Real World Data (RWD)

Real world data refers to health data collected outside of conventional clinical trials, drawn from sources that reflect routine clinical practice, patient behavior, and healthcare system operations. Common RWD sources include electronic health records (EHRs), medical and pharmacy claims databases, disease registries, patient generated health data from wearable devices and mobile apps, and vital statistics registries.

Clinical trials, while rigorous, enroll selected populations under controlled conditions that may not reflect everyday practice. RWD captures what happens to diverse patient populations in real clinical settings, including patients who are typically excluded from trials (such as those with multiple comorbidities, extremes of age, or rare conditions). Regulatory agencies such as the FDA and EMA increasingly accept evidence derived from RWD, known as real world evidence (RWE), in decisions on expanded indications, post marketing surveillance, and safety monitoring. In the United States, the 21st Century Cures Act of 2016 directed the FDA to evaluate the use of RWE to help support approval of new indications for approved drugs and to help support or satisfy post approval study requirements, and the FDA's 2018 framework for its RWE program defines RWD as data relating to patient health status and/or the delivery of health care routinely collected from a variety of sources. The EMA's definition is similar but explicitly limits RWD to sources other than traditional clinical trials. Health systems use RWD for quality measurement, resource planning, and comparative effectiveness research. The growing digitization of healthcare has made RWD increasingly available, comprehensive, and analytically tractable.

Most RWD is generated as a byproduct of healthcare delivery and patient activities, rather than being collected according to a research protocol; registries, which collect standardized data for defined purposes, are a partial exception. Electronic health records contain clinical notes, laboratory results, imaging reports, medication orders, and procedure records. Claims databases capture diagnoses, procedures, and costs submitted for reimbursement. Disease registries collect standardized data on patients with specific conditions for surveillance and quality improvement. Patient generated data from wearables and apps provide continuous physiological measurements and patient reported outcomes.

Using RWD for research requires careful attention to data quality, completeness, and structure. Unlike trial data, most RWD is not collected according to a standardized research protocol, and missing data, coding inconsistencies, and measurement variability are common. Researchers must clean and validate the data, apply appropriate inclusion and exclusion criteria, and use analytical methods designed for observational data (such as propensity score methods, instrumental variables, or difference in differences) to address confounding and selection bias.

Common data models, such as OMOP (Observational Medical Outcomes Partnership) and Sentinel, have been developed to standardize the structure and terminology of RWD across institutions, enabling multicenter studies and reproducible analyses. Data privacy regulations (such as HIPAA in the United States and GDPR in the European Union) govern access to and use of RWD. Under HIPAA, for example, research typically uses deidentified data, a limited data set under a data use agreement, or identifiable data with patient authorization or a waiver approved by an institutional review board or privacy board; under GDPR, pseudonymized data remain personal data that need a lawful basis, and only truly anonymous data fall outside the regulation.

Use RWD when the research question concerns the effectiveness, safety, or utilization of treatments in routine clinical practice, when a randomized trial is infeasible (due to ethical, logistical, or sample size constraints), or when the goal is to study rare outcomes, long term effects, or diverse populations not represented in trials. RWD is also appropriate for post marketing surveillance, health system performance measurement, and comparative effectiveness research. It is less appropriate for establishing initial efficacy, where the controlled conditions of a clinical trial are needed to isolate the treatment effect.

A common error is treating RWD analyses as equivalent to randomized trial evidence without acknowledging the potential for unmeasured confounding. Another mistake is using administrative codes (such as ICD diagnosis codes) as outcome definitions without validating them against clinical data, which can lead to misclassification. Researchers sometimes fail to account for immortal time bias, which occurs when follow up includes a period during which the outcome could not have occurred in one group (for example, the time between cohort entry and a first prescription), and that period is misclassified or ignored. Finally, ignoring data provenance (understanding how and why the data were generated) can lead to analyses that misinterpret administrative artifacts as clinical phenomena.

In a hypothetical example, a large integrated health system uses electronic health record data from 85,000 patients with type 2 diabetes to compare cardiovascular outcomes between two classes of glucose lowering medications. Propensity score matching on 32 baseline covariates creates 22,000 matched pairs. Over a median follow up of 3.2 years, the newer drug class is associated with a 15% lower rate of major adverse cardiovascular events (hazard ratio 0.85, 95% CI: 0.77 to 0.94). The results are consistent with findings from a recent cardiovascular outcomes trial, which is consistent with the idea that the trial's efficacy translates to real world effectiveness, although unmeasured confounding cannot be excluded.