In the realm of rare disease research, where patient populations often number in the mere hundreds globally, traditional randomized controlled trials (RCTs) face insurmountable hurdles like recruitment delays and ethical dilemmas in placebo use. Real-world evidence (RWE), derived from real-world data (RWD) sources such as electronic health records (EHRs), registries, and claims databases, offers a lifeline by providing larger, more diverse datasets to inform drug development and regulatory decisions. As of October 2025, the FDA’s updated guidance emphasizes RWE’s role in supplementing or even replacing RCTs for orphan drugs, particularly through externally controlled trials that integrate RWD to accelerate approvals. However, biostatisticians must navigate complex challenges in data integration to ensure validity, while deploying innovative solutions like synthetic controls and propensity score matching.
Key biostatistical challenges stem from RWD’s inherent limitations in rare diseases. Sparse data leads to low statistical power, wide confidence intervals, and vulnerability to finite sample bias, as seen in natural history studies (NHS) where small cohorts hinder reliable progression modeling. Heterogeneity across sources—EHRs capturing clinical details but lacking standardization, registries offering disease-specific insights but suffering from selection bias—complicates integration, risking confounding from unmeasured variables like time dependent changes in diagnostics or treatments. Confounding bias, immortal time bias, and violations of positivity assumptions further erode causal inference, especially in observational designs where RWD serves as external controls. Data quality issues, including missingness and inconsistent coding, amplify these problems, demanding rigorous assessment for “fit-for-purpose” use per FDA recommendations.
To surmount these, biostatisticians employ advanced data integration strategies. Standardization via common data models and FHIR APIs facilitates merging EHRs with registries, enabling comprehensive cohorts. Propensity score (PS) methods, such as matching or inverse probability weighting (IPW), balance covariates between treated and control groups, reducing confounding in hybrid designs that combine RCT data with RWD. For instance, PS matching has been pivotal in rare disease label expansions, like Prograf for transplants, by aligning registry data with trial populations. Synthetic control arms(SCAs) construct virtual comparators from weighted RWD, ideal for single-arm trials in conditions like Duchenne Muscular Dystrophy, where genetic matching ensures comparability and ethical feasibility. Bayesian approaches, including hierarchical modeling and power priors, borrow strength from historical RWD while inflating variances to account for bias, as demonstrated in rare events meta-analyses of vaccines like PPV23 against invasive pneumococcal disease. Targeted maximum likelihood estimation (TMLE) with super learning offers efficient, data-adaptive causal inference, minimizing variance in small samples. Tools like Python’s pandas for harmonization, R’s brms for Bayesian modeling, or SAS for sensitivity analyses streamline these workflows.
A compelling case is Skyclarys (omaveloxolone) for Friedreich Ataxia, approved in 2023 via an RCT augmented by PS-matched NHS data from registries, validating long-term efficacy through integrated RWE. Similarly, in diabetic ketoacidosis risk assessments for SGLT-2 inhibitors, design-adjusted synthesis of RWE with RCTs narrowed credible intervals, enhancing precision despite rare events. These examples underscore biostatisticians’ collaborative role with clinicians and regulators, using targeted learning roadmaps to define estimands, map causal graphs, and conduct sensitivity analyses for robustness.