Aetion makes health data answer for itself

A hospital bill is an excellent witness to a transaction. It is a rather less reliable witness to a cause. It can record that someone received a medicine, returned to a clinic, or entered a hospital. The difficult question is what would have happened to that person if the treatment had been different. Aetion has made that difficulty its business.

The bill is not the experiment

Healthcare records arrive with a complication: doctors choose treatments for reasons. Those reasons can also influence the outcome being studied. Compare two groups without accounting for their differences and a beautifully polished chart may deliver a thoroughly misleading answer. More rows do not automatically improve the question.

Aetion’s Evidence Platform, usually shortened to AEP, helps researchers apply epidemiological methods to these records and document how they did it. Its territory is real-world evidence: research drawn from the care people receive outside the controlled setting of a clinical trial. The business serves people who manufacture treatments, people who pay for them, and people who review the evidence. The 2025 acquisition announcement describes safety, effectiveness and value as the central questions.

That makes it an unusual software proposition. A result must be useful to its author, but also legible to a skeptical stranger. A reviewer needs to understand who entered the study, what counted as an outcome, and which comparisons were made. The trail behind the number is part of the deliverable.

Two scientists, and a missing piece of software

Jeremy Rassen and Sebastian Schneeweiss came to this problem through academic medicine. In a 2017 founder interview, Rassen described a mismatch: methods for assessing treatment effectiveness and value were well developed, while the technology to apply them at the required speed and scale was missing. He dates their founding to 2013.

Rassen brought an interesting second education. Before academic epidemiology, he had worked in Silicon Valley on large-database software. His investor biography records experience at Hewlett-Packard and Epiphany. Here was someone acquainted with both the statistical argument and the awkward business of making software usable.

The origin matters because it explains the product’s emphasis. Aetion sells tools for creating and examining evidence, with scientific services alongside them. Its advantage has to survive scrutiny of the research process. An attractive interface helps; an unexplained analytical choice can undo the work behind it.

“Science is our foundation.”Jeremy Rassen, March 2025 interview

The test that refused to flatter the software

One of the most revealing chapters is RCT-DUPLICATE, a research initiative comparing database analyses with randomized clinical trials. AEP supported implementation. A 2023 JAMA paper reported 32 trial emulations using three US claims databases, with protocols registered before analysis.

Across these selected trials, 75% met the study’s statistical-significance agreement criterion. In a post hoc subset of 16 trials with closer emulation of design and measurements, that figure was 94%; it was 56% in the other 16. These percentages describe agreement by one specified metric, not general software accuracy.

The failure point was partly the fit between the question and the records. Some trial design elements could not be closely represented in claims data. The authors also identify chance and residual confounding as possible causes of divergence. Their conclusion supports complementing trial evidence under suitable conditions. It does not issue a season ticket to skip trials.

Four applications, one chain of reasoning

Aetion’s applications divide the work into recognizable jobs. Discover, launched in May 2023, lets users explore populations, outcomes and treatment patterns, assess datasets and develop hypotheses. It gives early questions a place to go before a formal analysis. Substantiate supports more advanced descriptive and causal research.

Discover also records a useful change in the company’s thinking. Its launch announcement says customers wanted the technology extended to everyday exploratory insights, with a route into advanced evidence generation. Aetion responded by broadening the work its platform could support. The research journey needed an easier beginning.

Activate takes on the preparation stage. Launched in August 2025, it combines low-code tools with a hosted coding environment. Users can define measures, transform datasets and move into advanced analysis while retaining versioning and audit trails. Reusable definitions are a particularly practical feature: teams can spend less effort reinventing the same measure for each study.

Generate addresses another obstacle: access and privacy. Aetion acquired Replica Analytics in January 2022, adding technology for synthetic health data. The aim is to create useful representations that protect privacy while preserving relevant statistical properties. Synthetic data still needs evaluation for the intended task. A copy can be convenient without being adequate for every question.

Why the data connector bought the evidence company

Datavant announced its agreement to acquire Aetion in May 2025; Aetion confirmed completion that July. The combination joined data discovery, linkage and privacy capabilities with Aetion’s analytical tools. AEP also became available in AWS Marketplace. The official deal announcement did not state a purchase price.

The commercial logic is straightforward. A patient’s history can be fragmented across sources. Connecting those records improves what researchers can see. Researchers still have to decide whether the assembled information is adequate for a particular study. Connectivity and interpretation solve adjacent problems.

Datavant’s July 2026 discussion of evidence strategy makes that distinction explicit. Relevant endpoints may be absent from structured datasets or buried in clinical notes. Targeted record retrieval and enrichment can supply missing detail. Bringing the companies together creates opportunities to address these gaps in one workflow; it does not make every connected dataset fit for every purpose.

What a buyer is really paying for

Aetion’s business model is enterprise software and scientific services. Rassen described the platform as a service in 2017, with cloud, on-premises and hybrid deployment options. The current portfolio supports work across clinical development, regulatory evidence and market access. Customers buy a means of conducting research and making its reasoning inspectable.

A named example is Sanofi. Its 2019 enterprise collaboration paired its DARWIN real-world data platform with AEP to support transparent regulatory-grade studies. Aetion’s January 2022 announcement also describes collaborations with NICE and selection by the European Medicines Agency for safety and efficacy research. Those relationships establish concrete use cases; they do not confer blanket approval on every future analysis.

There are alternatives. IQVIA Real World Solutions combines data assets, analytics and scientific services. An organization can also build workflows internally with statistical programmers and epidemiologists. Aetion’s particular offer is a purpose-built evidence platform, now connected to Datavant’s infrastructure. The choice depends on the data required, the study, and the capabilities already inside the buyer’s organization.

The 2021 financing announcement gives a dated view of commercial momentum. Warburg Pincus led the $110 million round, with B Capital, Foresite Capital and existing investors participating. Aetion reported nearly doubled revenue and 100% customer renewals over the preceding year. Those are company-reported historical measures, rather than a description of today’s accounts.

Borrow the discipline before you buy the platform

The useful habit to copy is to design the study before falling in love with the dataset. Specify the population, comparison, outcome and timing. Check that the records can measure them. Keep definitions consistent across teams and preserve the steps between raw data and results. Treat missing information as a study-design problem.

That discipline now faces another test. In April 2026, Datavant described developing and deploying agents across Connect and AEP, with scientists retaining decision authority at defined review points. Its stated approach uses scientific evaluations and phased deployment. These are plans and company-described practices, rather than an independently established measure of agent performance.

Aetion’s story is useful because it makes an ordinary-looking feature consequential. Someone must be able to reconstruct the analysis. In healthcare, the answer travels: from a scientist’s screen to a meeting, a submission, a reimbursement discussion. The records begin the conversation. The work is making sure the conclusion can travel with them.