Real-World Data and Real-World Evidence

What is RWD and RWE

Real-world data (RWD)

= data are data relating to patient health status and/or the delivery of health care routinely collected from a variety of sources

• examples
  • electronic health records
  • medical claims
  • disease registry
  • wearables & mobile health

Real-world evidence (RWE)

= clinical evidence about the usage and potential benefits or risks of a medical product derived from analysis of RWD

• examples
  • safety (e.g. of a drug)
  • (comparative) effectiveness
  • long-term outcome
  • rare adverse events

Why "real-world" matters

real-world data is more realistic and less controlled than Randomized Clinical Trial (RCT) data

• RCT data are collected under a predefined protocol with controlled measurements and often randomized treatment assignment.
• RWD are generated in routine healthcare, so they are usually: less standardized, more heterogeneous, more incomplete, more representative of routine clinical practice

Key Methodological Concepts

Fit-for-Purpose Data

A dataset must be appropriate for the specific research question.

Coding and Phenotype Validity

Clinical concepts are often inferred from codes or combinations of records.

Missingness and Measurement Error

Missingness in RWD is often informative.

Healthcare-Utilization Bias

Patients who are sicker or have better healthcare access are often observed more frequently -> may partly reflect healthcare utilization rather than underlying disease biology.

Confounding

Most RWD studies are observational, so treatment or exposure is usually not randomly assigned.

Data Harmonization

Different datasets may represent the same concept differently.
Harmonization may require:

• unit conversion
• variable mapping
• coding standardization
• common phenotype definitions
• temporal alignment

Transportability and Generalizability

• Results from one dataset may not hold in another population or healthcare system.
• External validation is therefore important for both clinical prediction models and RWE studies.

From RWD to RWE

A useful framework is:

Research question
        ↓
Fit-for-purpose data
        ↓
Valid variables and phenotypes
        ↓
Assess missingness and measurement
        ↓
Address bias and confounding
        ↓
Appropriate analysis
        ↓
External validation / transportability
        ↓
Credible RWE

The key principle is:

RWE quality depends on the full study design and data-generation process, not simply on dataset size or statistical model performance.