Joseph L. Smith / Real-World Evidence

Evidence Is Only as Strong as Its Method

Evidence, not vibes. Strategy, not decks. Every study should start with a clear question and a protocol built to survive scrutiny.

Real-world evidence is not a shortcut around rigor. It is a different evidentiary terrain, with its own threats, advantages, and standards of credibility.

The evidence hierarchy still matters.

Real-world evidence becomes persuasive when the design is explicit about its place in the evidentiary stack. Not every question requires a randomized trial. Not every dataset can support a causal claim. The discipline is matching the question to the design and naming the limits honestly.

The best RWE programs do not pretend observational data is magic. They build credibility through target trial thinking, transparent cohorts, validated phenotypes, sensitivity analyses, and a clear explanation of what would change the decision.

Figure 1 / Evidence Hierarchy
  • 01
    Systematic Reviews & Meta-Analyses

    The strongest synthesis when the included studies are credible, comparable, and transparent.

  • 02
    Randomized Controlled Trials

    The benchmark for causal evidence, but often less representative than the real clinical world.

  • 03
    Prospective Cohorts

    Planned real-world follow-up with predefined protocols and better operational realism.

  • 04
    Retrospective Cohorts & Case-Control

    Fast, scalable, and powerful when confounding, selection, and measurement are handled directly.

  • 05
    Descriptive Studies

    The foundation: burden, prevalence, treatment patterns, and the commercial case for deeper work.

The ladder is not a status game. It is a way to prevent overclaiming. Descriptive studies can be exactly right for burden and market understanding. Comparative effectiveness needs a different burden of proof.

Data Sources

Every source has an argument and a flaw.

Data selection is not procurement. It is study design. Claims, EHR, registries, patient reports, and linked datasets each answer different questions and fail in different ways.

Source Note
Claims

Imperfect, but if you understand what it captures and misses, incredibly powerful.

Source Note
EHR

Rich clinical detail, messy data engineering, and the frontier where NLP becomes useful.

Source Note
Registries

Purpose-built for research, especially strong for rare disease and longitudinal follow-up.

Source Note
Linked Data

Claims plus EHR plus genomics plus patient voice: difficult to combine, but useful when the sources answer the same question.

Methods That Move the Needle

Credibility is designed.

The bar for RWE is not statistical significance. It is causal clarity, operational transparency, and interpretability for the audience that has to use the finding.

Target Trial Emulation

Specify eligibility, treatment strategies, time zero, follow-up, outcomes, and estimands as if a trial were being designed.

Synthetic Control Arms

Use historical patient data carefully when conventional controls are impractical, especially in rare disease and oncology.

Federated Evidence

Generate multi-site evidence while reducing unnecessary patient-level data movement.

The Discipline

Name the limits before someone else does.

The credibility of a real-world study is decided less by the headline result than by how honestly it handles confounding, missingness, and selection. The strongest programs publish their assumptions and pre-commit to what would change the conclusion.

Figure 2 / The Evidence Terrain
Internal validity →Decision relevance →DescriptiveCase-ControlRegistriesRetrospective CohortProspective CohortPragmatic TrialRCTTarget Trial EmulationTrial discipline, real-world terrain
Reviewer Test
Practical standard

If the protocol, cohort logic, and sensitivity analyses cannot be explained cleanly to a skeptical reviewer, the evidence is not ready to carry the decision.

Let's talk evidence.

Need a partner for RWE strategy, study design, or evidence generation? The best starting point is the decision your evidence has to support.

Start the Conversation