How Recursion Brought Real-Time Analytics to Clinical Trials with Quarto and Posit Connect

Summary

Recursion Pharmaceuticals built a real-time clinical monitoring portal with Quarto and Posit Connect, cutting dose escalation meeting turnaround from 7–10 business days to 24–48 hours with three analysts now supporting five or more active trials simultaneously. With that governed foundation in place, the team is moving toward AI-assisted clinical workflows designed to meet the FDA's emerging push for real-time clinical trials.

About:

Recursion Pharmaceuticals, Inc. (NASDAQ: RXRX) is a clinical stage TechBio company leading the space by decoding biology to radically improve lives.

Industry:

Life Science

Technology used:

Posit Workbench, Posit Connect, Posit Package Manager

“We do a lot of first-in-human dose escalation trials, where after enrolling a small cohort of patients we need to hold a meeting with investigators from different sites, and typically it would take a week or 10 days. With Posit and a lot of the work with the Quarto reports, we are able to constantly monitor the data and hold that meeting even within a day.”

Kasra Yousefi

Executive Director, Biostatistics at Recursion

The challenge

3 experts, 5 trials, and a PDF version control problem

Recursion runs a growing portfolio of early-phase drug trials, predominantly Phase 1 and Phase 2, where the central questions at every step are the same: is this drug safe, and is it working at the right dose? In dose escalation trials, a safety review committee needs to convene after every few patients to decide whether to move to a higher dose level, stay at the same dose level or move to a lower dose. The speed of that decision depends entirely on the speed of the data in front of the committee.

For the biometrics team at Recursion, getting data to investigators used to mean a multi-step, manual cycle. Raw data was extracted from the Electronic Data Capture (EDC) system, processed in R, and rendered into static documents: PDFs or HTML files distributed through email and/or Google Drive. The moment a file left the analyst's hands, the aging process began. Reviewers would ask follow-up questions. Late-entering data points would shift the picture. By the time a corrected version arrived, people would often be working from out-of-date information.

Chendi Liao, Executive Director of Statistical Science, describes the overhead that came with it: keeping track of which version was current, fielding requests for one more visualization or a corrected data point, and re-exporting documents each time. For a biometrics team of fewer than 10 people supporting five or more active trials, there was no slack to absorb it.

The solution

A clinical monitoring portal built in Quarto, deployed in a click

The shift began with a move from R Markdown to Quarto for clinical reporting, and from emailed exports to Posit Connect for delivery. In practice, it collapsed the distance between the analyst and the clinical scientist and medical monitor.

The workflow starts in Posit Workbench, where the team draws data from Recursion's clinical data storage environment through R. Using packages from the pharmaverse, they derive what are called ADaM datasets, analysis-ready data models derived from raw data, and build Quarto documents that combine structured summary tables with targeted interactivity.

The interactivity is deliberate and restrained. The portal is designed to answer the specific questions clinical scientists and medical monitors bring to it: whether a patient is showing a safety signal, how adverse event rates compare across dose levels, whether any patterns warrant escalation caution. As Kasra Yousefi, Executive Director of Biostatistics, describes it, the dashboard’s minimalism allows the team to focus on specific needs and perform targeted data review. Quarto gives enough interactivity without the cognitive load of too many buttons.

When a document is ready, the team publishes it to Posit Connect in one step, using a custom URL that ties the report to a specific study and reporting event. From that point on, there is no version management. If a data point arrives late, or a scientist asks for a new visualization, the report is refreshed and everyone sees the updated version immediately. No re-sending, no reconciling which PDF is current, no waiting.

24-48

Hour turnaround

Time from data availability to dose escalation committee meeting, down from 7-10 business days

3

Analysts supporting 5 active trials

A ratio that would conventionally require double or triple the headcount.

7x

Faster data-to-decision time

Estimated compression for dose escalation review cycles specifically.

The results

Dose escalation decisions in 24 hours, not 10 days

The impact is most visible in dose escalation reviews. In a Phase 1 trial, every new dose cohort requires a meeting of investigators from the sites enrolled in the study. They review all available safety data, adverse events, dose interruptions, and patient-level signals, before deciding whether to escalate. Scheduling those meetings and assembling the data used to take between a week and 10 days.

With Quarto deployed on Posit Connect, that process now completes in 24 to 48 hours. Internal clinical development stakeholders arrive already oriented, with an aggregated view of the cohort and the ability to zero in on individual patients where something warrants a closer look. The analyst is not the bottleneck in that cycle.

The team is also doing more with less in a way that is genuinely unusual for clinical biotech. Currently three Posit users on their biometrics team are supporting five or more active trials simultaneously with 40+ active users accessing and reviewing data in Posit Connect. As Yousefi notes, the conventional staffing model would require one or two data analysts per trial, which would mean doubling or tripling the team to cover the same workload. The automation and governed self-service that Posit Connect enables is what makes the ratio work.

The quality of information available to clinical scientists and medical monitors has improved in parallel. Rather than receiving a PDF snapshot at the end of a reporting cycle, internal clinical development stakeholders can access the portal at any time, viewing data from both the aggregated cohort level and the individual patient level. That continuous availability has, as Yousefi observes, started raising the analytical fluency of the scientists who use it. They are getting better at reading their own data.

Looking ahead

The next phase of AI in clinical workflows

Positron is beginning to enter the workflow at Recursion. For established workflows, the muscle memory of a familiar environment is hard to displace, but for new builds it is quickly becoming the default starting point. The feature the team is most interested in is the AI assistant.

The big question on the team's horizon is how to bring AI more deeply into clinical data workflows in a way that satisfies health authority requirements. At the moment the team is connected through GitHub Copilot, but conversations are underway with the IT infrastructure and data engineering team about integrating other large language models directly into the Posit Assistant. The team describes it as a “phase two” of their deployment, with a target later in 2026, noting that the regulatory dimension isn't a blocker so much as a box that needs to be checked carefully.

By laying an AI-driven GxP infrastructure early, the team is positioning itself to align seamlessly with the FDA's recent push toward Real-Time Clinical Trials. This preparation ensures the data pipeline is built for the continuous, secure streaming of endpoints and safety signals, turning real-time review requirements into a distinct operational advantage. In a first-in-human trial context, that care is essential.