AI Insight
Researchers developed a system to create disease case-finding models using summary data from the All of Us Research Program that can identify patients with potentially undiagnosed conditions without extracting protected individual patient data from the research enclave. They built nine models for conditions including Long COVID, chronic fatigue, and fibromyalgia, then deployed them through standard healthcare IT interfaces (FHIR, SMART on FHIR, and CDS Hooks) that can flag at-risk patients for clinician review. The models showed moderate to good performance in validation testing, with accuracy scores ranging from 0.61 to 0.82 for identifying self-reported conditions without diagnostic codes, and the system successfully reproduced exact results across different implementations in synthetic patient records.
Why it matters
This approach enables researchers to develop clinically useful diagnostic tools from large research databases while maintaining strict patient privacy protections, potentially helping clinicians identify patients with underdiagnosed conditions. If validated in real-world settings, this framework could bridge the gap between research data and clinical care by making enclave-derived models widely accessible through existing electronic health record systems.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Background. Research data enclaves such as the All of Us Research Program do not allow participant-level data, or models fitted to them, to leave, which makes it hard to deliver case-finding models, which flag patients with possibly uncoded conditions for clinician review, to patients and clinicians. Objectives. We describe and technically evaluate a system that fits models outside an enclave from released summary counts and delivers them through standard health information technology interfaces. Methods. We exported dissemination-compliant summary counts from the All of Us Researcher Workbench and, outside the enclave, fitted versioned models for Long COVID, chronic fatigue, fibromyalgia, endometriosis, Lyme disease, lupus, migraine, depression, and anxiety. A model contract (a model file and a written specification), a Python reference implementation, and synthetic Fast Healthcare Interoperability Resources (FHIR) test patients define the expected outputs. One JavaScript scorer runs in a patient demonstration, a SMART on FHIR clinician application, and a CDS Hooks service. An atlas displays aggregate results. We evaluated the system on synthetic records with automated tests, cross-implementation output comparison, launches through the public SMART sandbox, and a counterexample review of the scoring and display rules. Results. The deployed Long COVID model had an area under the receiver operating characteristic curve of 0.67 in a random validation split. The eight self-report models reached 0.61 to 0.82 in identifying self-reporters without a code. The JavaScript scorer reproduced all 162 reference outputs exactly, and all 437 automated tests passed. A counterexample review found 5 classes of failure, including scoring ineligible patients and treating unavailable data as absent, all fixed and added as tests. Conclusions. Models fitted from released summary counts can be delivered through FHIR, SMART on FHIR, and CDS Hooks with reproducible scoring on synthetic records. Real-record interoperability, usability, and prospective clinical performance remain to be evaluated.