Beyond the clinical trial: longitudinal real-world evidence across the drug lifecycle

What is real-world evidence?

Real-world data (RWD) are observations collected in routine care or everyday life: electronic health records, claims, registries, patient-reported outcomes and connected-device readings.

Real-world evidence (RWE) is what a well-designed analysis of suitable real-world data can support for a specified question. The paper argues that evidence quality comes from design and measurement, not from more data.

Why the patient's experience between visits matters

Randomized controlled trials remain the foundation for judging whether a medicine works and is safe, but they describe patients at a handful of scheduled moments. Routine-care records describe encounters, not the weeks between them. In conditions where symptoms fluctuate, such as migraine, myasthenia gravis or the early months of a metabolic therapy, much of what matters to patients happens out of view.

Linking electronic health records, validated patient-reported outcomes, medication records, connected-device observations and context into one longitudinal record can close some of these gaps. It can also introduce device variability, participation bias, reporting burden and missing data that is not random.

• Phase 2–3 development
• Natural history
• Post-authorization research
• Market access and HEOR

A four-level framework for real-world evidence use

Requirements for design, measurement and documentation rise with the strength of the claim.

1. Level 1 · Descriptive

   What happened, how often and in whom.

2. Level 2 · Exploratory

   Which factors move together; results are hypothesis-generating.

3. Level 3 · Fit-for-purpose measurement

   Whether a measure is valid for the concept it is meant to represent.

4. Level 4 · Comparative causal inference

   Whether strategy A led to a different outcome than strategy B, using target-trial designs.

Use cases: migraine, rare disease and metabolic therapy

Evidence economics: the cost per evaluable participant

The paper avoids hypothetical savings. Its arithmetic evidence-yield model shows that at 80% annual retention and 75% outcome completion, fewer than half of enrolled participants contribute complete 24-month data. Raising both rates to 90% would cut the cost per evaluable participant by about a third at the same budget. Sponsors should measure these rates prospectively, not assume them.

Seven practices for credible longitudinal evidence

1. 1. Name the evidence use first

   Choose the level, from descriptive to causal, before choosing sources or devices.

2. 2. Use established measures

   Prefer validated PROs and ClinROs; treat device data as context until shown otherwise.

3. 3. Pre-specify definitions

   Exposure, persistence, outcomes, windows and time zero, before the data arrives.

4. 4. Track yield monthly

   Retention, completion and burden, by channel and subgroup.

5. 5. Plan safety handling

   Agree the adverse-event review route with pharmacovigilance before enrollment.

6. 6. Keep lineage

   Provenance, versions and transformations for every value.

7. 7. Engage early

   Where data may support a regulatory question, discuss the plan with the agency early.