Take a deep breath.
There is an AI that makes a universal health record in America - Happy July 4th!
Happy July 4th!
It’s a hot summer Friday in Israel and in the US - 249 years of independence; the future looks bright for America and Israel.
Take a deep breath.
There is an AI that collects medical records from everywhere in the US and transforms them into meaningful data for patients, providers, and researchers.
My recent guest on the Life Sciences Today podcast was Troy Astorino, Co-Founder and CTO at PicnicHealth.
PicnicAI can collect medical records from everywhere in the US and transform them into meaningful data for patients, providers, and researchers.
They call it the “Universal Health record” .
PicnicAI is a 8 BN parameter LLM that outperforms frontier models on clinical data labeling and interpretation
Solving gnarly problems
11 years ago, Noga Leviner and Troy Astorino decided to solve a very gnarly problem: How to put US patients in control of their own medical data.
Today, Picnic Health specializes in non-interventional research and uses PicnicAI to address key problems in traditional trial models: cost reduction, efficiency improvement and eliminate geographical constraints.
In non-interventional research 70-80% of study data comes from standard of care information.
An example use case is Cell/gene therapy follow-up.
The traditional model requires 15-year site visits; unsustainable for patients to travel often across continents for their follow-up visit.
PicnicAI solution enables remote data collection through community health systems.
See the podcast here The Universal Patient Record - Life Sciences Today hosts Troy Astorino from Picnic Health.
Intro
This week, I zoomed-out after my interviews with Troy Astorino(Picnic Health), Alex Deyle (Flatiron Health) and Pamela Tenaerts (Medable).
I took some time off yesterday to analyze how the 3 tackle the problem of “how to get high-quality data efficiently”.
I found 3 powerful paradigms to attack the problem of getting good data.
The problem of getting good data
The problem of “how to get high-quality data efficiently” is universal.
In life sciences, the Medable, Flatiron, and PicnicHealth models represent 3 powerful paradigms to tackle this:
Improve the process (Medable - make data generation easier via SaaS),
Build the pipeline (Flatiron - exchange value for data access and own the aggregated asset), or
Mine the exhaust (Picnic Health - gather data that’s already out there and refine it with AI). Other industries mirror these approaches in their own contexts, each with successes and pitfalls.
Understanding the pros and cons of each model helps in strategizing data-driven initiatives—whether one is running clinical trials, optimizing farms, or tracking oil barrels, the core principles of data volume, variety, quality, speed, and ownership remain the guideposts for choosing the right approach.
Medable, Flatiron Health and Picnic Health.
Pharmaceutical companies increasingly seek high-volume, high-velocity patient data at lower cost for clinical research and real-world evidence.
Three companies — Medable, Flatiron Health, and PicnicHealth — exemplify different, almost orthogonal, go-to-market models to achieve this goal.
All aim to streamline data collection from patients and healthcare “endpoints” (clinics, health records, etc.) but each uses a distinct strategy:
Medable – Provides a decentralized clinical trial platform (software) sold to sponsors. Patients participate via local clinics or at home using Medable’s tech (eConsent, ePRO, sensors, telemedicine). The platform aggregates trial data in real time for the pharma client.
Flatiron Health – Offers a free oncology EHR system (OncoEMR) to community clinics, which in turn gives Flatiron access to those clinics’ patient data. Flatiron curates, normalizes, and aggregates these EHR data into rich datasets for pharma-sponsored research.
PicnicHealth – Operates a patient-centric data service: patients consent to share their medical records, and PicnicHealth passively collects their data from all the providers they visit. An in-house AI (a bespoke LLM built on Llama, ~8B parameters) then normalizes and structures the records. This yields longitudinal patient datasets across multiple health systems.
Each model has unique pros and cons in terms of data volume, quality, speed, customer acquisition, and defensibility.
Let’s now take a deeper look at each approach.
3 ways to get good data
Despite a common end goal (delivering high-quality patient data to pharma faster and cheaper), Medable, Flatiron, and PicnicHealth illustrate three distinct strategies.
In essence, they differ in what they offer to the data “source” and how they monetize the data downstream:
Medable’s “Sell Software” approach
Medable sells a technology platform to sponsors and CROs running clinical trials.
The value proposition to pharma is improved trial efficiency, patient reach, and data quality in a live study.
Medable doesn’t own clinical data. They own software that they sell in a SaaS model.
The advantage of this model is that it generates revenue immediately from software sales and embeds Medable into the trial execution process (high switching costs once a trial is underway)
It also ensures data is collected in a controlled manner (designed by the protocol, captured with purpose-built tools).
This yields cleaner data tailored for regulatory submission.
Additionally, Medable’s platform can enhance patient diversity and inclusion by reaching people remotely, a noted benefit of decentralized trials.
However, a drawback is that Medable operates in a highly competitive space (many platforms offer overlapping DCT capabilities).
Its model is service-oriented – success depends on continued adoption each time a new trial is initiated.
It does not accumulate a proprietary patient data asset; the deliverable is the trial outcome itself.
In the long run, Medable’s moat will depend on product excellence and enterprise relationships, rather than exclusive data.
Flatiron’s “Give Software to Get Data” approach
Flatiron effectively paid (subsidized) upfront by giving away a valuable oncology EHR system, in order to harvest data exhaust from clinical practice.
The genius of this model is turning an infrastructure play into a data goldmine.
Clinics got a modern EHR tailored to oncology for free, and in return Flatiron got permission to aggregate their de-identified patient records.
Over time, Flatiron accrued an extensive, real-world dataset on cancer patient care and outcomes, which it then monetized by selling access to pharma.
The pros: This model built a high barrier to entry – once Flatiron had years of longitudinal data from hundreds of clinics, a newcomer can’t easily replicate that (especially since those clinics are now locked in to Flatiron’s EHR).
The data asset appreciates with time: more patients, longer follow-up, more treatment events. Flatiron also heavily curated the data (using tech and human experts) to ensure its quality for research, something not all competitors can easily do at scale.
The value to pharma is clear: ready-made real-world evidence to complement clinical trials, useful for observational studies, external control arms, and understanding treatment patterns.
The cons: It required a long lead time and capital before paying off. Early on, Flatiron had to support a free product and invest in data processing with little immediate revenue. Only after aggregating enough data could they sign big pharma deals.
There’s also less control over data generation – Flatiron relies on what doctors naturally record. In oncology this works well (because key outcomes like tumor response, progression, survival are routinely recorded), but in other fields it might fail (as seen with Practice Fusion’s primary care data not yielding pharma value).
Flatiron mitigated this by focusing on a specialty where data density is high and by standardizing data post hoc.
In summary, Flatiron’s model is a longer-term “own the data supply chain” play, trading short-term revenue for a strategic dataset that became its moat (evidenced by Roche’s $1.9B acquisition purely for that data and capability).
PicnicHealth’s “Passive Aggregator with AI” approach
PicnicHealth skips providing software to sites and patients (like Medable) or running trials (like a CRO).
Instead it goes straight to patients and the existing healthcare systems for data.
The company acts as a patient’s authorized data agent, pulling records from any clinic or hospital the patient has visited.
This is enabled by legal rights (patients have the right to access their data) and technology (integrations with provider systems, patient portals, etc.).
Picnic’s innovation is automating the cleaning and structuring of these records using a domain-trained large language model - Picnic AI.
Picnic AI produces a comprehensive, chronologically merged medical history for each patient, which can then be aggregated (for those who opt in to research) into study datasets.
The strengths of this model are its breadth and efficiency. It can capture data across the fragmented U.S. healthcare landscape without requiring providers to adopt any new software. One patient’s 7-year journey might span a dozen different health systems – Picnic’s system stitches that together, something neither a single trial nor a single EHR would easily obtain.
This is incredibly useful for long-term safety monitoring or outcomes research (e.g. tracking gene therapy patients for a decade, even as they move or change doctors).
It also significantly reduces cost and burden: instead of paying investigator sites for visits or manual chart reviews, the data comes via digital collection and AI curation. By leveraging standard of care (which accounts for ~70-80% of needed data in many studies), Picnic’s approach eliminates many “source data capture” expenses of traditional trials.
Moreover, their LLM (trained on millions of real records) has shown it can label and interpret clinical text on par with or better than human experts, driving down the marginal cost of data processing.
The weaknesses/challenges: PicnicHealth’s model depends on patient participation – they must continually convince patients to sign up and stay engaged. This can be a bottleneck; unlike Flatiron (where data flows automatically from all patients at a clinic), Picnic only gets data from those individuals who consent. They have addressed this by offering personal health benefits to patients and by working through pharma-sponsored programs for specific diseases.
Another challenge is ensuring the quality and completeness of data solely obtained from routine care. If a patient skips an appointment or a test, that data just won’t exist (whereas a clinical trial would have mandated it). Picnic can’t directly fill such gaps except by volume (having large samples to find enough data points) or suggesting patients get certain follow-up (if they also run a care management service).
Finally, competition and moat: while PicnicHealth has a head start in AI-powered record curation, the concept of patient-mediated data sharing is becoming more common (e.g., Apple Health app, patient portals, etc., allow data portability).
Picnic’s bet is that its expertise in cleaning and structuring multi-source data at scale will set it apart.
Its growing repository of labeled data and refined model is a self-reinforcing advantage.
Still, it must keep that lead by continually improving the AI and expanding partnerships (for example, with patient advocacy groups or pharma) so that patients choose Picnic as the platform to share their data for research.
In summary
Medable is an application software company that optimizes how new data is collected in clinical trials using a decentralized model.
Flatiron is a data platform that transforms existing clinical data into research-grade assets.
PicnicHealth is a data aggregator that pulls disparate existing data into a unified format using AI.
All three address inefficiencies in the traditional clinical trial “data supply chain,” but at different points: Medable at the point of data generation (making new trials easier), Flatiron at the point of data aggregation from practice, and Picnic at the point of data integration across systems.
Each finds a way to deliver value to pharma, who increasingly appreciates all these approaches – indeed, they are complementary (pharma might use Medable for a prospective trial, Flatiron for a retrospective cohort, and PicnicHealth for long-term follow-up or real-world data capture, depending on needs).
About Me
I’m a former pharma-tech founder who bootstrapped to exit.
Now I run a private community with 900+ life science leaders.
📈Every Friday: sharp, no-fluff insights on how real techbio companies grow revenue—plus candid interviews with the builders.
Join the private network for techbio entrepreneurs.
About Picnic Health
PicnicAI can collect medical records from everywhere in the US and transform them into meaningful data for patients, providers, and researchers.
🔗 Visit Picnic Health