Evidence brief

Endpoint validation does not establish overall data reliability

Endpoint validation is essential in many real-world evidence studies. But confidence in a small number of well-validated endpoints cannot establish whether the broader set of analytic variables is accurate, complete, or traceable to clinical source documentation.

This evidence brief summarizes peer-reviewed findings on real-world data reliability and explains why non-endpoint variables should be evaluated empirically, not assumed reliable from endpoint performance.

Download the evidence brief

What the brief covers

Real-world evidence studies often depend on variables that define disease state, timing, severity, treatment context, and care pathways. These variables can shape cohort construction, confounding adjustment, subgroup assignment, and interpretation of the outcome.

The brief summarizes evidence showing that, under traditional structured approaches, routine clinical variables may be inaccurate, incomplete, or untraceable even when high-confidence endpoints perform well.

  • Endpoint validation and where it stops
  • Accuracy, completeness, and traceability as distinct reliability dimensions
  • Per-variable reliability findings for symptoms, severity, procedures, comorbidities, and pulmonary function findings
  • A four-criteria framework for evaluating real-world data reliability
  • The limits of the findings and when they should not be overextended

Why this matters

A real-world study can have a well-measured endpoint and still depend on analytic variables with unknown measurement properties. For pharmacoepidemiology, that distinction matters because non-endpoint variables often define the clinical context needed to interpret the outcome.

The practical question is not whether endpoint validation matters. It does. The question is whether endpoint validation is being overextended into a dataset-level reliability claim.

Inside the brief

The evidence brief summarizes a large peer-reviewed assessment of real-world data reliability across 120,616 patients with asthma, 58 US hospitals, and more than 1,180 outpatient clinics.

Under traditional structured approaches:

59.5%
Accuracy (F1)
Mean across clinical variables
46.1%
Completeness
Fewer than half of expected data sources present per patient-year
11.5%
Traceability
Fewer than 1 in 9 data elements traceable to concordant source documentation

Selected routine clinical variables showed substantial under-capture or misclassification, including asthma severity subtypes, cardinal symptoms, and pulmonary function findings.

Who should read it

This brief is intended for pharmacoepidemiologists, HEOR researchers, real-world evidence leaders, epidemiology methods teams, and others evaluating whether real-world data are fit for a specific research use case.

It is especially relevant when a study depends on clinical variables beyond a small number of high-confidence endpoints.