Colchicine in Acute Myocardial Infarction.
Jolly SS, d'Entremont MA, Lee SF, Mian R, Tyrwhitt J, Kedev S, Montalescot G, Cornel JH, Stanković G, Moreno R, Storey RF, Henry TD, Mehta SR, Bossard M, Kala P, Layland J, Zafirovska B, Devereaux PJ, Eikelboom J, Cairns JA, Shah B, Sheth T, Sharma SK, Tarhuni W, Conen D, Tawadros S, Lavi S, Yusuf S, CLEAR Investigators
- DOI
- 10.1056/NEJMoa2405922
- Record issued
- 2026-08-16
- 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/9db5c8f6-a3d3-473b-abb3-e930f58dc949 is authoritative.
How this rating was calculated
- StatisticsImpossible or misreported statistic ×2−2★
- IntegrityIntegrity concern−0.5★
- ReportingData & code availability partially met−0.25★
A demonstrable critical failure caps the rating at the minimum, regardless of the deductions above.
- The numeric-impossibility checks (GRIM/GRIMMER/DEBIT/SPRITE) did not run: 6 reported means were read, and their group size is not stated where the values are printed (this source has no machine-readable table structure). These checks need the count the mean was averaged over, so none was performed.
- No data or code availability links were detected to verify.
- 01Printed percentage does not match its own countdemonstrable
2.7% does not match the reported count 137/3528
“Coronary artery bypass grafting (%) 137 (2.7)”
Table 1Find in source - 02Printed percentage does not match its own countdemonstrable
2.8% does not match the reported count 139/3534
“Coronary artery bypass grafting (%) 139 (2.8)”
Table 1Find in source
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 well-reported randomized clinical trial with rigorous design, clear ethics approvals, and appropriate statistical methods. The main weaknesses are the absence of a data availability statement, lack of explicit statistical software identification, and minor copyedit issues including a potential denominator inconsistency.
This is an interventional clinical trial; all dimensions were applicable except those specific to animal or cell-line research. The two reviewers agreed on all dimensions; no divergence to reconcile. The statistics verification covered only a subset of tests (5 recomputed, 3 consistent, 2 inconsistent but no decision errors), so the statistical analysis is not fully verified.
Numerical inconsistencies
2 findings · worst criticalValues 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.
- Summary statistic impossible for the stated N (GRIM/GRIMMER)Recomputed
- Internal contradictions in the reported numbersAssessed
Recomputed 3 tests: 3 consistent, 0 inconsistent; 1 recomputed directly from the reported test statistics, 2 via agent-written checks. 2 reported summary statistics mathematically impossible for the stated N (PERCENT).
- PERCENT2.7% does not match the reported count 137/3528
“Coronary artery bypass grafting (%) 137 (2.7)”
Table 1Find in source - PERCENT2.8% does not match the reported count 139/3534
“Coronary artery bypass grafting (%) 139 (2.8)”
Table 1Find in source
- CONSISTENTreported p = .930 · recomputed p = .899Recomputed hazard ratio 0.99 (95% CI 0.85–1.16), reported p=0.93
“hazard ratio 0.99; 95% confidence interval (CI) 0.85-1.16, p=0.93”
Taken as given: 0.85–1.16 is a two-sided 95% confidence interval for the hazard ratio of 0.99, 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=0.93 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(0.99, 0.85, 1.16, 1) - CONSISTENTreported p = .930 · recomputed p = .899Reviewers 1, 2Primary outcome hazard ratio p-value
“The primary outcome occurred in 322 of 3528 patients (9.1%) in the colchicine group compared with 327 of 3524 patients (9.3%) in the placebo group (hazard ratio [HR], 0.99; 95% CI 0.85-1.16; P=0.93)”
Taken as given: The hazard ratio is 0.99 with 95% CI 0.85-1.16.; The CI is two-sided at 95%.; The p-value is two-sided.Method: Recomputed p-value from the hazard ratio and its 95% confidence interval using the normal approximation for the log hazard ratio.How we recomputed it: pCI(0.99, 0.85, 1.16, 1) - CONSISTENTreported p < .001 · recomputed p = <.001Reviewers 1, 2Diarrhea comparison p-value
“More colchicine patients (n=361; 10.2%) had diarrhea compared to placebo patients (n=233; 6.6%), p<0.001.”
Taken as given: The numbers 361 and 233 are the event counts in the colchicine and placebo groups, respectively.; The group totals are 3528 and 3534.; The test used is a two-sided chi-square test.Method: Recomputed p-value using Pearson's chi-square test for a 2x2 table.How we recomputed it: pChi2x2(361, 3528-361, 233, 3534-233)
- lowinternal contradictionThe denominator for the placebo group in the primary outcome sentence (3524) differs from the total randomized to placebo (3534) as stated elsewhere.
“The primary outcome occurred in 322 of 3528 patients (9.1%) in the colchicine group compared with 327 of 3524 patients (9.3%) in the placebo group”
ResultsFind in source
Overstated conclusions
1 finding · worst lowConclusions 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.
- Conclusions only partially backed by the presented evidenceAssessed
8 major claims checked against the paper's own evidence: 1 only partially supported (evidence backs part of the claim; gaps or caveats remain); the rest adequately supported.
- partialReviewer 2The trial provides the most robust evidence to date on colchicine in vascular disease.The trial is large with many events, but the claim of 'most robust' is an interpretation and not directly proven by the data.Evidence: Discussion mentions the trial is the largest with 649 primary events.
“The CLEAR trial is the largest trial in patients with coronary artery disease, with 649 primary outcome events, which is significantly more outcome events than prior trials in those with coronary artery disease.”
DiscussionFind in source - supportedReviewer 1Colchicine did not reduce the composite of cardiovascular death, myocardial infarction, stroke, or unplanned ischemia-driven coronary revascularization in patients post myocardial infarction.The primary outcome result (HR 0.99, 95% CI 0.85-1.16, p=0.93) directly supports this claim.Evidence: Primary outcome analysis in Table 2 and Results section.
“In patients post myocardial infarction, colchicine, when started early and continued for a median of 3 years, did not reduce the composite of CV death, MI, stroke or unplanned ischemia-driven coronary revascularization.”
ConclusionFind in source - supportedReviewers 1, 2Colchicine reduced C-reactive protein levels.The CRP analysis shows a significant reduction (MD -1.28, 95% CI -1.81 to -0.75).Evidence: CRP results in Results section.
“The least square mean (standard error) of CRP at 3 months, adjusted for baseline, was 2.98 (0.19) mg/L in 1,384 patients randomized to Colchicine and 4.27 (0.19) mg/L in 1,419 patients randomized to Placebo, with a MD of -1.28 (- 1.81, -0.75) mg/L.”
ResultsFind in source - supportedReviewer 1Colchicine increased diarrhea compared to placebo.The safety analysis shows a significantly higher rate of diarrhea in the colchicine group (10.2% vs 6.6%, p<0.001).Evidence: Safety results in Results section and Table 3.
“More colchicine patients (n=361; 10.2%) had diarrhea compared to placebo patients (n=233; 6.6%), p<0.001.”
ResultsFind in source - supportedReviewer 1The CLEAR trial is the largest trial in patients with coronary artery disease with 649 primary outcome events.The paper states this and provides the event count (649) which is the sum of primary events in both groups.Evidence: Discussion section.
“The CLEAR trial is the largest trial in patients with coronary artery disease, with 649 primary outcome events, which is significantly more outcome events than prior trials in those with coronary artery disease.”
DiscussionFind in source - supportedReviewer 1The results are robust as on-treatment analysis was consistent with intention-to-treat analysis.The paper states the on-treatment analyses were consistent with the primary analysis.Evidence: Results, Sensitivity analyses and Discussion.
“Our on-treatment analysis was consistent with our intention-to-treat analysis, suggesting our findings are robust.”
DiscussionFind in source - supportedReviewer 2Colchicine did not reduce the primary composite outcome in post-MI patients.The primary outcome result (HR 0.99, 95% CI 0.85-1.16, p=0.93) directly supports the null finding.Evidence: Primary outcome result in Results section.
“In patients post myocardial infarction, colchicine, when started early and continued for a median of 3 years, did not reduce the composite of CV death, MI, stroke or unplanned ischemia-driven coronary revascularization.”
ConclusionFind in source - supportedReviewer 2Colchicine increased diarrhea.Diarrhea was significantly more frequent with colchicine (10.2% vs 6.6%, p<0.001).Evidence: Safety results in Results section.
“More colchicine patients (n=361; 10.2%) had diarrhea compared to placebo patients (n=233; 6.6%), p<0.001.”
ResultsFind in source
Efficacy claim is anchored to an adequate endpoint and a meaningful effect.
- N/ASurrogate endpointThe primary outcome is a composite of hard clinical events (cardiovascular death, myocardial infarction, stroke, or unplanned ischemia-driven coronary revascularization), not a surrogate.
“The primary outcome was the composite of cardiovascular death, myocardial infarction, stroke or unplanned ischemia-driven coronary revascularization.”
- ADEQUATEEffect sizeThe primary outcome showed no significant difference between groups (HR 0.99, 95% CI 0.85-1.16, p=0.93), with event rates of 9.1% vs 9.3%. The effect size is reported with confidence intervals and is anchored to clinical outcomes, and the conclusion of no benefit is supported by the data.
“The primary outcome occurred in 322 of 3528 patients (9.1%) in the colchicine group and 327 of 3534 patients (9.3%) in the placebo group over a median follow-up of three years (hazard ratio 0.99; 95% confidence interval (CI) 0.85-1.16, p=0.93).”
Data authenticity concerns
None foundAn 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.
Checked — nothing surfaced.
Reporting gaps
1 finding · worst mediumRequired detail the manuscript never states — study design, biological variables, ethics approval and consent, key resources, statistical reporting, data and code availability, and overall transparency.
- Data/code availability incompleteAssessed
The introduction cites multiple prior trials (COLCOT, LoDoCo2, CHANCE, CONVINCE) and notes both beneficial and null results, acknowledging the need for more data. The rationale for the trial is clearly stated: 'Given the biological rationale and encouraging evidence, we conducted the CLEAR trial to provide examine the effect of colchicine in post-myocardial infarction patients.' Limitations of prior work are implicitly addressed by designing a larger trial with more events, as discussed in the Discussion.
“A trial of 4745 patients where colchicine was initiated within 30 days of a myocardial infarction and a trial of 5522 patients with stable coronary artery disease reported beneficial cardiovascular effects of colchicine, whereas 2 recent trials in patients with ischemic stroke (n=8343, n=3154) reported no reductions in cardiovascular events with colchicine with the caveat that one trial was a short term treatment trial (3 months).”
“Given the biological rationale and encouraging evidence, we conducted the CLEAR trial to provide examine the effect of colchicine in post-myocardial infarction patients.”
“A trial of 4745 patients where colchicine was initiated within 30 days of a myocardial infarction and a trial of 5522 patients with stable coronary artery disease reported beneficial cardiovascular effects of colchicine, whereas 2 recent trials in patients with ischemic stroke (n=8343, n=3154) reported no reductions in cardiovascular events with colchicine with the caveat that one trial was a short term treatment trial (3 months).”
“Given the biological rationale and encouraging evidence, we conducted the CLEAR trial to provide examine the effect of colchicine in post-myocardial infarction patients.”
Randomization used permuted blocks within a 24-hour computerized central system, stratified by center and STEMI/NSTEMI. Blinding was comprehensive: 'All participants, investigators, healthcare providers, data collectors and outcome adjudicators were blinded to treatment allocation.' The sample size was calculated and later increased based on blinded interim event rates. Inclusion/exclusion criteria were pre-specified and modified with a protocol amendment. The analysis population (intention-to-treat) and on-treatment sensitivity analyses are defined. Outlier handling is not explicitly discussed, but missing data are addressed as likely missing at random.
“Randomization was performed using permuted blocks within a 24-hour computerized central system at the Population Health Research Institute.”
“All participants, investigators, healthcare providers, data collectors and outcome adjudicators were blinded to treatment allocation.”
“The initial sample size calculation for cardiovascular death, recurrent myocardial infarction or stroke was based on a time-to-event analysis with an anticipated control event rate of 15% at three years, 80% power, a 2-sided type 1 error level of 5%, a 2% loss to follow-up in both arms, study drug discontinuation of 12.5%, and the assumption of no interaction with spironolactone.”
“Randomization was performed using permuted blocks within a 24-hour computerized central system at the Population Health Research Institute. Randomization was stratified by study center and whether the patient had a STEMI or NSTEMI.”
“All participants, investigators, healthcare providers, data collectors and outcome adjudicators were blinded to treatment allocation.”
“The initial sample size calculation for cardiovascular death, recurrent myocardial infarction or stroke was based on a time-to-event analysis with an anticipated control event rate of 15% at three years, 80% power, a 2-sided type 1 error level of 5%, a 2% loss to follow-up in both arms, study drug discontinuation of 12.5%, and the assumption of no interaction with spironolactone.”
The paper reports sex (20.4% female overall), age (mean 61 years), and detailed baseline characteristics including comorbidities, medications, and procedural details. Demographics include race/ethnicity and geographic region. Since both sexes are enrolled, sex_justified is not applicable. Species/strain and housing conditions are not applicable for a human trial.
“Baseline characteristics were well balanced between the groups, with the overall mean age being 61 years and 20.4% of patients being female (Table 1).”
“Race or ethnic group (%) American Indian or Alaskan Native 7 (0.2) 3 (0.1) Asian 95 (2.7) 89 (2.5) Black or African American 24 (0.7) 23 (0.7) Native Hawaiian or Other Pacific Islander 9 (0.3) 9 (0.3) White 3233 (91.6) 3249 (91.9) Other 153 (4.3) 159 (4.5)”
“Age, years (mean ±SD) 60.6 ±10.3 60.7 ±10.3”
“Diabetes Mellitus (%) 658 (18.7) 645 (18.3)”
The paper states: 'Ethics committees of participating centers and national regulatory authorities approved the trial. All patients provided written informed consent.' This satisfies both IRB approval and informed consent. Regulatory compliance is implied by the approval of national regulatory authorities, though not explicitly naming a framework like the Declaration of Helsinki, but the statement is adequate.
“Ethics committees of participating centers and national regulatory authorities approved the trial. All patients provided written informed consent.”
“Ethics committees of participating centers and national regulatory authorities approved the trial. All patients provided written informed consent.”
The study drug is identified: 'Study drugs were colchicine tablets of 0.5 mg and spironolactone tablets of 25 mg or matching placebos. Tiofarma, Netherlands, provided both study drugs with raw materials produced by Indena S.p.A., Milan, Italy.' This provides manufacturer and formulation. Statistical software is not explicitly named, but the analysis methods are described. Other bench resources are not applicable.
“Study drugs were colchicine tablets of 0.5 mg and spironolactone tablets of 25 mg or matching placebos. Tiofarma, Netherlands, provided both study drugs with raw materials produced by Indena S.p.A., Milan, Italy.”
“Study drugs were colchicine tablets of 0.5 mg and spironolactone tablets of 25 mg or matching placebos. Tiofarma, Netherlands, provided both study drugs with raw materials produced by Indena S.p.A., Milan, Italy.”
The primary analysis used a log-rank test and Cox proportional hazards model, with Fine-Gray for competing risks. Exact p-values are reported (e.g., p=0.93 for primary outcome). Effect sizes are reported with 95% CIs. Statistical software is not explicitly named, but the methods are described in detail. Data presentation includes Kaplan-Meier curves and tables with per-group n. Mathematical plausibility checks were not performed due to large N and continuous outcomes.
“A two-sided, log-rank test was used to compare the two randomized groups. A p-value of less than 0.05 was considered significant. The hazard ratio and 95% confidence interval were estimated using a Cox proportional hazards regression model with the treatment group as the independent variable stratifying by spironolactone or placebo and STEMI or NSTEMI.”
“The primary outcome occurred in 322 of 3528 patients (9.1%) in the colchicine group compared with 327 of 3524 patients (9.3%) in the placebo group (hazard ratio [HR], 0.99; 95% CI 0.85-1.16; P=0.93)”
“A two-sided, log-rank test was used to compare the two randomized groups. A p-value of less than 0.05 was considered significant. The hazard ratio and 95% confidence interval were estimated using a Cox proportional hazards regression model with the treatment group as the independent variable stratifying by spironolactone or placebo and STEMI or NSTEMI.”
“The primary outcome occurred in 322 of 3528 patients (9.1%) in the colchicine group compared with 327 of 3524 patients (9.3%) in the placebo group (hazard ratio [HR], 0.99; 95% CI 0.85-1.16; P=0.93)”
The paper does not include an explicit data availability statement in the provided text. The trial is registered (NCT03048825), but no repository or access mechanism is mentioned. Since this is a clinical trial with patient data, managed access would be acceptable, but the absence of any statement is a gap. Code sharing is not applicable as no custom code is mentioned.
The trial is registered (NCT03048825). Methods are detailed enough for replication. Limitations are explicitly discussed in a dedicated section. Conclusions are proportional to the results, acknowledging the null finding. Funding sources and conflicts are stated. A reporting guideline (CONSORT) is not explicitly mentioned, but the paper follows standard reporting.
“Trial registration: ClinicalTrials.gov number NCT03048825”
“First, based on the 95% CI of the primary outcome results, we cannot exclude a benefit of 15% or smaller, that could be clinically important.”
“Trial registration: ClinicalTrials.gov number NCT03048825”
“First, based on the 95% CI of the primary outcome results, we cannot exclude a benefit of 15% or smaller, that could be clinically important.”
Registered (1 ID: ClinicalTrials.gov). No reporting guideline cited.
Broken references and links
None found · partly checkedReferences 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.
Nothing surfaced — but not everything feeding this category ran (missing: data/code link verification), so read this as a partial clean bill.
Checked 12 references by DOI: 12 verified.
Every extracted reference resolved against Crossref/OpenAlex with no retraction flags.
Copyediting
7 minorWording, consistency and formatting errors that need correcting before submission.
No major wording or formatting errors. 7 minor suggestions below.
7 copyedit issues flagged: mostly typo, grammar, consistency.
- MINORtypoIntroduction, last sentence“we conducted the CLEAR trial to provide examine the effect of colchicine”→ Remove 'provide' to read 'to examine the effect'.Typographical error.
- MINORgrammarMethods, Statistical Considerations“The total event was analyzed using Prentice, William, Peterson (PWP model with the gap time and the Lin Wei Yang Ying (LWYY) model.”→ Add a closing parenthesis after 'PWP' and clarify the sentence structure.Missing parenthesis and awkward phrasing.
- MINORconsistencyResults, Efficacy“327 of 3524 patients (9.3%)”→ Check if the denominator should be 3534 (as in Table 2) instead of 3524.Potential inconsistency in the denominator for the placebo group.
- MINORclarityMethods, Study Design“we conducted the CLEAR trial to provide examine the effect of colchicine in post-myocardial infarction patients.”→ Rephrase to 'to examine the effect of colchicine in post-myocardial infarction patients.'Redundant word 'provide'.
- MINORtypoAbstract, Methods“we conducted the CLEAR trial to provide examine the effect”→ Remove 'provide'Typo: 'provide examine' should be 'examine'.
- MINORgrammarMethods, Statistical Considerations“The total event was analyzed using Prentice, William, Peterson (PWP model with the gap time and the Lin Wei Yang Ying (LWYY) model.”→ Add missing parenthesis: 'Prentice, William, Peterson (PWP) model'Missing closing parenthesis after 'PWP'.
- MINORconsistencyResults, Efficacy“327 of 3524 patients (9.3%)”→ Check if denominator should be 3534 as in Table 2.Inconsistent denominator: text says 3524, Table 2 says 3534.
The published work is robust and generally trustworthy, but readers should weigh the missing data availability statement and the minor internal inconsistency in the primary outcome denominator (3524 vs 3534). These do not undermine the main conclusions but warrant attention; an erratum or clarification may be appropriate for the denominator issue.
- 1.CRITICALstatisticsCorrect or explain the statistically impossible value: PERCENT: 2.7% does not match the reported count 137/3528Demonstrable critical failure — blocks the verdict from passing.
- 2.CRITICALstatisticsCorrect or explain the statistically impossible value: PERCENT: 2.8% does not match the reported count 139/3534Demonstrable critical failure — blocks the verdict from passing.
- 3.HIGHdata codeAdd a data availability statement in the Methods or a dedicated section, specifying how de-identified patient data can be accessed (e.g., via a data access committee or a repository like YODA/Vivli) with conditions and timeframe.The paper currently lacks any data availability statement, which is a reporting gap for a data-driven clinical trial.
- 4.HIGHstatisticsReconcile the denominator for the placebo group in the primary outcome sentence (Results, Efficacy) with Table 2 (3524 vs 3534) and correct the inconsistency.An internal contradiction in a key result undermines trust and could warrant an erratum.
- 5.MEDIUMreportingExplicitly name the statistical software used (e.g., SAS version 9.4, R version 4.2) in the Statistical Considerations section.Identifying software is part of methodological transparency and aids reproducibility.
- 6.MEDIUMreportingMention adherence to a reporting guideline such as CONSORT in the Methods or a footnote, and include the completed checklist as supplementary material.Explicitly stating CONSORT adherence is standard for clinical trials and improves transparency.
- 7.MEDIUMcopyeditFix the typo in the Introduction and Abstract: remove 'provide' from 'to provide examine the effect'.Typographical errors detract from professionalism.
- 8.MEDIUMcopyeditFix the missing parenthesis in Methods, Statistical Considerations: 'Prentice, William, Peterson (PWP) model' and clarify the sentence structure.Grammar and clarity issues in the methods section can confuse readers.
- 9.LOWrigorClarify the handling of outliers in the statistical analysis, even if only to state that no outliers were excluded or that sensitivity analyses were performed.Outlier handling is a minor reporting gap that reviewers may ask about.
- 10.LOWrigorProvide a more detailed description of the randomization implementation, including the specific software or system used for the 24-hour computerized central system.Additional detail on randomization sequence generation would enhance reproducibility.
- 11.LOWreportingConsider including a statement about the availability of the trial protocol and statistical analysis plan, perhaps as supplementary files or a link.Sharing the protocol and SAP is good practice and supports transparency.
- 12.LOWreportingInclude a CONSORT flow diagram in the main text or supplementary, as it is standard for clinical trials.A flow diagram improves clarity on patient disposition.
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.