of a 1:1 randomisation ratio.”
“Randomisation was done using a block randomisation method with a block size of eight, generated through a computer based allocation sequence created by an independent statistician.”
“Neither patients nor investigators were blinded to the treatment assignments, whereas all outcome assessors, data collectors, and analysts were blinded to the allocation.”
“A total of 144 disease-free survival events would provide 80% power to detect the difference between the two arms at a two sided significance level of 5%, on the basis of a 1:1 randomisation ratio.”
The paper reports sex (all female), age (median 47, range 22-70), and detailed baseline characteristics in Table 1, including menopausal status, histological grade, pathological T/N stage, lymphovascular invasion, Ki-67 index, and BRCA/HRR status. Since the study is human, species/strain and housing conditions are not applicable.
“The median age was 47 (range 22-70) years at the time of study entry.”
“Median (range) age, years | 48 (24-70) | 47 (22-70)”
“The median age was 47 (range 22-70) years at the time of study entry.”
“Median (range) age, years | 48 (24-70) | 47 (22-70)”
The paper states: 'The trial was conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki and was approved by the independent ethics committee at the participating institution.' It also states 'All participants gave written informed consent before their enrolment in the study.' The ethics statement in the supplementary section names the committee: 'This study was approved by the independent ethics committee of Fudan University Shanghai Cancer Center.'
“This study was approved by the independent ethics committee of Fudan University Shanghai Cancer Center.”
“All participants gave written informed consent before their enrolment in the study.”
“The trial was conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki”
“This study was approved by the independent ethics committee of Fudan University Shanghai Cancer Center.”
“All participants gave written informed consent.”
“The trial was conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki”
The trial uses standard chemotherapeutic agents, which are identified by generic name with dosing regimens. The paper does not provide manufacturer or lot numbers, but for a drug trial, the investigational product is adequately identified by name, dose, and schedule. Statistical software is identified as SPSS version 22.0 and R 4.4.1.
“We used SPSS version 22.0 and R 4.4.1 for statistical analyses.”
“We used SPSS version 22.0 and R 4.4.1 for statistical analyses.”
The paper names the tests used (log-rank, Cox proportional hazards, piecewise Cox, restricted mean survival analysis). It reports exact p-values (e.g., P=0.03) and effect sizes with 95% CIs. The proportional hazards assumption was tested and found violated, and the paper appropriately used piecewise models and restricted mean analyses. Statistical software is identified. Data presentation includes Kaplan-Meier curves and forest plots, appropriate for a clinical trial.
“We used an unadjusted Cox proportional hazards model to estimate hazard ratios and confidence intervals, and we tested the comparison of survival between trial arms by log-rank testing.”
“We assessed the assumptions of proportional hazards by assessment of the Schoenfeld residuals. When proportional hazards assumption was not valid, we provided restricted mean survival analysis and the piecewise hazard ratio within pre-defined time periods (0-12, 12-36, and >36 months).”
“hazard ratio 0.64, 95% confidence interval (CI) 0.43 to 0.95; P=0.03”
“We used an unadjusted Cox proportional hazards model to estimate hazard ratios and confidence intervals, and we tested the comparison of survival between trial arms by log-rank testing.”
“We assessed the assumptions of proportional hazards by assessment of the Schoenfeld residuals.”
“hazard ratio 0.64, 95% confidence interval (CI) 0.43 to 0.95; P=0.03”
The data availability statement states: 'The data underlying the findings in this paper are openly and publicly available and can be found at Fudan University Shanghai Cancer Center Clinical Trial Database ( http://117.72.180.52:8081 [search for CT2025-00002]).' This is a concrete route. The code is available in supplementary materials. For a clinical trial, repository deposit and accession numbers are not applicable for patient-level data, but the data are openly available.
“The data underlying the findings in this paper are openly and publicly available and can be found at Fudan University Shanghai Cancer Center Clinical Trial Database ( http://117.72.180.52:8081 [search for CT2025-00002]).”
“The code used to analyse the data in the paper can be found in the supplementary materials.”
“The data underlying the findings in this paper are openly and publicly available and can be found at Fudan University Shanghai Cancer Center Clinical Trial Database ( http://117.72.180.52:8081 [search for CT2025-00002]).”
“The code used to analyse the data in the paper can be found in the supplementary materials.”
The trial is registered at ClinicalTrials.gov (NCT04296175). The paper states it was reported according to the CONSORT 2025 statement. All pre-specified endpoints (primary and secondary) are reported, including negative/exploratory results. Limitations are explicitly discussed in the Discussion. Funding sources and competing interests are declared.
“ClinicalTrials.gov NCT04296175 (https://clinicaltrials.gov/ct2/show/NCT04296175)”
“The trial was reported according to the CONSORT 2025 statement.”
“This study has several other limitations that warrant consideration.”
“ClinicalTrials.gov NCT04296175”
“The trial was reported according to the CONSORT 2025 statement.”
“This study has several other limitations that warrant consideration.”
Registered (1 ID: ClinicalTrials.gov). Reporting guideline cited: CONSORT.
Broken references and links
None foundReferences checked against Crossref, OpenAlex and Retraction Watch for retractions and resolvability, plus declared data and code links probed for whether they resolve to content matching the paper.
Checked — nothing surfaced.
Checked 50 references by DOI: 50 verified.
Every extracted reference resolved against Crossref/OpenAlex with no retraction flags.
1 data/code link checked; 1 live.
- datahttp://117.72.180.52:8081LIVEHTTP 200Resolves, but the content could not be matched to the paper.
Copyediting
3 minorWording, consistency and formatting errors that need correcting before submission.
No major wording or formatting errors. 3 minor suggestions below.
3 copyedit issues flagged: mostly typo, consistency, clarity.
- MINORtypoDiscussion, Limitations of study“he KEYNOTE-522 trial”→ the KEYNOTE-522 trialTypo: 'he' should be 'the'.
- MINORconsistencyAbstract, Results“hazard ratio 0.64, 95% confidence interval (CI) 0.43 to 0.95; P=0.03”→ Ensure consistent use of 'hazard ratio' vs 'HR' throughout.Abbreviation 'HR' is used later; consider defining at first use.
- MINORclarityMethods, Statistical analysis“which we modelled as a continuous variables using splines”→ which we modelled as continuous variables using splinesGrammar: 'a continuous variables' should be 'continuous variables'.
The published work is robust and methodologically sound. An informed reader should weigh the non-proportional hazards and the modest effect size, but the paper's transparent reporting and appropriate statistical handling support its conclusions. Minor reporting gaps (e.g., missing CONSORT flow diagram, exact p-values for piecewise analyses) are worth noting but do not warrant a correction.
- 1.MEDIUMreportingAdd a CONSORT flow diagram to the main text (currently referenced as Fig 1 but not shown in the provided text) and report the number of patients screened and excluded.CONSORT requires a flow diagram to fully document participant flow and exclusions.
- 2.MEDIUMstatisticsReport exact p-values for the piecewise hazard ratios (currently only CIs are given) and for the interaction tests in subgroup analyses.Exact p-values allow readers to assess significance in each time period and subgroup.
- 3.MEDIUMstatisticsReport the results of the Schoenfeld residual test for secondary endpoints in a supplementary table.The proportional hazards assumption may also be violated for secondary endpoints, and reporting the test results would enhance transparency.
- 4.MEDIUMdata codeProvide the analytical code in a public repository (e.g., GitHub) with a DOI, rather than only in supplementary materials.A versioned public repository improves reproducibility and long-term access.
- 5.MEDIUMdata codeAdd a detailed data access procedure for the clinical trial database, including any conditions for access and a timeline.A clear access procedure aligns with best practices for clinical data sharing.
- 6.MEDIUMreportingAdd a statement about the availability of the study protocol in a public repository (e.g., ClinicalTrials.gov).Public protocol availability enhances transparency and allows verification of pre-specified analyses.
- 7.MEDIUMstatisticsAdd a note on the handling of missing data for the primary analysis (e.g., censoring rules) in the statistical analysis section.Clarifying missing data handling helps readers interpret the analysis.
- 8.LOWreportingReport the manufacturer and lot numbers for the chemotherapeutic agents used.This is sometimes expected for investigational products to ensure traceability.
- 9.LOWcopyeditFix the typo 'he KEYNOTE-522 trial' to 'the KEYNOTE-522 trial' in the Discussion, Limitations of study.Correcting typos improves professionalism and readability.
- 10.LOWcopyeditFix the grammar 'a continuous variables' to 'continuous variables' in Methods, Statistical analysis.Correcting grammar improves clarity.
- 11.LOWcopyeditEnsure consistent use of 'hazard ratio' vs 'HR' throughout the manuscript; define at first use.Consistent terminology avoids confusion for readers.
The star rating is the report’s one-glance summary. Every paper starts at 5★ and loses stars for the concrete problems the review finds — so a rating is never a vague average, it’s a running total you can read line by line under “How this rating was calculated.”
- Reporting — 8 dimensionseach dimension that fully fails−½★
- each dimension partially met−¼★
- Statistics · Integrity · Claimseach serious problem−1★
- each medium problem−½★
- Citationseach retracted or unverifiable reference−¼★
- Copyeditonly when the manuscript needs a full edit−½★
The rating never drops below 1★, and a demonstrable critical failure (an impossible statistic, a proven ethics violation) caps it at 1★ on its own — so the stars can never look healthy when the verdict is CRITICAL.
The rating draws on a panel of agents. Three independent Kaimen Rigor reviewers grade the eight dimensions below across several independent passes (the shown verdict is their majority vote — steadier than any single run), isolate the paper’s major claims and check its own evidence backs them, and flag integrity concerns. Alongside them, a citation agent resolves every reference against Crossref, OpenAlex, and Retraction Watch; a statistics agent recomputes reported tests; and rule-based checks verify that declared data/code links actually resolve. Full text is required — an abstract-only submission is not analyzed.
Graded against NIH, MDAR, ARRIVE 2.0, CONSORT, EQUATOR, and RRID guidelines. A dimension that doesn’t apply to the study type is skipped, never penalized.
This Kaimen Rigor review is model-assisted and is not a substitute for formal expert review. It complements human evaluation by surfacing potential methodological concerns — verify each finding against the source.