A real-world evidence (RWE) platform pools the health records and insurance claims of millions of ordinary patients, and analyses them to learn what medicines do outside trials. The records themselves are called real-world data. What is learned from them is called real-world evidence. The records are de-identified: names and other identifying details are removed before anyone analyses them.
A trial tests a medicine on a few thousand carefully chosen patients. After launch, millions of different people take it: older people, people with other illnesses and people taking other medicines. Many of them would not have been allowed into the trial. A side effect that affects 1 patient in 20,000 will probably not appear in a trial of 3,000 patients. It shows up only across millions of records.
RWE platforms are used for three things. They find rare side effects. They find new uses for a medicine, when patients who took it for one illness do better at another. And more and more, they produce evidence of the quality that drug regulators accept. The largest platforms are built on records from the United States, often pooled by groups of American hospitals.
Consider a blood thinner that 400,000 patients in the pooled records have taken. A drug safety team asks whether it causes liver injury. The platform finds the patients who took it. It also finds a comparison group of similar patients who took an older blood thinner. It matches the two groups on age, sex, other illnesses and other medicines. Much of this information is written in doctors' notes, not in structured fields. So a language model reads the notes and turns them into data. Reading notes in this way is called clinical NLP. In the end, 48 of the 400,000 patients had liver injury, or 12 in every 100,000. In the comparison group, the rate was 4 in every 100,000. A difference like this is called a signal. It is a reason to investigate, not yet proof. The company reports it to the regulator and studies it more closely.
The records come from healthcare data platforms and from clinical data abstraction, in which models or trained staff turn notes into data. For cancer, the platform adds each tumour's genes from tumour genomic profiling, because the right medicine often depends on them. This is one use of genomics. The same records can also supply the comparison group for a trial, which is called a synthetic control arm.
An RWE platform is different from a clinical trial data platform. A trial platform holds data collected for one trial, under that trial's rules. An RWE platform holds data that was recorded during ordinary care, for other reasons. So its results need careful analysis before anyone trusts them.