OM1
OM1 is a Boston real world data and AI company that Richard Gliklich, a physician, founded in 2015. It sells to drug and diagnostics makers, health systems and payers, and the companies it names include Amgen, AbbVie, GSK, Johnson & Johnson, Hologic and Exact Sciences.
Three things make up the offer. The Real-World Data Cloud links medical records, claims, lab results and specialty networks for more than 350 million US patients, deidentified and followed over time. PhenOM, OM1's own foundation model, is trained on more than a billion patient years of those histories and predicts diagnoses, clinical events and treatment outcomes for single patients or whole cohorts, through a web app, an API, an EHR embed or file delivery. Agentic data processing extracts concepts from clinical notes and standardizes records, with clinical and scientific staff reviewing the workflows.
Research teams use it to define cohorts, find hard to reach patients, run registries and observational studies, and build regulatory evidence. OM1 helped power the 650,000 patient real world study behind FDA approval of Hologic's Aptima HPV assay for primary cervical cancer screening in February 2026. In an October 2026 preprint, about 80% of 1,770 PhenOM prediction tasks reached an AUROC of 0.70 or better. Scientific consulting rounds out the work, and everything is priced by agreement.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
PhenOM sits at the center of the platform. It generates patient phenotypes and probabilities for future diagnoses, clinical events and treatment outcomes, for individual patients and for cohorts, and OM1 holds 10 AI patents around it. AI agents also do the data work underneath, acquiring, linking and enriching records and pulling concepts out of clinical notes.
The datasets, registry operations and scientific consulting have value without the model, and study design and clinical interpretation still rest with people. OM1 sells an AI engine together with data and research services rather than a model alone.
Clinical and data science staff review the data processing workflows, and extraction is traceable through quality dashboards. In the predictive medicine view, a clinician sees a patient's absolute risk, percentile and the factors behind a flag before acting on it. The API terms require qualified professionals to review and independently verify generated results before relying on them.
The automation stops at data processing and prediction; treatment decisions stay with the clinician. No confidence threshold stops a prediction automatically, no escalation rule is set, and there is no process for recalling a result after an error is found.
An October 2026 preprint describes PhenOM as a GPT style transformer of about 1 billion parameters, with a vocabulary of about 280,000 medical events. It was trained on 200 million deidentified patient histories, with a separate holdout pool of similar size, and reads longitudinal clinical sequences so that one model covers many outcomes instead of one model per outcome.
OM1 owns the model, and its weights and code are proprietary. Detailed development methods and versioned benchmarks are promised for later, and no production release history is available.
OM1 builds PhenOM itself and trains it on its own Real-World Data Cloud, so the core model and its main data source have a named owner rather than an anonymous supplier.
The surrounding steps involve more parties: data acquisition from health systems and specialty networks, record linkage, note processing and agentic workflows. The infrastructure vendors, linkage providers and any additional models used in those steps are not named. The API terms acknowledge third party materials without a subprocessor register, so the retention terms of everyone handling records around the model are not visible.
The October 2026 PhenOM preprint tested the current model on 1,770 prediction tasks in a held out population. About 80% of tasks reached an AUROC of 0.70 or better and about 30% reached 0.80. An ISPOR Europe 2026 poster on seven major cardiovascular events reported AUROCs from 0.73 for hemorrhagic stroke to 0.87 for heart failure.
All authors work for OM1, labels come from claims, and validation stays inside OM1's own network. A 2025 randomized trial of the earlier Joint Insights decision aid improved decision quality in knee osteoarthritis, but it used regression and gradient boosting rather than the current transformer.
PhenOM trains on deidentified longitudinal records. The API terms let OM1 use customer inputs to run the service, improve PhenOM and generate outputs, so submitted data can feed model improvement. Protected health information can be submitted only under a separate business associate agreement, with encryption described and identifiers kept out of metadata and logs.
How long patient inputs are retained and when they are deleted are not stated. The professional verification duty addresses reliance on uncertain outputs, but no technical safety thresholds, failure monitoring detail or incident process for a harmful prediction is set out.
OM1's June 2025 privacy notice acknowledges business associate relationships and the agreements behind them, then excludes the protected health information they govern from its own scope. The API terms allow PHI only under a separate agreement and set out encryption and user duties, but that agreement is not public.
Retention, permitted uses and subcontractor obligations for patient data therefore sit in each service contract. The deidentified research datasets and identifiable API submissions travel on different contractual paths, and only the first is described in any detail.
The API documentation describes authentication scoped to each organization and encryption in transit and at rest, and the privacy statement refers to physical, administrative and technical safeguards. No SOC 2 report, HITRUST certification, ISO 27001 certificate or penetration testing schedule is named for OM1.
Its EcoVadis Gold rating covers sustainability and corporate responsibility, not information security. For a company holding linked records on more than 350 million patients, the security controls are stated rather than independently audited.
The PhenOM API terms, dated January 2026, say the API and model are not FDA cleared for clinical use and limit them to research and informational purposes. The predictive medicine pages, though, describe point of care risk panels, patient screening and finding candidates for treatment.
How the clinical applications are separated from the API restriction, in contract or in the software, is not explained. OM1's registry work generates evidence for regulatory submissions, as in the Hologic HPV approval, which is a different matter from clearance of the prediction software itself.
Prediction tasks are defined by clinical experts, data scientists and product managers working together. Each task gets an identifier, eligibility criteria, an outcome definition and a held out evaluation, which makes the evaluation repeatable, and the data workflows have human review. The preprint reports how discrimination differs across clinical domains and age cohorts.
It stops short of a demographic fairness assessment of the commercial service, and labels from claims and testing inside OM1's own network limit how representative the results are. Monitoring thresholds, who owns bias findings and external validation are not described.
The API terms put the risk of a wrong prediction on the customer. They disclaim any warranty of accuracy or reliability, require professional verification, cap OM1's liability at the fees paid in the prior 12 months or $100 where no fees were paid, and have users indemnify OM1 for specified claims arising from their use. Separate agreements may govern a given customer.
The published discrimination results let a buyer judge the model's performance, and explanations behind each flag support review, but no error guarantee, correction service or compensation for a wrong prediction is offered.
Medical records are one of OM1's main inputs, linked with claims, lab results and other sources. On the way out, results reach customers through an API, a web application, an EHR embed, partner modules and file delivery.
No production EHR connector is named with its supported resources, authentication or read and write behavior, and standard code sets in the research data describe normalization rather than a two way clinical interface. The Joint Insights trial relied on manual data collection and chart cross checks, which says nothing about automated exchange for the wider platform.
Customers use OM1 through its hosted application or API, embedded partner experiences or file delivery, and software partners build PhenOM into their own products, which adds their hosting to the picture.
The data covers US patients and the API's default contractual territory is the United States, but neither is a commitment about where inference, storage or backups physically run. Cloud regions, tenant isolation and any customer choice of residency are not described, and nothing says whether a large sponsor can get a dedicated environment.
Buyers choose among licensed clinical datasets, registry subscriptions or custom study programs, predictive applications, API access and scientific services, and start with a demo or a direct conversation. API use runs under a separate agreement that sets authorized users, purposes, territory and term, with commercial use allowed under it.
No subscription rate, per prediction charge, minimum commitment or implementation fee is listed, and signing up for the documentation portal does not open a free production tier. Each component can carry its own scope, with no combined price schedule.
Life sciences teams, researchers, providers and payers use OM1 for research and care management, and the named companies include Amgen, AbbVie, GSK, Johnson & Johnson, Hologic, Exact Sciences, PPD and Cencora. The prediction catalog spans cardiovascular, metabolic, renal, neurological, behavioral health, oncology and surgical outcomes. Registries cover product safety, effectiveness, natural history, rare diseases and long term follow up, and the neurology specialty network grew by 500,000 patients.
Validation is uneven across that breadth, from one academic clinic in the knee osteoarthritis trial to tasks tested only inside OM1's own US data.
Alternatives to OM1
Other products that do the same kind of work as OM1, and what each one does.
Komodo Health sells real world data and AI analytics to drug makers, payers, providers and the consultants and investors who serve them.
Real world data company built on records patients choose to contribute.
QuantHealth is a clinical trial simulation company with offices in Tel Aviv and Worcester, Massachusetts.
Frekil builds AI agents that run observational real world evidence studies for pharmaceutical, biotech and contract research teams on data they already hold or license, from protocol through final report.
ConcertAI's Accelerated Clinical Trials platform helps pharmaceutical sponsors and contract research organizations design, plan and run oncology studies.
Truveta is owned by 30 US health systems that contribute de-identified health records every day to a shared research dataset.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Custom agreement
|
Dataset access, registry subscriptions, custom study programs, predictive applications, and API access are arranged directly with OM1. API rights depend on a separate agreement defining authorized users, purposes, territory, and term. | A separate business associate agreement governs API submissions that contain PHI; its terms sit in the service agreement. | Implementation and scientific service fees are not listed. | Vendor Published |
Clinical datasets can be licensed, and registries subscribed to or built for a specific program. Scientific consulting supports study design and evidence generation. Per prediction charges, subscription minimums and free production access are not listed. Product scope and contract terms are negotiated directly.