Real-World Evidence Studies

Validate RWD against source documents (medical records, billing records) to assess accuracy of diagnoses, treatments, and outcomes. Random audit of 50–200 records often reveals error rates (misclassification, missing data). Use validated case definitions (e.g., published algorithms for identifying diabetes from claims codes) rather than single diagnosis codes. Test sensitivity/specificity of case definitions against manual review. Document limitations (coding errors, missing variables) and perform sensitivity analyses using alternative definitions. Use multiple data sources (e.g., link EHR + claims to increase completeness). Transparent reporting of validation efforts increases confidence in findings.

Both are 'real-world' in spirit, but differ fundamentally: Pragmatic trials are randomized (exposure allocated by randomization), though in real settings. RWE is observational (no randomization, exposure determined by clinical practice). Pragmatic trials maintain randomization's causal power while testing in real practice. RWE sacrifices randomization for speed and scale. Pragmatic trials enroll prospectively (researchers know enrollment); RWE often leverages retrospectively collected data. Pragmatic trials can assess implementation factors; RWE naturally captures real-world implementation. For establishing causality, pragmatic trials are stronger. For rapid, broad evidence, RWE is faster.

Propensity score matching: calculate likelihood of treatment given baseline variables (age, sex, comorbidities, etc.), match treated to untreated on propensity score, compare outcomes. This mimics randomization on observed variables. Regression adjustment: include exposure and potential confounders in logistic/Cox regression, report adjusted effect. Inverse probability weighting (IPW): weight observations by inverse propensity, creating pseudo-population where exposure is independent of confounders. Stratification: divide population by confounder (e.g., age groups), compare exposure within strata. Choose method based on sample size and overlap (region of positivity: do treated and untreated exist at all confounder values?). Document balance of baseline characteristics before and after adjustment. Always acknowledge that only measured confounders are controlled; unmeasured confounding remains.

No. RWE and trials serve different purposes and are complementary. Trials establish efficacy (can it work in ideal conditions?) with high internal validity. RWE establishes effectiveness (does it work in real practice?) with high external validity but lower internal validity (confounding). For approving new drugs or devices, regulatory agencies require trials demonstrating efficacy and safety in defined populations. RWE supports trials by testing generalizability, assessing safety in excluded populations, quantifying real-world effectiveness, and informing post-approval decisions. The gold standard is both: trials for causal efficacy evidence, RWE for real-world application and long-term outcomes.