I am a physician, clinical pharmacologist and Director & Head of Clinical Sciences at nference. My work connects clinical questions, electronic health records and evidence generation with drug discovery and development. This guide brings together my public research and interactive explanations of the methods behind credible comparisons.
What is the difference between real-world data and real-world evidence?
Real-world data (RWD) are the observations; real-world evidence (RWE) comes from analyzing them to answer a clinical question. Routine electronic health records, claims and disease registries can contribute data about health and care. Their usefulness depends on what was recorded, why it was recorded and whether it supports the question being asked. FDA’s RWE overview provides the regulatory definitions and examples.
For example, a fictional database may contain two treatment groups with different hospitalization rates. Those records are data. A useful comparison requires a defined population, an outcome and time window, a defensible design and an assessment of confounding and missing information. The size of a dataset does not establish that a treatment caused an outcome.
How can electronic health records inform clinical trials and drug development?
EHR studies can describe care patterns, characterize populations and outcomes, and inform research questions across drug development. Depending on the purpose and data quality, real-world information may also support trial planning, safety evaluation or observational treatment comparisons. Each use needs its own fit-for-purpose assessment. A record created for clinical care is not automatically a validated research measurement.
In drug discovery, clinical patterns can help generate hypotheses that need further evaluation. In clinical development, external comparators and target trial emulations demand careful attention to eligibility, time zero, treatment assignment and outcome measurement. Observational adjustment does not create randomization. See the FDA guidance on EHR and claims data and Hernán and Robins’ target-trial framework.
A practical starting point is to write the question before extracting the cohort: who is eligible, which alternatives are compared, when follow-up begins, and which outcome matters? Then check the availability and quality of those variables, the design assumptions, and how results change under plausible alternatives. Make the reasoning traceable from the record to the conclusion.
Choose the question you want to understand
The Understanding Deeply series uses entirely fictional examples to make assumptions and calculations visible. Each lesson has a readable explanation as well as an interactive experiment.
| Your question | Explore the method |
|---|---|
| How can I compare treatment groups? | IPTW and propensity score matching |
| Which treatment effect am I trying to estimate? | Intercurrent events and estimands |
| How well does a test or phenotype identify patients? | Clinical AI and diagnostic performance |
| How should I design an observational treatment comparison? | Target trial emulation |
| How large is the benefit or harm? | Risk, odds, NNT and NNH |
| What could the unobserved outcomes change? | Missing data and sensitivity analysis |
| Could the data-generating process mislead me? | Bias in real-world data |
| How much uncertainty remains? | Statistical power and p-values |
| Do reviewers classify the same records consistently? | Rater agreement and validation |
| What process could generate many zero counts? | Zero-inflated count models |
Connect the methods to published research
Explore my clinical AI and real-world data publications, wider research catalogue and granted patents. The catalogue links to original sources and distinguishes verified findings from study-focus summaries. The source publication is the place to assess the full methods and limitations.
My professional background and ORCID record provide author context. For a discussion about a research question, get in touch.
Primary references
- FDA: Real-world evidence
- FDA: Assessing EHR and medical claims data for drug and biological product decisions
- Hernán and Robins: Using big data to emulate a target trial
These explanations and views are my own. They do not represent my employer or affiliated organizations. The learning examples are educational and contain no real patient data.