by Mathieu Ghadanfar, MD, FESC, FAHA
Cardiovascular medicine has entered an era of extraordinary randomized evidence. DAPA-HF, EMPEROR-Reduced, EMPA-KIDNEY, the SELECT trial (1, 2, 3): each enrolling tens of thousands of patients, each reshaping guidelines. And yet the questions that dominate daily clinical decision-making remain unanswered by any of them. No large randomized trial has directly compared empagliflozin against dapagliflozin. None has characterized treatment effects in elderly patients with stage 4 chronic kidney disease (CKD) and atrial fibrillation who could not have qualified for the pivotal studies. None tells us what happens to the cardiorenal benefit curve at year five, when most trials had already closed. Real-world evidence (RWE), derived from secondary use of routine healthcare data, is increasingly positioned to answer exactly these questions—provided it meets the rising methodological and regulatory standards that now define the field.
A Regulatory Inflection Point
In March 2024, the FDA released draft guidance on Real-World Evidence for Non-Interventional Studies, emphasizing prespecification, bias mitigation, and fit-for-purpose data (4). In May 2024, EMA issued its reflection paper outlining parallel methodological expectations for regulatory use of RWD (5). These signals do not equate RWE with randomized trials, but they raise the bar decisively: observational cardiovascular programs must now be causal, transparent, and decision-grade. Critically, both agencies demand prespecified protocols and early scientific advice before data are accessed—a discipline most cardiovascular observational programs have not historically applied. The implication for sponsors is direct: a study designed retrospectively to support a regulatory submission will not pass scrutiny under these frameworks.
Secondary Use of Data in Cardiovascular Medicine
Secondary data sources (claims, electronic health records - EHRs, registries, and emerging digital streams) offer scale and longitudinal follow-up unavailable to any single trial, but each carries structural limitations. Claims files contain no ejection fraction, no New York Hearth Association (NYHA) functional class, no biomarkers. In EHR-based analyses, structured data systematically underperforms for key cardiovascular endpoints.
Validation studies have shown that myocardial infarction (MI) is captured with a recall of under 30% in structured EHR problem lists, not because the diagnosis is missing from the record, but because clinicians routinely document prior MI as a brief notation in free-text notes rather than entering it as a coded diagnosis. Automated extraction from structured fields alone misses it entirely, making chart abstraction or Natural Language Processing (NLP) mandatory before MI can serve reliably as an endpoint or eligibility criterion.
Target Trial Emulation: The New Baseline
Target trial emulation (TTE), formalized by Hernán and colleagues (6, 7), requires investigators to explicitly specify the hypothetical randomized trial their observational analysis aims to emulate before touching the data. The framework eliminates several classes of bias that have undermined cardiovascular observational research for decades. Immortal time bias, for instance, arises when patients must survive a period of follow-up before being classified as treated, attributing survival time to drug exposure that preceded it—an artefact that can generate apparent hazard ratios below 0.70 for treatments that have no real effect on mortality. Prevalent user bias inflates apparent benefit by excluding patients who already discontinued therapy due to side effects or progression. TTE addresses both by anchoring the analysis to treatment initiation at a prespecified time zero, mirroring the randomization moment of a trial. In September 2025, JAMA published the TARGET Statement, codifying reporting standards for trial emulation studies equivalent to CONSORT for RCTs (8). As illustrated in Figure 1, the methodological contrast between a naïve analysis and a properly emulated trial is not a question of statistical refinement, it is the difference between a finding that survives regulatory scrutiny and one that does not.
Figure 1. Three structural biases: immortal time bias, prevalent user bias, and confounding by indication, drive the flawed results that characterize naïve observational analyses CV research (left), including overestimated treatment benefit and false associations. Target trial emulation (right) addresses each directly: defined eligibility criteria with washout enforce a clean new-user start; rigorous confounding adjustment replaces uncontrolled channeling. The output is not simply a more cautious estimate, it is a study design capable of generating credible, decision-grade conclusions.
Where Cardiovascular RWE Adds the Most Value
RWE is particularly valuable in four domains. First, comparative effectiveness when head-to-head trials are absent: no randomized trial has compared empagliflozin against dapagliflozin, yet formulary and guideline decisions require this signal, a gap that a well-designed target trial emulation using Danish registry data began to address in 2024. Second, treatment effects in populations systematically excluded from trials: patients with estimated glomerular filtration rate (eGFR) below 20, severe frailty, or complex polypharmacy represent a large share of real prescribing but a marginal share of pivotal trial enrolment. Third, long-term outcome trajectories beyond trial duration: DAPA-HF followed patients for a median of 18 months; patients now take dapagliflozin for years, and the eGFR trajectory beyond that window is a legitimate clinical and regulatory question (1). Fourth, implementation gaps and therapy underuse: despite Class I guideline recommendations, Sodium-Glucose Co-Transporter 2 (SGLT2) inhibitors and Glucagon-like peptide 1 (GLP-1) receptor agonists remain dramatically underprescribed in eligible patients with heart failure and CKD, quantifying and correcting this gap is a domain where RWE has no substitute.
A Practical Illustration
Consider what a flawed study looks like when it reaches a formulary committee or a guideline panel. A naïve analysis defines exposure as any prescription within six months before an outcome and reports a dramatic mortality benefit—an association driven largely by immortal time: the treated patients, by construction, survived long enough to receive a prescription. A regulatory-grade emulation defines time zero at initiation, enforces a new-user design that excludes prevalent users who already tolerated the drug, aligns eligibility criteria to the target population, and prespecifies causal estimands with preregistered sensitivity analyses. The difference is not cosmetic. When the RCT-DUPLICATE initiative systematically compared cardiovascular trial emulations against their original RCTs, alignment depended almost entirely on whether these design disciplines had been applied. Studies that skipped them diverged. Studies that applied them converged. That is the operational meaning of “decision-grade” evidence.
What RWE Still Cannot Fully Solve
Residual confounding persists regardless of analytical sophistication, particularly confounding by indication: the cardiologist’s prescription decision encodes clinical intuitions about prognosis that no database captures. Endpoint misclassification in structured EHRs is not a marginal concern, it is a validity threat requiring chart abstraction or NLP-based phenotyping for any study where cardiovascular events are endpoints. And the regulatory acceptance boundary deserves plain statement: to date, no cardiovascular RWE study has achieved single-arm adequacy as primary evidence for a marketing authorization in a competitive indication. Both FDA and EMA have used RWE to support label extensions and post-marketing commitments, but the threshold for replacing a controlled trial in a primary efficacy submission has not yet been crossed in this therapeutic area. Sponsors who design studies without that boundary in mind risk producing evidence that satisfies publication reviewers but fails regulatory scrutiny. The workflow in Figure 2 maps the design disciplines that separate studies likely to survive that scrutiny from those that will not.
Figure 2. Target Trial Emulation Workflow: design steps that distinguish a credible cardiovascular non-interventional study from a naïve observational analysis, mapped against the corresponding RCT design element each step emulates.
Conclusion
The trial-first, real-world-second paradigm is breaking down in cardiovascular medicine, not because randomized trials have lost their authority, but because the therapeutic landscape has outpaced their capacity to answer every relevant question. Cardiovascular RWE, when anchored in causal inference frameworks and aligned with the regulatory standards now in force, is becoming an essential and non-substitutable component of the evidence ecosystem. The starting point for any serious program is no longer a dataset. It is a clearly specified question: what is the exact trial I am trying to emulate, and what would a rigorous reviewer at the FDA or EMA conclude about my ability to answer it from the data I have? That question, asked early and answered honestly, is what separates cardiovascular RWE that changes practice from cardiovascular RWE that fills journals.
References
- McMurray JJV, Solomon SD, Inzucchi SE, Køber L, Kosiborod MN, Martinez FA, Ponikowski P, Sabatine MS, Anand IS, Bělohlávek J, Böhm M, Chiang CE, Chopra VK, de Boer RA, Desai AS, Diez M, Drozdz J, Dukát A, Ge J, Howlett JG, Katova T, Kitakaze M, Ljungman CEA, Merkely B, Nicolau JC, O'Meara E, Petrie MC, Vinh PN, Schou M, Tereshchenko S, Verma S, Held C, DeMets DL, Docherty KF, Jhund PS, Bengtsson O, Sjöstrand M, Langkilde AM; DAPA-HF Trial Committees and Investigators. Dapagliflozin in Patients with Heart Failure and Reduced Ejection Fraction. N Engl J Med. 2019 Nov 21;381(21):1995-2008. doi: 10.1056/NEJMoa1911303. Epub 2019 Sep 19. PMID: 31535829.
- EMPA-KIDNEY Collaborative Group. Design, recruitment, and baseline characteristics of the EMPA-KIDNEY trial. Nephrol Dial Transplant. 2022 Jun 23;37(7):1317-1329. doi: 10.1093/ndt/gfac040. PMID: 35238940; PMCID: PMC9217655.
- Lincoff AM, Brown-Frandsen K, Colhoun HM, Deanfield J, Emerson SS, Esbjerg S, Hardt-Lindberg S, Hovingh GK, Kahn SE, Kushner RF, Lingvay I, Oral TK, Michelsen MM, Plutzky J, Tornøe CW, Ryan DH; SELECT Trial Investigators. Semaglutide and Cardiovascular Outcomes in Obesity without Diabetes. N Engl J Med. 2023 Dec 14;389(24):2221-2232. doi: 10.1056/NEJMoa2307563. Epub 2023 Nov 11. PMID: 37952131.
- US Food and Drug Administration. Real-World Evidence: Considerations Regarding Non-Interventional Studies for Drug and Biological Products. Draft Guidance. March 2024.
- European Medicines Agency. Reflection Paper on the Use of Real-World Data in Non-Interventional Studies to Generate Real-World Evidence. EMA/CHMP/150527/2024. 2024.
- Hernán MA, Sauer BC, Hernández-Díaz S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016; 79:70–75.
- Hernán MA, Dahabreh IJ, Dickerman BA, Swanson SA. The target trial framework for causal inference from observational data: why and when is it helpful? Ann Intern Med. 2025;178(3):402–407. doi:10.7326/ANNALS-24-01871.
- TARGET Statement Writing Group. Transparent Reporting of Observational Studies Emulating a Target Trial (TARGET Statement). JAMA. 2025;334(12):1084–1093. doi:10.1001/jama.2025.13350.
- Wang SV, Pottegård A, Crown W, et al. HARmonized Protocol Template to Enhance Reproducibility of Hypothesis Evaluating Real-World Evidence Studies (HARPER). Pharmacoepidemiol Drug Saf. 2023;32(1):44–55.
- Wang SV, Pottegård A, Crown W, et al. Emulation of randomized clinical trials with nonrandomized database analyses: results of 32 clinical trials. JAMA. 2023;329(16):1376–1385.
