Phase 2 of MVP-CHAMPION confirmed that strategic alignment of data, computation, and multidisciplinary expertise can not only generate new scientific knowledge but also produce tools, models, and insights directly ready for clinical application. The lessons learned here, about integration, infrastructure, multimorbidity, equity, and translation, provide a blueprint for scaling this approach nationally and sustaining its impact over time. The success of Phase 2 of MVP-CHAMPION demonstrates that federal collaboration, when intentionally structured around real-world clinical needs, high-performance computation, and scalable data assets, can meaningfully accelerate both scientific discovery and health system transformation. Yet Phase 2 represents not an endpoint, but a foundation, a first proof that a new paradigm for precision health is achievable at national scale. Building on this momentum, we envision a future in which the VA-DOE collaboration continues to evolve into a flagship national model for personalized, data-driven, preventive, and efficient healthcare. Efforts towards these goals are actively ongoing.

1. Federal integration and collaboration is essential. The most successful projects embedded VA clinicians and researchers, DOE data scientists, DOE computational biologists, and VA operational leaders into a single, iterative workflow. Close integration ensured that clinical questions were framed in a computationally tractable way, analytic methods were chosen with real-world deployment in mind, and outputs were immediately interpretable by clinicians. Early and ongoing collaboration across disciplines accelerated both scientific innovation and translational readiness. The VA-DOE collaboration provides a scalable blueprint for larger national efforts in the future. By expanding partnerships with other federal agencies, healthcare systems, and industry collaborators, MVP-CHAMPION can help accelerate discovery, validation, and deployment of precision health solutions not only for Veterans, but for the American public at large.
2. Building a flexible, reusable secure infrastructure and pipeline for implementation saves time and costs. Investments in shared computational assets, such as the VA Large Language Model (VA-LLM), the GRIN, VISION, and MENTOR network analysis tools, and common EHR pipelines, dramatically reduced project startup time, eliminated duplication of effort, and allowed knowledge and methods to flow across disease areas. Projects that could immediately leverage these assets progressed more rapidly from concept to discovery to translation. Many of the predictive models, AI tools, and computational pipelines developed in Phase 2 are now poised for broader operational deployment. In the future, expanding precision outreach frameworks, dynamic risk prediction dashboards, and multimorbidity-aware clinical decision support tools across the VA system could deliver immediate improvements in Veteran health, particularly for high-burden conditions such as heart failure, lung cancer, and diabetes.
3. Modeling Multimorbidity Is Critical for Real-World Relevance Projects that explicitly incorporated multimorbidity into their models consistently produced findings that were more predictive, clinically meaningful, and operationally actionable. Veterans rarely present with isolated disease states; efforts that acknowledged the complex, interacting nature of multiple conditions were better aligned with actual clinical realities and provided more accurate risk stratification and intervention targets. Phase 2 made substantial advances in integrating structured EHR data, unstructured clinical notes, genomics, and environmental data. The next frontier is seamless fusion across even more modalities—imaging, wearable device data, social determinants of health—to model the full complexity of Veteran health. Moreover, expanding multimorbidity modeling from a secondary consideration to a primary analytic focus will better reflect real patient trajectories and therapeutic needs. The emergence of digital twin technologies, computational representations of individual patients that evolve with longitudinal multimodal data, offers a powerful opportunity to simulate disease progression and assess personalized interventions. These virtual patient models can help test therapeutic strategies in silico, reducing risk and accelerating translation into clinical practice.
4. Exascale Computing Unlocks Healthcare Scale and Complexity Analyses of the scale required—encompassing millions of Veterans, multimodal data, billions of clinical notes, and multiscale biological data—would not have been feasible without access to DOE’s leadership-class computing infrastructure. High-throughput modeling, large-scale machine learning, and integration of structured and unstructured data all depended on DOE’s computational environment and expertise. Future expansion will continue to rely on scalable HPC and AI ecosystems. As healthcare data continues to accumulate, static risk models rapidly become outdated. Future MVP-CHAMPION efforts will aim to focus on creating real-time, continuously learning systems, adaptive models that update dynamically with new EHR entries, genomic discoveries, environmental exposures, and treatment innovations. This will move VA closer to a true learning health system where prediction and intervention evolve alongside clinical practice.
5. Early Attention to Operational Translation Increases Impact Projects that engaged operational partners from the outset, such as the Suicide Risk prediction effort, were better positioned to move from discovery to deployment. Benefits of early engagement were numerous, including assessments of fruitful directions for improvement, understanding of existing processes, and detailed discussions of appropriate validation strategies and problem formulations which produce actionable information and scalable implementation.
6. The VA Continues to be a Valuable Source of Multi-Ancestry Genomic Analyses As found previously within MVP, Genetic discoveries validated across ancestry groups were more robust and generalizable. Additionally, 12% of identified genomic risk loci in the MVP data were only significant in people of non-European ancestry (Verma, ref 14 below). Future phases must continue to prioritize within and cross-ancestry analyses, validation across demographic groups, and ensuring that precision health benefits extend to all Veterans, regardless of background or service history. Investment in equitable data collection, inclusive model development, and focused deployment in underserved populations will be critical.
7. Administrative and Regulatory Alignment Enables Speed The Interagency Agreement between VA and DOE, along with clear regulatory frameworks for data sharing and security, was foundational. All projects leveraged pre-approved environments, such as the VA KDI environment at the Oak Ridge National Laboratory S and DOE’s exascale computing facilities with appropriate authority to operate (ATO), were able to move faster without compromising privacy or compliance. The healthcare challenges of the next decade will require even greater computational innovation. Continued collaboration with DOE will ensure that VA researchers and clinicians can access next-generation AI methods (e.g., federated learning, explainable multimodal transformers), petascale-to-exascale computing, and cutting-edge simulation tools to push the boundaries of predictive, preventive, and personalized medicine.