The Story
Real-world evidence is a phrase that means, roughly, "what actually happened to real patients, as opposed to what happened to the carefully selected few in a clinical trial." Regulators increasingly want it. Drug companies desperately need it. And almost nobody can produce it cleanly, because the raw material is a mess. Century Health's pitch is that it has automated the un-messing.
The problem is boring, which is the point
If you want to study, say, whether a multiple sclerosis therapy is working across thousands of patients, you first need those patients' records in a form a computer can reason about. In practice that record is a neurologist's typed note, a scanned lab report, an imaging summary, and a billing code, spread across systems that were never designed to talk to each other. Historically, turning that into a structured dataset meant hiring humans to read charts and type the answers into a spreadsheet. This is slow, expensive, and - because humans get tired - inconsistent.
Century Health's answer is a platform called CHARM, which stands for Century Health Abstraction & Retrieval Model. CHARM reads the notes, the reports, and the codes, and extracts the clinical variables a researcher would otherwise pay a person to hunt for. The company says it does this at 97% accuracy measured against clinical expert judgment, which is the polite way of saying "about as well as the person you were going to hire, minus the fatigue."
Why 97% is a specific number
Ninety-seven percent is interesting precisely because it is not one hundred. A company selling you certainty would round up. Century Health is selling you a tradeoff: near-expert accuracy at machine speed and machine cost, benchmarked honestly against the humans it replaces. In a field where "AI" is often a synonym for "trust us," publishing a number you can be held to is itself a strategy.
The other number worth staring at is 60x. That is how much the company says its clinical data network grew over the past year. Networks like this compound: each specialty clinic that joins makes the resulting registries deeper, which makes them more valuable to pharma, which funds more clinics joining. The trick is starting - and Century Health started by going narrow.
Narrow, then wide
Rather than trying to ingest all medicine at once, Century Health built disease-specific registries in areas where the data is deep and the questions are urgent: neurology, nephrology, ophthalmology, respiratory, metabolic, and immunology. It partners with specialty provider networks - Nira Medical in neurology (4,000-plus MS patients), Balboa Nephrology, Nimbus Health, Memory Treatment Centers - to curate their records, then licenses de-identified, analysis-ready datasets to the drug companies studying those diseases.
This is a two-sided business wearing a data-infrastructure costume. On one side are clinics sitting on valuable but unusable records. On the other are life-sciences companies - reportedly including multiple "Top 5" pharma names - that will pay well for records made usable. Century Health stands in the middle, running the compliance gauntlet so neither side has to.
Compliance is the moat
The unglamorous truth of health data is that the reason it stays trapped is legal, not technical. Move it carelessly and you have a HIPAA problem. Century Health leans into this: de-identification, SOC 2 certification, role-based access control, encryption, and data traceability are not footnotes on the website - they are the product. A pharma company hands you patient data only if it believes you will not get it, or them, sued. Treating that as the core deliverable, rather than a tax, is how a 16-person startup ends up in enterprise contracts.
It also explains the founders. CEO Vish Srivastava spent time at Boston Consulting Group and then Evidation Health, where he led product for a consumer health platform that reached five million users and scaled real-world evidence offerings. His co-founder and CTO, Sanjay Hariharan, came out of McKinsey's QuantumBlack machine-learning group. One knows how evidence gets sold to pharma; the other knows how to make a model behave. The company also traces its motivation to something less strategic - loved ones facing neurodegenerative disease - which is the kind of origin that keeps a team pointed at neurology when the spreadsheet says do something easier.