Epidemiological Analysis and Causal Modeling of Non-AIDS-Defining Skin Cancers in People with HIV - PROJECT SUMMARY/ABSTRACT This application seeks to understand the epidemiological patterns and complex interplay of risk factors driving skin cancer incidence and severity among people with HIV (PWH) and to develop predictive tools that address disparities in diagnosis and prognosis. PWH face a higher likelihood of developing non-AIDS-defining skin cancers (NADSCs), including but not limited to basal cell carcinoma, squamous cell carcinoma, and melanoma, due to multifactorial biopsychosocial determinants of health. Despite this risk, the mechanisms underlying NADSC incidence and severity in PWH remain under-researched, and no tailored predictive models exist to guide targeted interventions for this vulnerable population. This proposal describes the application of epidemiological and machine learning approaches, including causal artificial intelligence (AI) methods, specifically probabilistic graphical models (PGMs), with two objectives: 1) characterizing patterns of NADSC burden among PWH and 2) identifying direct effectors among risk factors contributing to NADSC in PWH and developing robust predictive models accordingly. By leveraging multimodal healthcare data containing sociodemographic and clinical data, I aim to address critical gaps in the identification of high-risk subgroups and the development of targeted detection and prevention strategies. The central hypothesis of this proposal is that NADSC incidence and severity in PWH are influenced by a complex interplay of causal factors that can be modeled to inform personalized clinical interventions. In Aim 1, I will characterize the epidemiological patterns of NADSC burden (prevalence, disease sub-types, severity) among PWH compared to people without HIV. In Aim 2, I will identify factors that directly affect NADSC incidence and severity in PWH. Specifically, I will integrate high-dimensional data sources to uncover interactions between risk factors such as immunosuppression, comorbidities, medication history, and sociodemographic identifiers. I will then build and validate predictive models for NADSC incidence and severity among PWH, focusing on the integration of multimodal data to develop robust and clinically actionable models. These models will be validated using an independent dataset to ensure generalizability. This work is significant because it addresses a critical gap in understanding and predicting NADSC incidence and severity in PWH, a population with unique sociomedical vulnerabilities. The project is innovative in its use of causal AI and multimodal data to elucidate novel relationships and create predictive tools tailored to the needs of PWH. These findings will advance the field of onco-dermatology and contribute to reducing disparities in skin cancer outcomes through data-driven clinical and public health strategies.
