Clinical trials and real-world evidence represent two fundamentally different ways of learning whether medical treatments work, and the tension between them shapes nearly every drug approval, insurance coverage decision, and treatment guideline you encounter. Randomized controlled trials (RCTs) have been called the “gold standard” for establishing whether a treatment causes a health benefit, while real-world evidence (RWE) draws on data from routine clinical practice to show how treatments perform outside of tightly controlled study conditions. Neither approach alone gives the full picture, and understanding where each excels and where each falls short has become one of the more consequential debates in modern medicine.
Why Randomized Trials Earned Their Reputation
The core strength of a randomized trial is its ability to isolate cause and effect. When researchers randomly assign patients to receive either the treatment being tested or a comparison (a placebo or an existing therapy), they create groups that are, on average, alike in every way except the treatment itself. That means any difference in outcomes can be attributed to the treatment rather than to some other factor like age, disease severity, or lifestyle. Randomization distributes these confounding factors across groups, and blinding (keeping patients and sometimes doctors unaware of who receives what) further reduces bias from expectations or subjective assessments.1PubMed Central. Selection of Control, Randomization, Blinding, and Allocation Concealment
This design is powerful enough that regulatory agencies and medical journals treat RCTs as the primary basis for claims about treatment effects. The AMA Manual of Style permits cause-and-effect language for randomized trials while instructing that observational studies use softer language like “association” or “correlation.”2JAMA. Randomized Trials vs Real-world Evidence: How Can Both Inform Decision-making? That distinction is not just editorial convention. It reflects a genuine difference in how confidently you can draw conclusions from each type of study.
But the very features that make RCTs strong also limit what they can tell you. Trials enroll a carefully selected group of patients, often excluding people with multiple health conditions, older adults, pregnant women, and others who represent large portions of the people who will eventually use the drug. Trials run for defined periods, often shorter than the years or decades patients will take a medication. And they happen in clinical settings where patients receive closer monitoring and more consistent care than they would in an average doctor’s office. The result is that what works in a trial does not always translate to the same results in everyday practice.
What Real-World Evidence Actually Is
Real-world evidence comes from analyzing data that was not collected for the purpose of a controlled experiment. Instead, it is drawn from the routine documentation of healthcare: electronic health records, insurance claims, product registries, disease registries, and similar sources.3NIH Pragmatic Trials Collaboratory. Acquiring Real-World Data: Common Real-World Data Sources When a doctor prescribes a medication and records the outcome in your chart, that information becomes part of the real-world data (RWD) ecosystem. When your insurer processes a claim for a hospitalization, that is another data point. When a device manufacturer tracks implant performance through a registry, that too feeds the pool.
The appeal is scale and representativeness. A single hospital system’s electronic records might capture tens of thousands of patients on a given drug, including the elderly, the chronically ill, and the noncompliant, all the people trials tend to exclude. And because these records accumulate over years, they can reveal patterns that a two-year trial would miss entirely: slow-developing side effects, interactions with other medications, and real-world adherence patterns where patients skip doses or stop treatment altogether.
The weakness is that nobody designed this data for research purposes. A doctor’s note about a patient’s blood pressure was written to guide that patient’s care, not to answer a research question. Missing values, inconsistent coding, and records scattered across unconnected systems all introduce noise. And because patients were not randomly assigned to treatments, comparing outcomes between groups is fraught with confounding. Sicker patients may have been prescribed a different drug than healthier ones, and disentangling the drug’s effect from the patient’s baseline health is where much of the methodological difficulty lies.
The Gap Between Trial Results and Real-World Outcomes
One of the most striking findings in this area comes from cancer treatment. A study comparing clinical trial results with real-world outcomes for 29 cancer drug indications found that nearly all of them (97%) showed worse survival in real-world settings, with a median difference of about five months. The overall pattern suggested that patients treated outside of trials faced roughly 58% higher mortality compared to trial participants.4PubMed Central. Assessing the efficacy-effectiveness gap for cancer therapies: A comparison of overall survival and toxicity between clinical trial and population-based, real-world data for contemporary parenteral cancer therapeutics Hospitalization rates were also about 14 percentage points higher in the real world than what trial reports showed for serious adverse events.
This gap between efficacy (what a treatment achieves under ideal conditions) and effectiveness (what it achieves in routine care) is not unique to oncology, but cancer data illustrates it starkly. Several factors drive it. Trial participants tend to be younger, fitter, and more closely monitored. They receive care at academic medical centers with experienced teams. They are more likely to complete the full course of treatment. And trial protocols often exclude patients with complicating conditions that are common in the real patient population. None of this means trial results are wrong, but it does mean they describe a best-case scenario that most patients will not replicate.
How Regulators Are Using Real-World Evidence
The U.S. Food and Drug Administration’s relationship with real-world evidence has evolved considerably. The legislative push for requiring drug effectiveness began in the early 1960s, prompted by the thalidomide disaster, when Congress passed laws that the FDA enforced largely through placebo-controlled randomized trials. By the 1980s, the RCT had been widely labeled the gold standard. But the 21st Century Cures Act of 2016 sent a clear mandate for the FDA to evaluate new forms of evidence, including RWE.5Journal of Orthopaedic Trauma. Randomized Clinical Trial or Real-World Evidence: How Historical Events, Public Demand, and the Resulting Laws and Regulations Shaped the Body of Medical Evidence
This is not just theoretical. An analysis of FDA labeling expansions for small molecules and biologics from 2022 through 2024 found that roughly a quarter of approvals included real-world evidence in the supporting documentation, with the proportion hovering between 23% and 28% each year. Oncology led the way, accounting for about 44% of submissions containing RWE, followed by infectious disease and dermatology.6PubMed Central. Real-World Evidence in FDA Approvals for Labeling Expansion of Small Molecules and Biologics The FDA has been especially receptive to RWE for effectiveness comparisons in rare diseases and cancers, where running traditional randomized trials is difficult or sometimes ethically impossible.7PubMed. Application of Real-World Evidence to Support FDA Regulatory Decision Making
That said, real-world evidence does not replace trials in the regulatory framework. It supplements them. The most common pattern is that a drug is initially approved based on trial data, and then real-world evidence is used later to expand the label to new patient populations, new indications, or new dosing regimens. In some rare disease contexts, RWE has served as the external comparator arm when enrolling a control group in a trial would be impractical or unethical.
Rare Diseases and Special Populations
Rare diseases present a unique challenge for evidence generation. By definition, few patients exist, which makes it difficult to enroll enough people for a well-powered randomized trial. Some conditions are so uncommon that a single trial site might see only a handful of eligible patients over several years. In these situations, real-world evidence from patient registries, electronic health records, and insurance claims can provide the clinical context that trials alone cannot.
A systematic review of RWE used in FDA approvals for rare diseases found that well-conducted real-world studies can strengthen the case for a new treatment, especially when a large treatment effect is observed. RWE in this context most often serves to provide historical comparisons (how patients with the condition fare on existing treatments or no treatment) and to contextualize trial results within the broader patient population.8PubMed Central. A systematic review of real-world evidence supportive of new drug and biologic license application approvals in rare diseases
The same logic applies to other underrepresented groups. Pediatric patients, pregnant women, elderly patients with multiple conditions, and people from underrepresented racial and ethnic backgrounds are routinely excluded from or underenrolled in clinical trials. Real-world data can fill these gaps by documenting how treatments actually perform in populations that trials never studied. This is not a matter of preference. For some groups, it is the only ethical source of clinical evidence available.
Safety Monitoring After Approval
Clinical trials are designed primarily to measure efficacy, and their ability to detect safety problems is limited. A trial with a few hundred or even a few thousand participants, running for a year or two, will catch common side effects. But rare adverse events that occur in one out of every 10,000 patients, or problems that take years to develop, are essentially invisible to trials. This is why real-world data plays such a critical role in post-market safety surveillance.
Linked real-world data from electronic health records, claims databases, and adverse event reporting systems offers a more comprehensive view of safety outcomes, including rare events, complications from off-label use, and long-term risks that trials cannot capture.9PubMed Central. Leveraging real-world data for safety signal detection and risk management in pre- and post-market settings The FDA’s Adverse Event Reporting System (FAERS), for instance, collects spontaneous reports of suspected drug side effects from patients, healthcare providers, and manufacturers. Researchers use data-mining techniques on this database to detect signals, patterns of reported events that occur more often than expected for a given drug, and then investigate whether the signal represents a genuine safety concern.10PubMed. Disproportionality Analysis of Abemaciclib in the FDA Adverse Event Reporting System: A Real-World Post-Marketing Pharmacovigilance Assessment
This is an area where RWE is not competing with clinical trials but filling a gap that trials were never designed to address. Many of the most consequential drug safety discoveries of the past few decades, from cardiovascular risks of certain pain medications to rare blood clots associated with specific vaccines, emerged from post-market real-world surveillance rather than from the original approval trials.
The Methodological Challenges That Keep Researchers Up at Night
The biggest methodological concern with real-world evidence is confounding. In a trial, randomization ensures the treatment and control groups are comparable. In observational data, doctors choose treatments based on patient characteristics, meaning the treatment group and the comparison group are almost never directly comparable. A sicker patient might receive the more aggressive treatment, so if that patient has worse outcomes, you cannot tell whether the drug failed or whether the patient was just sicker to begin with.
Researchers have developed statistical tools to address this. Propensity score methods, which estimate the probability that a patient would receive a given treatment based on their recorded characteristics and then use that score to create more balanced comparisons, have become widely used in healthcare database research.11Endocrine Reviews. Conducting Real-world Evidence Studies on the Clinical Outcomes of Diabetes Treatments But propensity scores only account for factors that are measured and recorded. If a key variable, like disease severity or a patient’s motivation to follow treatment, is not captured in the data, no statistical method can adjust for it.
Selection bias adds another layer. The patients who appear in a particular database are there for a reason: they have insurance (claims data), they visited a specific hospital system (EHR data), or they enrolled in a registry. None of these are random samples of the general population. Researchers have proposed weighting approaches to reduce selection bias, but the problem is inherent to data that was collected for non-research purposes.12PubMed Central. A framework for understanding selection bias in real-world healthcare data
Then there are the well-known investigator errors that plague observational research: immortal time bias (artificially inflating survival because you only count patients who lived long enough to start treatment), adjustment for variables that lie on the causal pathway between treatment and outcome, and reverse causation (confusing the direction of cause and effect).11Endocrine Reviews. Conducting Real-world Evidence Studies on the Clinical Outcomes of Diabetes Treatments These are not obscure concerns. They regularly produce findings that later turn out to be wrong, which is why RWE has had a credibility problem even as its use has expanded.
Target Trial Emulation
One of the most important methodological advances in recent years is a framework called target trial emulation. The idea is straightforward in principle: before touching the observational data, researchers first design the hypothetical randomized trial that would ideally answer their question. They define who would be eligible, what treatment comparison they would make, when follow-up would start, and what outcome they would measure. Then they use the observational data to emulate that trial as closely as possible.13PubMed Central. The Target Trial Framework for Causal Inference From Observational Data: Why and When Is It Helpful?
This sounds simple, but the discipline it imposes prevents many of the biases that have historically undermined observational research. Immortal time bias, for example, becomes nearly impossible when you define “time zero” the same way a trial would. Vague eligibility criteria that inadvertently mix different patient populations get caught when you have to write an explicit protocol. The approach has been credited with dramatically improving the quality of causal inference from observational data and has gained popularity across multiple medical specialties.14PubMed Central. Target Trial Emulation to Improve Causal Inference from Observational Data: What, Why, and How?
Pragmatic Trials as a Middle Ground
The clean division between “clinical trials” and “real-world evidence” obscures an increasingly important hybrid: the pragmatic clinical trial. These studies maintain randomization, the core feature that allows causal inference, but they run in real-world clinical settings rather than in the tightly controlled environments of traditional trials. They enroll broader patient populations, use routine care procedures, and measure outcomes that matter to patients and clinicians rather than surrogate biomarkers.15PubMed Central. Pragmatic Clinical Trials and Real-World Evidence: An Introduction
Pragmatic trials aim to answer effectiveness questions from the start, rather than leaving the translation from efficacy to effectiveness as an exercise for real-world observation after the fact. They can provide evidence on how a treatment strategy performs in routine practice while retaining the strength of randomization.16PubMed. Pragmatic trials and real world evidence: Paper 1. Introduction The trade-off is that pragmatic trials sacrifice some internal control. With less restrictive eligibility criteria and more variable care settings, there is more noise in the data, and isolating the treatment effect requires larger sample sizes.
Decentralized clinical trials push this boundary further. Using telemedicine, mobile health providers, and wearable monitoring devices, these studies allow patients to participate from home rather than traveling to a study site. Wearable devices can continuously measure physiological parameters like heart rhythm, activity levels, or glucose, creating a stream of real-world data within a trial framework.17PubMed Central. The role of remote data capture, wearables, and digital biomarkers in decentralized clinical trials The result is data that is both more representative of how patients live and more granular in what it measures.
Data Quality and the Interoperability Problem
For real-world evidence to be credible, the underlying data needs to be consistent, complete, and structured in ways that allow meaningful analysis across different sources. This is harder than it sounds. A diagnosis code in one hospital system might mean something slightly different from the same code in another. Lab results may be stored in different units. Medication records may not capture what the patient actually took, only what was prescribed.
Common data models have been developed to address this, essentially standardizing how different data sources organize their information so that analyses can run across multiple institutions without manual translation. A systematic review evaluating several of these models found that the OMOP Common Data Model scored highest overall, followed by PCORnet. Among data exchange standards, FHIR achieved the maximum score in every category evaluated, with OpenEHR as a strong alternative.18PubMed Central. Common data models and data standards for tabular health data: a systematic review The adoption of these standards has been uneven, but the trend is toward greater harmonization, which is essential if real-world evidence is to be compared or combined across different health systems.
Artificial intelligence is also playing a growing role in making real-world data usable for research. Much of the most clinically rich information in health records sits in unstructured text: doctors’ notes, pathology reports, radiology readings. Natural language processing systems can extract structured data from these free-text fields, pulling out diagnoses, treatment responses, genetic test results, and other information that would otherwise require manual chart review at enormous cost.19PubMed Central. Generating real-world evidence from unstructured clinical notes to examine clinical utility of genetic tests: use case in BRCAness Some platforms combine rule-based algorithms with machine learning and neural networks to handle both structured and unstructured fields.20JTO Clinical and Research Reports. Automating Access to Real-World Evidence The quality of these extraction tools matters enormously, because errors at the data stage propagate through every downstream analysis.
Insurance Coverage and Health Technology Assessment
The comparison between RCTs and RWE is not just an academic exercise. It directly affects which treatments get covered by insurers and at what price. Health technology assessment (HTA) bodies, the organizations that advise governments and payers on whether a new drug or device is worth paying for, have traditionally relied on trial data to judge a product’s value. But HTA agencies increasingly want to see real-world evidence as well, particularly for questions that trials do not answer: how long patients actually stay on a drug, what happens to patients who would have been excluded from the trial, and whether a treatment’s benefits hold up over years of use in routine practice.21PubMed Central. Real-world evidence to support health technology assessment and payer decision making: is it now or never?
For drug manufacturers, this creates a dual evidence burden. The trial that earns regulatory approval may not be sufficient to convince a payer that the treatment is cost-effective. Real-world studies demonstrating sustained benefits, manageable side effects, and practical adherence rates in a broader population can be the difference between widespread reimbursement and limited access. For patients, this means that how well a treatment is studied in the real world directly influences whether their insurance will cover it.
Ethical and Privacy Dimensions of Real-World Data
Using patient data generated during routine care for research purposes raises a distinct set of ethical questions. In a traditional clinical trial, patients give informed consent: they know they are in a study, they understand the risks, and they agree to participate. Real-world evidence studies often use data collected before any research question was formulated, sometimes years earlier. Patients may not know their records are being used for research, and the original consent they gave for medical care did not contemplate this use.22PubMed Central. Ethical considerations for real-world evidence studies
De-identification is the standard solution: stripping records of names, dates, and other identifying information before analysis. But de-identification is not foolproof. With enough data points, it is sometimes possible to re-identify individuals, especially in small patient populations like rare diseases. Data sharing across institutions, which is necessary for the multi-site analyses that give RWE its statistical power, multiplies the points at which breaches can occur. These are not hypothetical concerns. They shape how institutional review boards evaluate RWE protocols and how regulatory agencies set standards for the data that supports drug approvals.
There is also a subtler ethical issue about representativeness. If real-world data is drawn primarily from large health systems and insured populations, it may systematically exclude uninsured patients, rural communities, and people who face barriers to healthcare access. Evidence derived from these datasets then reflects the experience of those who already have better access to care, potentially widening rather than narrowing health disparities. Awareness of this limitation is growing, but practical solutions remain uneven.