Proactive HIV Care: AI-Based Comorbidity Prediction from Routine EHR Data

2025-08-26

AI Summary

Early detection of multiple comorbidities in people living with HIV (PLWH) remains challenging due to reliance on symptomatic presentation, often leading to diagnostic delays. Method: We propose a demographic-aware, multi-label comorbidity prediction framework leveraging routine electronic health record (EHR) data from 2,200 PLWH. The model integrates 30 laboratory biomarkers and 7 sociodemographic variables within an enhanced machine learning architecture; we systematically compare demographic-aware versus agnostic variants. Results: Our framework significantly improves predictive accuracy for key comorbidities—including diabetes, chronic kidney disease, and cardiovascular disease. Notably, we demonstrate for the first time that gender and age can be implicitly inferred from laboratory values alone, revealing potential algorithmic bias and underscoring the critical need for fairness and interpretability in clinical AI. This work establishes a novel paradigm for proactive, equitable, and interpretable comprehensive health management in HIV care.

Technology Category

• Machine Learning: Hardware-aware ML
• Humans and AI: Human-in-the-loop Machine Learning
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

People living with HIV face a high burden of comorbidities, yet early detection is often limited by symptom-driven screening. We evaluate the potential of AI to predict multiple comorbidities from routinely collected Electronic Health Records. Using data from 2,200 HIV-positive patients in South East London, comprising 30 laboratory markers and 7 demographic/social attributes, we compare demographic-aware models (which use both laboratory/social variables and demographic information as input) against demographic-unaware models (which exclude all demographic information). Across all methods, demographic-aware models consistently outperformed unaware counterparts. Demographic recoverability experiments revealed that gender and age can be accurately inferred from laboratory data, underscoring both the predictive value and fairness considerations of demographic features. These findings show that combining demographic and laboratory data can improve automated, multi-label comorbidity prediction in HIV care, while raising important questions about bias and interpretability in clinical AI.

Problem

• Predicting HIV comorbidities from routine EHR data using AI
• Comparing demographic-aware versus unaware AI prediction models
• Evaluating fairness and bias in clinical AI for comorbidity screening

Innovation

• AI predicts HIV comorbidities from EHR data
• Demographic-aware models outperform unaware counterparts
• Combining demographic and lab data improves prediction