Real-World Evidence & Registry Data

Real-world evidence (RWE) is generated from routinely-collected real-world data — disease registries, electronic health records (EHRs), and administrative or insurance-claims databases — rather than from a controlled trial. It captures what happens to ordinary patients in ordinary care, which is exactly its strength and its weakness: it is generalisable but it is observational.

How treatments, devices and pathways actually perform in unselected patients over long periods. RWE shines where RCTs struggle: it is large and generalisable, good for rare events and harms, long-term and real-world effectiveness, and questions where a trial would be impractical or unethical to run.

First ask the question type. For incidence, harms, prognosis or coverage, RWE is often ideal. For a causal treatment effect, be sceptical: the groups were never randomised, so they differ systematically — sicker patients get the active treatment (confounding by indication). Look for rigorous adjustment (matching, propensity scores, multivariable models) — and remember even the best adjustment leaves residual confounding.

Regulators and guideline groups increasingly lean on RWE, and the ED runs on registries and audit data. Knowing it is observational stops you from reading a registry “treatment effect” as if it were trial-grade proof — the single most common appraisal error with this evidence.

• RWE = observational: association, not proof of causation
• Great for harms, rare events, long-term & real-world effectiveness
• Beware confounding by indication + data quality / missingness

TARN — the Trauma Audit & Research Network is the UK’s national trauma registry (the largest in Europe, now the National Major Trauma Registry). It powers case-mix-adjusted benchmarking and a vast body of observational trauma research — injury patterns, outcomes and process measures across the whole system. Its registry studies are observational: they describe associations and adjust for case mix, but cannot randomise — so a survival difference between groups is hypothesis-generating, not proof that a treatment caused it.

• Confounding by indication — the sicker patients were the ones who got (or missed) the treatment.
• Data quality and missingness — coding errors and incomplete records distort the picture.
• Immortal time bias — mis-classifying the period before a treatment could even be given.