Certification
The first AI-powered health IT product certified for predictive decision support risk
Drummond pDSI-Risk certification means an independent third party reviewed how Ozwell manages and discloses the risks of its AI — not just that we say we do. Announced July 15, 2025.
At a glance
What was certified
Certification is a specific, checkable claim. These are its terms, so you can verify them rather than take them on trust.
- Certification
Drummond pDSI-Risk
Issued by Drummond Group, an ONC-Authorized Certification Body.
- Criterion
ASTP/ONC § 170.315(b)(11)
The decision support interventions criterion, covering predictive DSI.
- Announced
July 15, 2025
Ozwell was the first AI-powered health IT product to achieve it.
- Developer
BlueHive Health, LLC
Part of Medical Informatics Engineering, Inc., the team behind WebChart and Enterprise Health.
The definition
What a predictive decision support intervention is
ONC defines a pDSI — generically understood as AI — with a three-prong test. Ozwell meets all three, which is why it is certified as one rather than exempt from the question.
Supports decision making based on models
Ozwell uses third-party foundation models to inform and augment the clinical documentation workflow.
Derives relationships from training data
The models behind Ozwell learn relationships from training data rather than following hand-written rules.
Produces a prediction, classification, recommendation, evaluation, or analysis
Ozwell outputs recommendations, evaluations, and analyses — drafted SOAP notes, clinical summaries, and suggested documentation actions.
FAVES
Fair, Appropriate, Valid, Effective, Safe
ONC groups the expectations for decision-support AI under five principles. Here is what each one requires, and what we have published against it.
Fair
Outputs should be unbiased and equitable regardless of patient demographics or clinical context.
What we publish: Our disclosure describes the approach taken to ensure fairness, the methods used to manage and reduce bias, and the results of fairness testing on both internal and external data — including the limitation that testing relied on simulated scenarios rather than real-world demographic data.
Appropriate
The tool should support healthcare professionals rather than replace clinical judgment.
What we publish: Ozwell’s intended decision-making role is published as “inform and augment decision-making, but not replace clinical management.” Nothing reaches a patient record without a clinician approving it.
Valid
Outputs should be accurate, relevant, and consistent with clinical guidelines and the user’s prompt.
What we publish: We publish validity testing on data from the same source as the training data and on external data, the metrics used, and the standing advice that users verify outputs against current clinical guidelines and patient-specific information.
Effective
Results should be actionable and useful in real clinical workflow.
What we publish: The value the intervention produces is described concretely — drafted SOAP notes, medical record entries, and clinical summaries — and was validated with clinician feedback from outpatient clinics, hospitals, and administrative offices.
Safe
The tool should be built and maintained to minimize risk to patients, users, and organizations.
What we publish: We publish known risks and limitations, and the settings and populations where Ozwell should not be used — emergency and critical care, and highly specialized care such as advanced pain management or rare disease treatment.
Source attributes
Everything we are required to disclose, published in full
Certification under § 170.315(b)(11) requires a developer to answer thirty-one questions about how the intervention was built, tested, and bounded. Ours are published in full — not summarized, and not behind a sales conversation.
- Details and output of the intervention
- Who developed it, who funded it, what value it produces, and whether the output is a prediction, classification, recommendation, evaluation, or analysis.
- Purpose of the intervention
- Intended use, intended patient populations, intended users, and the decision-making role it was designed for.
- Cautioned out-of-scope use
- The tasks, situations, and populations a user is cautioned against applying it to, plus known risks and limitations.
- Development and training data
- Exclusion and inclusion criteria for the training set, and the relevance of that data to the deployed setting.
- Fairness in development
- The approach taken to ensure output is fair, and the approaches used to manage, reduce, or eliminate bias.
- External validation
- The data source, clinical setting, and environment where validity and fairness were assessed, and who conducted the testing.
- Quantitative measures of performance
- Validity and fairness in test data from the same source as the training data, and in data from a different source.
- Ongoing maintenance and update
- How the intervention is monitored and updated once deployed, and how emerging bias is identified and addressed.
Questions
About the certification
What certification does and does not mean, in the plainest terms we can put it.
What is Drummond pDSI-Risk certification?
It is an independent certification, issued by Drummond Group, that a predictive decision support intervention meets the intervention risk management and source attribute disclosure requirements of ASTP/ONC § 170.315(b)(11). Ozwell was the first AI-powered health IT product to achieve it, announced on July 15, 2025.
What does the certification actually test?
It confirms alignment with intervention risk assessment and mitigation benchmarks, transparency standards, and the intervention risk management and source attribute disclosure requirements for predictive decision support. It reviews how the risks of the AI are assessed, mitigated, and disclosed — it is not a claim that the AI is never wrong.
Does certification mean Ozwell can be trusted to practice medicine?
No, and it is not intended to. Ozwell’s published intended decision-making role is to inform and augment clinical decision-making, not to replace clinical management. Certification is evidence that the limits of the tool are honestly assessed and disclosed, which is a different and more useful thing than a claim of infallibility.
Where can I read the full disclosure?
All thirty-one source attributes are published on our blog, covering intended use, cautioned uses, training data, fairness and validity testing, known limitations, and ongoing maintenance. Our intervention risk management practices are published alongside them.
Who issued the certification?
Drummond Group, an ONC-Authorized Certification Body and an established independent testing and certification body for health IT. Drummond announced the certification on July 15, 2025.
Certified, and still supporting
See what a clinician-approved draft looks like
Start free on the web. Nothing reaches a patient record until you approve it.
Also worth reading: how Ozwell handles the record.