Advancing Multimodal AI/ML to Enhance HIV Clinical Care

Multimodal data have been generated and collected as part of HIV care delivery and research including, but not limited to, structured data and unstructured data in electronic health records (EHRs), claims data, pharmacy records data, imaging data, omics data and other molecular biomarker data. Such rich data offer great opportunities for harnessing the transformative power of artificial intelligence (AI) and machine learning (ML) to enhance personalized clinical decision support and address unmet needs in HIV prevention and care. Multimodal AI that can integrate multiple modalities of data encountered in clinical practice has been shown to yield superior performance over simpler, unimodal models in various disease areas outside of HIV. However, multimodal biomedical data are typically complex and heterogeneous, and are fraught with missing data and other sources of biases. For example, patients with less access to healthcare or lower socio-economic status tend to have more incomplete data in their EHRs. Thus, advancing multimodal AI for HIV applications faces significant technical challenges in the training, validation, and implementation, including, but not limited to, quantifying the dimension of heterogeneity, identifying interconnections, and addressing missing data. Another major barrier in advancing multimodal AI in HIV applications is that multimodal data in HIV are typically not publicly available. Our project seeks to address these and other challenges through three specific aims. In Aim 1, we will develop novel accurate, efficient and unbiased multimodal AI models for HIV care and prevention. In Aim 2, we will adapt and create causal knowledge graphs to enhance interpretability for applications in HIV care and prevention. In Aim 3, we will develop synergistic integration of knowledge graphs and multimodal AI models for more precise model and increased usability in HIV care and prevention. We will train and test the proposed multimodal AI models and knowledge graphs using multimodal data from the Veteran Health Administration, the largest integrated health system in the US, and the Veteran Aging Cohort Study for three important use cases in HIV prevention and care, namely, 1) identification of HIV patients at risk of medication non-adherence and/or loss to care; 2) prediction of complications of HIV patients; and 3) identification of patients at high risk of HIV infection. Our model development will be guided by ethical principles to ensure data privacy, security, and transparency. We will adopt a human-centered approach that seeks valuable inputs from and meaningful engagements with key stakeholders informed by the theory of Participatory Action Research. Once successfully completed, our project is expected to advance the state-of- the-art multimodal AI and knowledge graphs that can be applied/adapted to other use cases in HIV and transform HIV care and prevention. Project Number: 1R01MH143267-01 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Mental Health (NIMH) | Principal Investigator: Qi Long (+1 co-PI) | Institution: UNIVERSITY OF PENNSYLVANIA, PHILADELPHIA, PA | Award Amount: $1,223,841 | Activity Code: R01 | Study Section: Special Emphasis Panel[ZRG1 NV-J (53)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11309283