Real-world evidence and comparative effectiveness
- Concept
Uses data from everyday care (registries, health records, billing) to see how treatments perform outside trials.
What it is
Real-world data are generated in routine care: electronic health records, disease registries, claims and billing databases, and digital health devices. Real-world evidence is what is concluded from analysing them. They allow millions of patients, rare effects and groups excluded from trials to be studied, and treatments used in practice to be compared (comparative effectiveness); their great risk is confounding by indication.
Worked example
- Registry of 10 000 heart failure patients: 2 000 get a new medicine and 8 000 do not. Deaths at 1 year: 300 and 800.
- Crude: 300 / 2 000 = 15 % versus 800 / 8 000 = 10 %; RR = 1.5. It looks harmful.
- Severe: treated 270/1 500 (18 %) versus untreated 240/1 000 (24 %); RR = 0.75.
- Mild: treated 30/500 (6 %) versus untreated 560/7 000 (8 %); RR = 0.75.
- Check: 270 + 30 = 300 and 240 + 560 = 800. 75 % of treated patients (1 500 / 2 000) were severe.
- Reading: sicker patients were prescribed it more; stratifying by severity moves the RR from 1.5 to 0.75. It was confounding by indication.
Real-world data sources
| Source | What it offers | Main limitation |
|---|---|---|
| Disease registry | Standardised clinical data and follow-up | Only patients from participating centres |
| Electronic health record | Clinical detail, lab results, vital signs | Free text, missing data, nothing from other institutions |
| Billing databases | Many patients and service use | Little clinical data; codes designed for billing |
| Randomised trial (for comparison) | Groups made comparable by chance | Selected patients, limited size and duration |
Markers
- RWD
- RWE
- Confounding by indication
- Propensity score
In practice
- Safety surveillance of medicines and vaccines after approval.
- Studying children, older adults, pregnant women or people with several diseases, often excluded from trials.
- Outcomes research: quality of life, service use and costs in practice.
Common pitfalls
- Comparing treated with untreated without adjusting for severity or the reason for prescribing.
- Believing millions of records remove bias: size reduces chance, not confounding.
- Defining the start of follow-up badly and giving one group risk-free time (immortal time bias).
Test yourself
Think of the answer, then tap the question to see it.
What is confounding by indication?
A bias in which the reason a treatment is prescribed (for example, greater severity), not the treatment itself, explains the outcome.
What are three sources of real-world data?
Disease registries, electronic health records and claims or billing databases; digital devices too.
Why does a very large database not guarantee unbiased results?
Because size reduces random error but does not correct confounding or measurement error.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Real-World Evidence (U.S. Food and Drug Administration, 2026); Confounding by indication | Catalog of Bias (Catalogue of Bias Collaboration, 2018); Chapter 24: Including non-randomized studies on intervention effects (Cochrane, 2024).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.