Real-world data and study design
Real-world data and study design · RWD / RWE
Real-world data are health and healthcare records routinely collected through sources such as electronic health records, prescriptions, claims and registries. Analysis of suitable data can generate evidence about a medicine’s use, benefits or risks.
How to read it
Separate the source of data from the design of the comparison. Examine population comparability, exposure and outcome definitions, missing information and confounding.
A common pitfall
Real-world data are not synonymous with an observational design or proof of causality. They may support different designs, including randomized trials embedded in routine practice. Person-years describe a time denominator, not data provenance.
Records do not assign treatment
Routine records may omit symptoms that were never brought to care or whether a prescription was actually taken. More records do not automatically resolve missing or inaccurately measured information.
When treatment is not randomized, medicine users may differ in underlying conditions, baseline risks or care-seeking. Adjustment for measured factors cannot automatically remove unmeasured confounding.
Keep the GLP-1 analyses separate
The GLP-1 case includes pooled randomized trials and a separate Sentinel observational comparison. The 6.4 versus 3.8 events per 100000 person-years came from the pooled trials, not Sentinel. Ask who was compared, when follow-up began and how events were identified.
Practice with a case
Read the Chinese case: GLP-1 Why can a rare risk warrant a warning →The Chinese case separates person-time rates in pooled trials, routinely collected data and observational comparisons.
Podcast link not yet supplied.
Examples in other articles
Related readings will be added as the collection grows.
- FDA Submitting Documents Using Real-World Data and Real-World Evidence
- FDA GLP-1 safety communication 2026-11-09
Source
Read the primary source ↗Explanations by Before the Conclusion · November 10, 2026