Policy-Embedded Graph Expansion: Networked HIV Testing with Diffusion-Driven Network Samples

Researchers have developed Policy-Embedded Graph Expansion (PEGE), a novel AI framework for improving HIV testing efficiency in real-world settings. The approach combines intelligent sequential testing algorithms with diffusion-based network modeling to identify cases more effectively, achieving 15.4% more HIV detections while testing only 25% of populations.

This research addresses a critical gap between theoretical network-based testing algorithms and practical public health deployment. Traditional approaches assume complete network information upfront, but real HIV referral networks emerge incrementally through contact tracing and testing cascades. PEGE solves this by embedding probabilistic graph expansion directly into the decision-making policy rather than attempting full network reconstruction—a pragmatic shift that acknowledges real-world data constraints.

The introduction of Dynamics-Driven Branching (DDB) leverages diffusion models to predict how transmission networks naturally expand through forest-like structures inherent in referral chains. This design choice reflects domain expertise: HIV contact networks in resource-limited settings typically follow hierarchical patterns rather than complex topologies. The 17.3% improvement in discounted reward and 15.4% increase in detection rates while testing only 25% of populations demonstrates substantial efficiency gains.

For global health systems, this represents tangible progress toward UN Sustainable Development Goal 3.3. Early intervention and case detection directly reduce transmission rates and improve patient outcomes. The framework's viability in collaboration with the WHO and University of Witwatersrand suggests near-term implementation potential in sub-Saharan Africa and other high-burden regions.

The work validates a broader trend: combining reinforcement learning with domain-specific generative models outperforms generic sequential decision-making. Future deployment will depend on integration with existing surveillance systems and stakeholder adoption, but the research establishes both algorithmic feasibility and public health impact metrics that funding agencies and governments track closely.

• →PEGE + DDB framework achieves 15.4% more HIV detections while testing only 25% of populations, demonstrating significant efficiency gains.
• →Policy-embedded graph expansion avoids impractical full-network reconstruction by embedding probabilistic distributions directly into decision policies.
• →Diffusion-based network modeling is optimized for forest-like referral structures natural to real-world HIV contact tracing.
• →Research collaboration with WHO and University of Witwatersrand indicates readiness for real-world deployment in resource-limited settings.
• →Framework advances progress toward UN SDG 3.3 by enabling targeted, efficient testing strategies for epidemic control.