Screening Electronic Records for Veterans whoshould be referred for specialty care Evaluations(SERVE)

Background: Missed and delayed diagnosis of chronic diseases contribute to damaging consequences that can be prevented with timely detection and treatment. Axial spondyloarthritis (axSpA) is an example of a rheumatic disease with frequent and potentially devastating diagnostic delays, averaging 6-14 years. AxSpA affects approximately 174,000 Veterans, with chronic pain and damage to the spine and joints. AxSpA progressively impairs function, such that 75% of affected Veterans over the age of 60 experience walking limitations and 45% experience severe limitations in activities of daily living. We developed a machine learning (ML) algorithm that classifies Veterans as having or not having axSpA, and we discovered many Veterans at high risk for undiagnosed axSpA. The objectives of this proposal are to develop a ML algorithm to identify Veterans who should receive a diagnostic evaluation for axSpA, incorporate Veteran preferences for applying the algorithm, and prospectively validate the screening process (ML algorithm + Veteran engagement + diagnostic evaluation) with in-clinic diagnostic evaluations. Significance: This research provides the foundation for a disease screening program that will identify Veterans at risk for a treatable but frequently undiagnosed disease. It will also develop and test processes for connecting these Veterans with the soonest and best care for their disease. Innovation & Impact: The translation of an advanced ML screening approach from computer bench to clinical care is innovative with complex chronic diseases. Furthermore, this screening approach does not require Veterans or their providers to complete screening procedures prior to a diagnostic evaluation; thus deploying this screening approach in large populations is practical. We estimate detecting ~13,000 Veterans with previously undiagnosed axSpA, when our screening process is applied nation-wide. This study will also provide valuable insights for engaging, educating, and referring Veterans at risk of undiagnosed disease to appropriate specialty care providers. Specific Aims: Aim 1: Derive and validate a ML algorithm to predict an incident axSpA diagnosis. Aim 2: Learn and incorporate Veteran and provider perspectives for applying the screening process. Aim 3. Evaluate the impact of the axSpA screening process in real-world clinic settings. Methodology: Aim 1: We will extend our prior work that uses ML to identify prevalent axSpA to derive a new ML algorithm using an ensemble method including traditional ML techniques and deep learning. We will optimize algorithm performance by adding additional data sources, input features, and time series modeling that reflect the temporal and sequential nature of the clinical process (i.e. recurrent neural networks). Aim 2: We will conduct focus groups with Veterans, primary care providers and rheumatologists to learn their preferences regarding communication, education, and referral options. Aim 3: We will prospectively validate the screening process with Veterans who screen positive by the algorithm with a diagnostic evaluation with a rheumatologist. Next Steps/Implementation: We will study implementation strategies for widespread dissemination of the screening process and seek opportunities to expand our screening approach to other diseases. Project Number: 1I01HX003705-01A2 | Fiscal Year: 2026 | NIH Institute/Center: Veterans Affairs (VA) | Principal Investigator: Jessica Walsh | Institution: VA SALT LAKE CITY HEALTHCARE SYSTEM, SALT LAKE CITY, UT | Activity Code: I01 | Study Section: HSR-3 Healthcare Informatics & Access to Care[HSR3] View on NIH RePORTER: https://reporter.nih.gov/project-details/11107307