Outcome-specific differences between
randomized trials and real-world studies
of CDK4/6 inhibitors in hormone receptor-
positive, HER2-negative advanced breast
cancer: a meta-epidemiological analysis

Real-world evidence (RWE) is increasingly used alongside evidence from randomized clinical trials (RCTs) in clinical decision-making, regulatory evaluation, guideline development, and health technology assessment.¹⁻⁴ In oncology, RWE is particularly relevant because RCT participants are often highly selected, whereas patients treated in routine practice are more clinically heterogeneous.⁵˒⁶ Clinicians, guideline developers, and decision-making bodies commonly interpret outcomes from RCTs and real-world studies (RWSs) in parallel.⁶⁻⁸ However, outcomes derived from RWSs are generated under routine-care conditions rather than protocol-driven trial settings. Differences in patient selection, treatment composition, follow-up intensity, outcome ascertainment, documentation, and reporting may therefore shape observed outcomes, including objective response rates and adverse event proportions.⁹˒¹⁰ A central uncertainty in the use of RWE is not whether RWD should be dismissed, but how such data should be interpreted: whether outcome estimates derived from routine practice are sufficiently valid, reliable, and comparable with RCT outcomes, and under which outcome domains direct interchangeability may break down. ⁷˒¹⁰˒¹¹

This uncertainty reflects not only differences in patient characteristics and treatment composition, but also systematic differences in outcome generation, measurement, and reporting.⁷˒¹²˒¹³ Previous meta-epidemiological studies have largely examined how methodological features within RCTs affect treatment-effect estimates, whereas comparisons of RCTs and RWE have more often focused on survival outcomes or relative effects from comparative designs.¹⁴⁻²⁰ Systematic cross-design comparisons of single-arm outcomes remain limited.⁷˒²¹˒²² The comparability of such outcomes may also vary by ascertainment mechanism: response outcomes are influenced by assessment criteria and denominator construction; laboratory-based adverse events depend more on objective testing; symptom-based adverse events depend on active elicitation and documentation; and management-based adverse events are linked to clinical decision records.²¹˒²³⁻²⁵ Without stratification by ascertainment mechanism, cross-source comparisons may obscure the sources of differences between RCTs and RWSs.¹³˒²¹˒²²

CDK4/6 inhibitors combined with endocrine therapy for hormone receptor–positive, ERBB2-negative advanced or metastatic breast cancer provide a suitable case for this comparison.²⁶ Multiple RCTs have established their efficacy, and real-world cohorts from multiple countries and regions are available.²⁶⁻³² Reported outcomes also span objective response, laboratory-based adverse events, symptom-based adverse events, and treatment management outcomes.³³˒³⁴ We therefore conducted a meta-epidemiological analysis in which the quantity under study was the reported outcome proportion rather than the treatment effect. We quantified differences between RCTs and RWSs in single-arm efficacy and safety outcome proportions, assessed whether these differences persisted after adjustment for observable study-level covariates, and tested whether residual differences varied according to outcome ascertainment mechanism.

Protocol and Reporting Standards

This meta-epidemiological analysis was based on a systematic review. The protocol was registered with PROSPERO (CRD420261397529) and locked before specification of the primary analytic model. The study followed the PRISMA 2020 statement and MOOSE reporting guideline.³⁵˒³⁶ Complete search strategies, variable definitions, protocol deviations, and reporting checklists are provided in Additional file 1.

Search Strategy and Study Selection

We searched PubMed, Embase, the Cochrane Library, and Web of Science from database inception through December 2025. Searches combined terms for hormone receptor–positive, ERBB2-negative breast cancer; CDK4/6 inhibitors plus endocrine therapy; and randomized clinical trial or real-world study designs. Full database-specific strategies are provided in the eMethods. Reference lists of included studies and relevant systematic reviews were also screened.

Two reviewers independently screened titles, abstracts, and full texts, with disagreements resolved through discussion or adjudication by a third reviewer. The selection process is shown in PRISMA 2020 flow diagrams.

Eligibility Criteria

Eligible studies included adults with pathologically confirmed hormone receptor–positive, ERBB2-negative locally advanced or metastatic breast cancer. Hormone receptor positivity was defined as estrogen receptor positivity and/or progesterone receptor positivity. Eligible interventions were palbociclib, ribociclib, abemaciclib, or dalpiciclib combined with endocrine therapy, including aromatase inhibitors or fulvestrant, as first- or second-line treatment for advanced disease.

Eligible designs were phase 2 or 3 RCTs and prospective or retrospective observational cohorts, registry studies, or postmarketing surveillance studies. For RCTs, only CDK4/6 inhibitor treatment arms were extracted; for RWSs, corresponding real-world intervention cohorts were extracted. We excluded studies of early breast cancer; CDK4/6 inhibitor monotherapy; third-line or later treatment; combination treatment with chemotherapy, nonendocrine targeted therapy, or immunotherapy; ERBB2-positive or hormone receptor–negative disease; phase 1 dose-finding studies; abstracts, reviews, case reports, and duplicate publications; and studies lacking extractable event counts, denominators, or essential study information.

Data Extraction and Risk of Bias Assessment

Two reviewers independently extracted data using a prespecified standardized form, with disagreements resolved by discussion or third-reviewer adjudication. Extracted data included study characteristics, baseline patient characteristics, treatment characteristics, risk-of-bias information, and event counts and denominators for each outcome. Complete variable definitions are provided in eMethods 3.

Risk of bias was assessed using the Cochrane Risk of Bias 2 tool for RCTs and ROBINS-I for RWSs.³⁷˒³⁸ Because the analysis focused on single-arm outcome proportions rather than within-study relative treatment effects, these assessments were used primarily to describe evidence-source quality and inform interpretation of heterogeneity.

Outcomes

Efficacy outcomes included objective response rate, clinical benefit rate, complete response, partial response, stable disease, and progressive disease.

Safety outcomes were classified according to their expected ascertainment mechanisms and outcome characteristics. Laboratory-based adverse events included grade 3 or higher neutropenia, leukopenia, and anemia. Symptom-based adverse events included diarrhea, nausea, and vomiting. All-grade and grade 3 or higher adverse events were considered composite safety outcomes encompassing both symptom-based and objectively measured events. Management-based adverse events included adverse event–related treatment discontinuation and dose adjustment.

Adverse events of special interest included venous thromboembolism, interstitial lung disease or pneumonitis, and QT prolongation; these were described exploratorily and evaluated for certainty of evidence when data permitted. Median progression-free survival and median overall survival were presented descriptively and were not included in pooled analyses of single-arm event proportions.

Statistical Analysis

All analyses used RCT treatment arms or RWS cohorts receiving CDK4/6 inhibitor treatment as the unit of analysis. The estimand of this study was not a treatment effect. It was the difference in the outcome proportion reported for the same intervention when that intervention is observed through two different evidence-generating processes. The object of comparison was therefore the measurement, documentation, and reporting pathway rather than the pharmacological activity of CDK4/6 inhibitors, and single-arm event proportions were the analytic level appropriate to this question. Consistent with this estimand, no cross-source estimate was interpreted as evidence of comparative effectiveness or comparative safety. For each binary outcome, event proportions were pooled separately for RCTs and RWSs using logit-transformed random-effects models, with between-study variance estimated by restricted maximum likelihood. No overall pooled estimate combining the 2 evidence sources was reported, to avoid implying direct interchangeability. Pooled proportions were reported with 95% CIs and 95% prediction intervals.

Baseline covariate balance between RCTs and RWSs was assessed using standardized mean differences (SMDs), with an absolute SMD greater than 0.10 indicating potential imbalance.³⁹ Balance was displayed using a Love plot, and descriptive baseline characteristics were summarized with weights proportional to treatment arm or cohort sample size.

Mixed-effects meta-regression was used to evaluate adjusted differences in single-arm outcomes between RCTs and RWSs. The dependent variable was the logit-transformed event proportion, and the primary explanatory variable was evidence source, defined as RWS vs RCT.⁴⁰˒⁴¹ The evidence-source coefficient represented the adjusted association between RWS, relative to RCT, and the outcome proportion, and was reported as an odds ratio with 95% CI.

Covariate adjustment followed a prespecified imbalance-and-outcome-relevance algorithm. Because this study compared single-arm proportions between RCT treatment arms and RWS cohorts rather than within-trial randomized relative effects, randomization did not ensure baseline balance between evidence sources. Candidate covariates were first restricted to variables imbalanced between RCTs and RWSs, defined as an absolute SMD greater than 0.10. Among these, variables associated with the corresponding outcome in univariable meta-regression, defined as P < .10, were included in the multivariable model. Unlike within-trial relative treatment effects, single-arm outcome proportions are influenced by baseline prognostic characteristics, which may differ in their relevance across outcomes. Therefore, the prespecified outcome-specific adjustment strategy was used to account for measured between-source imbalances while considering outcome relevance and model parsimony. The prespecified common-core model was retained as a sensitivity analysis to assess the robustness of findings to a shared adjustment set.

Two prespecified sensitivity analyses assessed the robustness of the adjustment strategy. First, all outcomes were adjusted using a common core covariate set, including age, ECOG performance status of 2 or higher, visceral metastasis, bone-only metastasis, and 3 or more metastatic sites. Second, models were adjusted for the proportion receiving first-line therapy and the proportion receiving palbociclib, to examine whether treatment line and drug composition explained the main findings. Covariates included in each outcome-specific model are listed in the corresponding adjusted results figures. An additional study-dependence sensitivity analysis excluded identifiable overlapping reports and applied cluster-robust variance estimation (CR2) with Satterthwaite small-sample correction, clustering by unique trial or cohort family while retaining the original model specifications and covariates. Details are provided in the Supplementary Methods. Between-study heterogeneity was described using τ², I², the Cochran Q test, and 95% prediction intervals.⁴²˒⁴³ For outcomes with at least 10 contributing studies, funnel plots and Egger tests were used to assess small-study effects or publication bias.⁴⁴ All tests were 2-sided, with α = .05. Analyses were performed using R version 4.6 with the metafor and meta packages.

Study Selection

The electronic search identified 6771 records, including 3607 from the RCT search and 3164 from the RWS search. After removal of duplicates, 5637 records underwent title and abstract screening, and 737 full-text articles were assessed for eligibility. A total of 86 studies were included: 31 RCTs and 55 RWSs. These 86 studies contributed 141 CDK4/6 inhibitor treatment arms or cohorts for single-arm analysis, including 40 treatment arms from 31 RCTs involving 7464 patients and 101 cohorts from 55 RWSs involving 30 858 patients. The full study selection process is shown in eFigure 1 for RCTs and eFigure 2 for RWSs.

Study and Patient Characteristics

When weighted by the sample size of each CDK4/6 inhibitor treatment arm or cohort, patient characteristics differed substantially between RWSs and RCTs. Patients in RWSs were older than those in RCTs, with a mean age of 64.2 years vs 60.0 years; a higher proportion received first-line therapy, 94.7% vs 64.5%; and a higher proportion received palbociclib, 76.1% vs 47.9%. Because first-line treatment proportion and palbociclib proportion were the most prominent compositional differences between the 2 evidence sources, these variables were examined in subsequent sensitivity analyses. RCT reports had a median publication year of 2022 (range, 2015–2025), compared with 2024 (range, 2020–2025) for RWS reports. Geographically, most RCT reports were multinational (20 of 31), whereas RWS reports were predominantly country-specific, most commonly from the United States (17 of 55), Italy (6 of 55), and Türkiye (5 of 55).

Unadjusted Between-Source Estimates

Unadjusted single-arm syntheses showed that RWSs generally reported more favorable efficacy estimates and lower adverse event proportions than RCTs. In contrast, management-based adverse events reflecting explicit clinical decisions, including treatment discontinuation and dose adjustment, were more similar between the 2 evidence sources. Median progression-free survival was similar between RCTs and RWSs, at 25.1 months among RCTs (k = 37) and 23.0 months among RWSs (k = 81). However, median overall survival showed substantial between-study heterogeneity, with numerically higher reported values in RCTs than in RWSs, at 57.7 months vs 45.5 months, respectively (k = 10 and k = 53). Detailed unadjusted estimates are provided in eFigure 3 for efficacy outcomes, eFigure 4 for safety outcomes, and eFigure 5 for progression-free and overall survival distributions.

Covariate Balance

Among the 19 baseline covariates assessed, 11 showed imbalance between RCTs and RWSs, defined as an absolute standardized mean difference greater than 0.10. Four covariates showed pronounced imbalance, defined as an absolute standardized mean difference greater than 0.25: mean age (+0.79), 3 or more metastatic sites (−0.64), ECOG performance status of 2 or higher (+0.35), and bone-only metastasis (+0.32) (Figure 1).

Figure 1. Covariate Balance Between Randomized Trial Arms and Real-World Study Cohorts

Signed standardized mean differences (SMDs) are shown for baseline patient, disease, endocrine-sensitivity, and treatment-history characteristics comparing real-world study (RWS) cohorts with randomized controlled trial (RCT) treatment arms. Positive values indicate higher values or proportions in RWS cohorts than in RCT arms, whereas negative values indicate higher values or proportions in RCT arms. Dashed vertical reference lines indicate absolute SMD thresholds of 0.10, 0.20, and 0.30. An absolute SMD greater than 0.10 was considered evidence of potential imbalance, and larger values indicated more pronounced between-source imbalance. Covariates are grouped by characteristic domain.

Primary Adjusted Analysis

After multivariable adjustment using covariates selected for each outcome on the basis of standardized mean differences and univariable meta-regression, the pattern of differences between RCTs and RWSs varied substantially by outcome type.

For efficacy outcomes, the adjusted objective response rate was significantly higher in RWSs than in RCTs (51.8% vs 36.6%; P = .031) (Figure 2). Clinical benefit rate (85.4% vs 74.7%), complete response (6.4% vs 3.0%), partial response (42.7% vs 32.9%), and progressive disease (12.3% vs 8.9%) were also numerically higher in RWSs, although the corresponding P values ranged from .06 to .08 and did not meet conventional thresholds for statistical significance. Overall, the direction of all efficacy outcomes suggested more favorable estimates in RWSs than in RCTs.

Figure 2. Primary Adjusted Estimates for Efficacy Outcomes by Evidence Source

Adjusted proportions of efficacy outcomes are shown for randomized controlled trial (RCT) treatment arms and real-world study (RWS) cohorts. Estimates were derived from mixed-effects meta-regression models using the prespecified imbalance-and-outcome-relevance adjustment algorithm. Blue squares indicate adjusted estimates for RCTs, and orange circles indicate adjusted estimates for RWS; horizontal bars represent 95% confidence intervals. The P value corresponds to the adjusted association between evidence source and the logit-transformed outcome proportion. Residual I² represents the proportion of residual between-study heterogeneity remaining after covariate adjustment, rather than the unadjusted heterogeneity from single-arm pooled estimates. The k column reports the number of contributing treatment arms or cohorts overall and by evidence source, shown as total (RCT/RWS). Listed covariates are those included in the outcome-specific adjusted model.

For safety outcomes, the adjusted differences varied across laboratory-based, symptom-based, composite, and management-based outcomes (Figure 3). Differences in laboratory-based adverse events, including grade 3 or higher neutropenia, leukopenia, and anemia, were largely attenuated after adjustment, with no significant differences between the 2 evidence sources. In contrast, both symptom-based adverse events, including diarrhea (6.2% vs 37.7%), nausea (8.7% vs 28.9%), and vomiting (4.2% vs 19.8%), and composite safety outcomes, including all-grade adverse events (77.1% vs 99.0%) and grade 3 or higher adverse events (32.6% vs 76.1%), were substantially lower in RWSs than in RCTs after adjustment; all corresponding P values were ≤.001. Among management-based adverse events, adverse event–related treatment discontinuation was similar between RWSs and RCTs (10.6% vs 8.6%; P = .523), whereas adverse event–related dose adjustment was significantly higher in RWSs than in RCTs (50.8% vs 33.4%; P = .016), a direction opposite to that observed for most other outcomes.

Figure 3. Primary Adjusted Estimates for Safety Outcomes by Evidence Source

Adjusted proportions of safety outcomes are shown for randomized controlled trial (RCT) treatment arms and real-world study (RWS) cohorts. Estimates were derived from mixed-effects meta-regression models using the prespecified imbalance-and-outcome-relevance adjustment algorithm. Blue squares indicate adjusted estimates for RCTs, and orange circles indicate adjusted estimates for RWS; horizontal bars represent 95% confidence intervals. The P value corresponds to the adjusted association between evidence source and the logit-transformed outcome proportion. Residual I² represents the proportion of residual between-study heterogeneity remaining after covariate adjustment, rather than the unadjusted heterogeneity from single-arm pooled estimates. The k column reports the number of contributing treatment arms or cohorts overall and by evidence source, shown as total (RCT/RWS). Listed covariates are those included in the outcome-specific adjusted model.

Sensitivity Analyses

Two prespecified sensitivity analyses were conducted to assess the robustness of the primary findings. In the core covariate model, which applied prespecified dimension-reduction rules across outcomes, the direction of results was consistent with the primary analysis, and differences in symptom-based adverse events remained highly significant (eFigures 6 and 7). The model jointly adjusting for first-line therapy proportion and palbociclib proportion addressed the most prominent compositional differences between the 2 evidence sources. In this model, the key differences observed in the primary analysis remained statistically significant after adjustment, including objective response rate, complete response, all-grade adverse events, grade 3 or higher adverse events, diarrhea, nausea, vomiting, and adverse event–related dose adjustment (eFigures 8 and 9). These findings suggest that the observed differences were not explained by treatment line or major differences in drug composition. In the additional study-dependence sensitivity analysis, most between-source effect estimates remained directionally consistent after excluding potentially overlapping reports and accounting for within-study clustering (eFigures 10 and 11). Reliable cluster-robust inference was available for 8 of 19 outcomes, with persistent differences observed for diarrhea (P=.010) and AE-related dose reduction (P=.007). For ORR, the direction and magnitude of the difference were preserved, although reliable cluster-robust inference was limited by the small number of independent real-world cohorts.

Risk-of-bias assessments of the underlying RCTs and RWSs are presented in eFigures 12 and 13; these assessments do not directly establish the risk of bias in the between-source comparisons. Funnel plots and Egger test results are provided in eFigure 14A-Q.

This study compared single-arm outcomes of CDK4/6 inhibitors for hormone receptor–positive, ERBB2-negative advanced or metastatic breast cancer across 31 RCTs and 55 RWSs. Three findings were central. First, RWSs generally reported higher response proportions and lower symptom-based adverse event proportions than RCTs. Second, after adjustment for observable study-level covariates, cross-source differences varied by outcome type: laboratory-based adverse event differences were attenuated or absent, symptom-based adverse event differences persisted, and management-based adverse events showed a distinct pattern. Third, key differences remained after further adjustment for treatment line and drug composition. These findings demonstrate outcome-specific differences in reported single-arm outcomes between RCTs and RWSs, which may be associated with differences in patient characteristics, treatment composition, outcome ascertainment, documentation, and reporting. However, the present analysis cannot establish the relative contributions of these factors or identify the mechanisms responsible for the observed differences.⁷˒¹²˒²¹˒²² The distinct patterns across laboratory-based, symptom-based, composite, and management-based safety outcomes further characterize these between-source differences. Laboratory-based adverse events rely on relatively objective and standardized testing procedures, which may explain why adjusted differences between evidence sources were small.²⁵˒³³ By contrast, symptom-based adverse events depend on active elicitation, patient reporting, clinician documentation, and completeness of medical records. The lower reported proportions of diarrhea, nausea, and vomiting in RWSs may be associated with differences in outcome ascertainment, clinical documentation, or selective reporting, although differences in actual event occurrence cannot be excluded.¹²˒⁴⁵⁻⁴⁸ Management-based adverse events are more closely linked to explicit clinical decision records. The higher proportion of dose adjustment in RWSs may reflect individualized dose management in routine care, although differences in dose-adjustment definitions and documentation may also contribute.³³˒³⁴

Response outcomes should also be interpreted through outcome generation mechanisms rather than as direct evidence of treatment-effect differences. RCTs typically use more standardized response assessment procedures, whereas response evaluation in RWSs may vary by assessment timing, imaging frequency, physician judgment, and documentation completeness.²¹⁻²³ RWS response rates are often calculated among evaluable patients, potentially excluding those with early loss to follow-up, rapid progression, or incomplete imaging assessment and thereby increasing observed response proportions.¹²˒¹³˒²¹ However, an exploratory analysis restricted to RWS cohorts using RECIST 1.1 criteria and all-treated denominators did not yield a lower pooled ORR, suggesting that differences in response criteria and denominator definitions alone may not fully explain the observed between-source discrepancies. Nevertheless, the contribution of these factors cannot be isolated from differences in patient characteristics, treatment composition, and other study-level factors.

For clinicians, guideline developers, regulatory agencies, and health technology assessment bodies, these findings suggest that the value of real-world data should be judged according to the evidence question.⁴˒¹²˒⁴⁹ High-quality real-world data can complement RCT evidence by characterizing patient populations treated in routine practice, treatment pathways, adherence, dose adjustment, discontinuation, and long-term management outcomes. It may also help assess whether a treatment has observable effectiveness in usual care settings.⁴˒¹²˒⁴⁹ However, when RWSs lack comparable internal control groups, single-arm response rates or adverse event proportions should not be treated as direct substitutes for RCT treatment-arm outcomes, particularly for relative effectiveness assessments or cross-source comparisons.⁷˒⁹˒⁵⁰

Safety evidence should also be interpreted according to ascertainment mechanism. Laboratory-based adverse events and clearly documented treatment management outcomes are generally more readily captured in real-world data and may provide useful complements to RCT evidence. In contrast, symptom-based adverse events depend on active elicitation, patient reporting, and clinical documentation, and may be systematically underestimated in routine care data.⁴⁵˒⁴⁶ Lower reported rates of diarrhea, nausea, or vomiting in RWSs therefore should not be interpreted simply as better real-world tolerability. Future real-world studies should report adverse event ascertainment methods, follow-up frequency, grading criteria, denominator definitions, and drug composition more transparently to improve interpretability for guideline, regulatory, and health technology assessment decisions.

This study complements previous RCT–RWE comparisons, which have more often focused on relative treatment effects from comparative designs or survival outcomes. By focusing on single-arm outcome proportions, this analysis demonstrates that the extent of agreement between RCTs and RWSs varies across outcome domains, potentially in relation to differences in outcome ascertainment and reporting. Stratifying safety outcomes by laboratory testing, symptom reporting, and treatment management decisions may help identify sources of cross-source differences and provides a reproducible framework for future evidence synthesis.

Limitations

The adapted, exploratory certainty assessment generally rated the single-arm evidence as low to very low (eTable 5), largely reflecting the noncomparative nature of the outcomes and substantial heterogeneity. This adaptation has not been specifically validated for cross-source comparisons of single-arm proportions and should not be interpreted as a conventional GRADE assessment of comparative treatment effects.

This study has several limitations. First, all analyses relied on aggregate-level data, and the observed between-source differences represent study-level associations subject to ecological bias rather than causal effects. Second, although measured patient and treatment characteristics were adjusted for, residual confounding may remain due to differences in patient selection, treatment patterns, calendar period, geographic setting, follow-up, outcome assessment, and clinical documentation. Accordingly, this study identifies outcome-specific discrepancies between RCTs and RWSs but cannot establish the mechanisms responsible for these differences. Third, CDK4/6 inhibitors in hormone receptor–positive, ERBB2-negative advanced or metastatic breast cancer were used as a case example, and generalizability to other drug classes or tumor types requires validation. Fourth, substantial residual heterogeneity persisted after adjustment, limiting the generalizability of average between-source estimates to individual study populations. In addition, no formal adjustment for multiple testing was applied, and individual P values should therefore be interpreted as exploratory. Fifth, median progression-free survival and overall survival were presented descriptively because of differences in follow-up duration and data maturity and were not used as primary cross-source comparison outcomes.

Real-world efficacy and safety outcomes for CDK4/6 inhibitors were not uniformly consistent with randomized trial outcomes. Discrepancies were most apparent for response outcomes and symptom-based adverse events, whereas laboratory-based and management-based safety outcomes showed greater concordance or different patterns. These findings indicate that real-world evidence should be interpreted according to the endpoint being evaluated rather than treated as uniformly reliable or unreliable. In clinical, regulatory, guideline, and health technology assessment settings, response rates and symptom-reported adverse events from routine-care data require particular caution because they may be sensitive to differences in assessment intensity, documentation, and reporting.

List of abbreviations

CDK4/6, cyclin-dependent kinase 4/6; CENTRAL, Cochrane Central Register of Controlled Trials; CI, confidence interval; ECOG, Eastern Cooperative Oncology Group; ERBB2/HER2, human epidermal growth factor receptor 2; MOOSE, Meta-analysis Of Observational Studies in Epidemiology; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses; PROSPERO, International Prospective Register of Systematic Reviews; RCT, randomized controlled trial; ROBINS-I, Risk Of Bias In Non-randomized Studies - of Interventions; RWE, real-world evidence; RWS, real-world study; SMD, standardized mean difference.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Availability of data and materials

All data analyzed in this study were extracted from previously published studies and publicly available sources cited in the manuscript and Additional file 1. The extracted aggregate study-level dataset, data dictionary, risk-of-bias assessments, and statistical code are available from the corresponding author on reasonable request. No individual participant data were collected.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported by the National Natural Science Foundation of China (grant No. 82403949) and the Shenzhen Municipal Science and Technology Innovation Committee Project (grant No. RCBS20231211090721033). The funders had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.

Authors' contributions

LS: Conceptualization, methodology, formal analysis, investigation, data curation, visualization, writing-original draft, writing-review and editing.

XY: Investigation, data curation, validation, writing-review and editing.

DZ: Methodology, validation, interpretation of findings, writing-review and editing.

MZ: Validation, interpretation of findings, writing-review and editing.

XK: Investigation, data curation, validation, writing-review and editing.

LZ: Investigation, data curation, validation, writing-review and editing.

JH: Clinical interpretation, supervision, writing-review and editing.

LT: Conceptualization, methodology, supervision, funding acquisition, project administration, writing-review and editing.

All authors had access to the relevant study data, critically revised the manuscript for important intellectual content, approved the final version of the manuscript, and agreed to be accountable for the accuracy and integrity of the work.

Acknowledgements

The authors thank all investigators of the randomized clinical trials and real-world studies included in this systematic review and meta-epidemiological analysis. Their published data made this comparative evidence assessment possible.

Additional file

Additional file 1. DOCX. Supplementary methods, tables, figures, and reporting checklists. Contains full database search strategies, detailed eligibility criteria and variable definitions, supplementary analyses, certainty-of-evidence assessments, reporting checklists, and eTables/eFigures.

1. US Food and Drug Administration. Real-World Evidence. Published April 2, 2026. Accessed May 18, 2026. https://www.fda.gov/science-research/science-and-research-special-topics/real-world-evidence

2. US Food and Drug Administration. Framework for FDA’s Real-World Evidence Program. Published December 2018. Accessed May 18, 2026. https://www.appliedclinicaltrialsonline.com/view/framework-fda-s-real-world-evidence-program-1

3. Heads of Medicines Agencies: HMA/EMA Joint Big Data Steering Group. Accessed May 18, 2026. https://www.hma.eu/about-hma/working-groups/hma/ema-joint-big-data-steering-group/hma/ema-joint-big-data-steering-group.html

4. Overview | NICE real-world evidence framework | Guidance | NICE. June 23, 2022. Accessed May 18, 2026. https://www.nice.org.uk/corporate/ecd9/chapter/overview

5. Orcutt X, Chen K, Mamtani R, Long Q, Parikh RB. Evaluating generalizability of oncology trial results to real-world patients using machine learning-based trial emulations. Nat Med. 2025;31(2):457-465. doi:10.1038/s41591-024-03352-5

6. Azoulay L. Rationale, Strengths, and Limitations of Real-World Evidence in Oncology: A Canadian Review and Perspective. The Oncologist. 2022;27(9):e731-e738. doi:10.1093/oncolo/oyac114

7. Bartlett VL, Dhruva SS, Shah ND, Ryan P, Ross JS. Feasibility of Using Real-World Data to Replicate Clinical Trial Evidence. JAMA Netw Open. 2019;2(10):e1912869. doi:10.1001/jamanetworkopen.2019.12869

8. Zong J, Rojubally A, Pan X, et al. A Review and Comparative Case Study Analysis of Real-World Evidence in European Regulatory and Health Technology Assessment Decision Making for Oncology Medicines. Value Health. 2025;28(1):31-41. doi:10.1016/j.jval.2024.09.007

9. Yap TA, Jacobs I, Baumfeld Andre E, Lee LJ, Beaupre D, Azoulay L. Application of Real-World Data to External Control Groups in Oncology Clinical Trial Drug Development. Front Oncol. 2021;11:695936. doi:10.3389/fonc.2021.695936

10. Ma X, Nussbaum NC, Magee K, et al. Comparison of real-world response rate (rwRR) to RECIST-based response rate in patients with advanced non-small cell lung cancer (aNSCLC). Ann Oncol. 2019;30:v651. doi:10.1093/annonc/mdz260.103

11. Lin L, Lucassen MJJ, van der Noort V, Egberts TCG, Beijnen JH, Huitema ADR. The feasibility of using real world data as external control arms in oncology trials. Drug Discov Today. 2025;30(3):104324. doi:10.1016/j.drudis.2025.104324

12. US Food and Drug Administration. Real-World Data: Assessing Electronic Health Records and Medical Claims Data to Support Regulatory Decision-Making for Drug and Biological Products. Published March 2, 2025. Accessed May 18, 2026. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/real-world-data-assessing-electronic-health-records-and-medical-claims-data-support-regulatory

13. Friends of Cancer Research. Considerations for Use of Real-World Evidence in Oncology. Published 2020. Accessed May 18, 2026.

14. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA. 1995;273(5):408-412. doi:10.1001/jama.273.5.408

15. Wood L, Egger M, Gluud LL, et al. Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: meta-epidemiological study. BMJ. 2008;336(7644):601-605. doi:10.1136/bmj.39465.451748.AD

16. Page MJ, Higgins JPT, Clayton G, Sterne JAC, Hróbjartsson A, Savović J. Empirical Evidence of Study Design Biases in Randomized Trials: Systematic Review of Meta-Epidemiological Studies. PloS One. 2016;11(7):e0159267. doi:10.1371/journal.pone.0159267

17. Wang Y, Parpia S, Couban R, et al. Compelling evidence from meta-epidemiological studies demonstrates overestimation of effects in randomized trials that fail to optimize randomization and blind patients and outcome assessors. J Clin Epidemiol. 2024;165:111211. doi:10.1016/j.jclinepi.2023.11.001

18. Lakdawalla DN, Shafrin J, Hou N, et al. Predicting Real-World Effectiveness of Cancer Therapies Using Overall Survival and Progression-Free Survival from Clinical Trials: Empirical Evidence for the ASCO Value Framework. Value Health J Int Soc Pharmacoeconomics Outcomes Res. 2017;20(7):866-875. doi:10.1016/j.jval.2017.04.003

19. Signorovitch J, Moshyk A, Zhao J, et al. Overall survival in the real-world and clinical trials: a case study validating external controls in advanced melanoma. Future Oncol. 2022;18(11):1321-1331. doi:10.2217/fon-2021-1054

20. Peters S, Salomonsen RJB, Diaz Perez I, et al. Real-world outcomes versus clinical trial results of first-line immunotherapy with or without chemotherapy in patients with metastatic non-small-cell lung cancer: the CORRELATE study. ESMO Open. 2026;11(1):106031. doi:10.1016/j.esmoop.2025.106031

21. Ma X, Bellomo L, Magee K, et al. Characterization of a Real-World Response Variable and Comparison with RECIST-Based Response Rates from Clinical Trials in Advanced NSCLC. Adv Ther. 2021;38(4):1843-1859. doi:10.1007/s12325-021-01659-0

22. Feinberg BA, Zettler ME, Klink AJ, Lee CH, Gajra A, Kish JK. Comparison of Solid Tumor Treatment Response Observed in Clinical Practice With Response Reported in Clinical Trials. JAMA Netw Open. 2021;4(2):e2036741. doi:10.1001/jamanetworkopen.2020.36741

23. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228-247. doi:10.1016/j.ejca.2008.10.026

24. Miller TP, Li Y, Getz KD, et al. Using electronic medical record data to report laboratory adverse events. Br J Haematol. 2017;177(2):283-286. doi:10.1111/bjh.14538

25. National Cancer Institute. CTCAE and AE Reporting. Published September 2, 2025. Accessed May 18, 2026. https://dctd.cancer.gov/research/ctep-trials/for-sites/adverse-events

26. Rugo HS, Brufsky A, Liu X, et al. Real-world study of overall survival with palbociclib plus aromatase inhibitor in HR+/HER2- metastatic breast cancer. NPJ Breast Cancer. 2022;8(1):114. doi:10.1038/s41523-022-00479-x

27. Finn RS, Martin M, Rugo HS, et al. Palbociclib and Letrozole in Advanced Breast Cancer. N Engl J Med. 2016;375(20):1925-1936. doi:10.1056/NEJMoa1607303

28. Hortobagyi GN, Stemmer SM, Burris HA, et al. Overall Survival with Ribociclib plus Letrozole in Advanced Breast Cancer. N Engl J Med. 2022;386(10):942-950. doi:10.1056/NEJMoa2114663

29. Goetz MP, Toi M, Huober J, et al. Abemaciclib plus a nonsteroidal aromatase inhibitor as initial therapy for HR+, HER2- advanced breast cancer: final overall survival results of MONARCH 3. Ann Oncol Off J Eur Soc Med Oncol. 2024;35(8):718-727. doi:10.1016/j.annonc.2024.04.013

30. Xu B, Zhang Q, Zhang P, et al. Dalpiciclib or placebo plus fulvestrant in hormone receptor-positive and HER2-negative advanced breast cancer: a randomized, phase 3 trial. Nat Med. 2021;27(11):1904-1909. doi:10.1038/s41591-021-01562-9

31. Provenzano L, Dieci MV, Curigliano G, et al. Real-world effectiveness comparison of first-line palbociclib, ribociclib or abemaciclib plus endocrine therapy in advanced HR-positive/HER2-negative BC patients: results from the multicenter PALMARES-2 study. Ann Oncol. 2025;36(7):762-774. doi:10.1016/j.annonc.2025.03.023

32. Duchnowska R, Soter K, Pogoda K, et al. Real-world outcomes with palbociclib, ribociclib, and abemaciclib plus endocrine therapy in HR+/HER2− advanced breast cancer: A multicenter retrospective study. Breast Off J Eur Soc Mastology. 2025;85:104665. doi:10.1016/j.breast.2025.104665

33. Stanciu IM, Parosanu AI, Nitipir C. An Overview of the Safety Profile and Clinical Impact of CDK4/6 Inhibitors in Breast Cancer-A Systematic Review of Randomized Phase II and III Clinical Trials. Biomolecules. 2023;13(9):1422. doi:10.3390/biom13091422

34. Pu D, Wu Y, Xu D, et al. The adverse events of CDK4/6 inhibitors for HR+/ HER2- breast cancer: an umbrella review of meta-analyses of randomized controlled trials. Front Pharmacol. 2024;15. doi:10.3389/fphar.2024.1269922

35. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. Int J Surg. 2021;88:105906. doi:10.1016/j.ijsu.2021.105906

36. Stroup DF, Berlin JA, Morton SC, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA. 2000;283(15):2008-2012. doi:10.1001/jama.283.15.2008

37. Sterne JAC, Savović J, Page MJ, et al. RoB 2: a revised tool for assessing risk of bias in randomised trials. BMJ. 2019;366:l4898. doi:10.1136/bmj.l4898

38. Sterne JA, Hernán MA, Reeves BC, et al. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ. 2016;355:i4919. doi:10.1136/bmj.i4919

39. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25):3083-3107. doi:10.1002/sim.3697

40. Viechtbauer W. Conducting Meta-Analyses in R with the metafor Package. J Stat Softw. 2010;36:1-48. doi:10.18637/jss.v036.i03

41. Balduzzi S, Rücker G, Schwarzer G. How to perform a meta-analysis with R: a practical tutorial. Evid Based Ment Health. 2019;22(4):153-160. doi:10.1136/ebmental-2019-300117

42. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560. doi:10.1136/bmj.327.7414.557

43. Higgins JPT, Thompson SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. J R Stat Soc Ser A. 2009;172(1):137-159. doi:10.1111/j.1467-985X.2008.00552.x

44. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315(7109):629-634. doi:10.1136/bmj.315.7109.629

45. Basch E, Jia X, Heller G, et al. Adverse symptom event reporting by patients vs clinicians: relationships with clinical outcomes. J Natl Cancer Inst. 2009;101(23):1624-1632. doi:10.1093/jnci/djp386

46. Atkinson TM, Dueck AC, Satele DV, et al. Clinician vs Patient Reporting of Baseline and Postbaseline Symptoms for Adverse Event Assessment in Cancer Clinical Trials. JAMA Oncol. 2020;6(3):437-439. doi:10.1001/jamaoncol.2019.5566

47. National Cancer Institute Division of Cancer Control and Population Sciences. Patient-Reported Outcomes Version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE) Increases Well-Tolerated Treatment and Optimizes Treatment Selection. Accessed May 19, 2026. https://cancercontrol.cancer.gov/research-emphasis/ending-cancer-as-we-know-it/patient-reported-outcomes?

48. Fares CM, Williamson TJ, Theisen MK, et al. Low concordance of patient-reported outcomes with clinical and clinical trial documentation. JCO Clin Cancer Inform. 2018;2:1-12. doi:10.1200/CCI.18.00059

49. Jaksa A, Arena PJ, Hanisch M, Marsico M. Use of Real-World Evidence in Health Technology Reassessments Across 6 Health Technology Assessment Agencies. Value Health. 2025;28(6):898-906. doi:10.1016/j.jval.2025.02.012

50. Lambert J, Lengliné E, Porcher R, Thiébaut R, Zohar S, Chevret S. Enriching single-arm clinical trials with external controls: possibilities and pitfalls. Blood Adv. 2023;7(19):5680-5690. doi:10.1182/bloodadvances.2022009167