Beyond Prompting: AI for Climate-Sensitive Disease Reviews (23880)
Background
Pacific Islands regularly encounter outbreaks of climate sensitive diseases such as leptospirosis and dengue. Robust disease spread models are required to guide policy and preparedness in a changing climate. In the absence of complete and robust field data, parametrising these models requires systematic review of evidence from large bodies of scientific literature, which remains slow, labour intensive, and difficult to scale. Artificial intelligence offers a path forward, but only if the underlying tools are reproducible, explainable and able to generalise across disease contexts.
Methods
Study ranking and title-abstract screening are two sequential stages of an automated literature screening pipeline. We investigated AI based approaches for both tasks. For ranking, a large language model (LLM) was used to assess article relevance by framing selection criteria as structured question-answer tasks across four rainfall-focused zoonotic disease reviews. For screening, we evaluated automatic prompt optimisation using student-teacher LLM configurations across climate-sensitive zoonotic disease reviews and external benchmark datasets to examine when performance gains could be achieved.
Results
AI-assisted ranking reduced manual screening effort by at least 70% at a 95% recall threshold, producing transparent rationales alongside each decision. Automatic prompt optimisation improved screening performance in 10 of 11 reviews, with gains observed across both zoonotic disease and out-of-domain reviews. Variation in screening outcomes was more strongly associated with review-level characteristics than subject area.
Conclusion
Together these studies demonstrate the potential of AI to accelerate evidence synthesis while supporting explainability and reproducibility. Despite the limited evidence base for small island contexts, particularly in the Pacific, AI-based approaches can facilitate systematic navigation of the literature, gap identification, and synthesis of available evidence. These studies form the foundation of CliZod, a Climate Sensitive Zoonotic Disease Database being developed as an open collaborative platform for identifying, extracting and storing disease parameters from scientific literature.
- Bolanos, F., Salatino, A., Osborne, F., & Motta, E. (2024). Artificial intelligence for literature reviews: Opportunities and challenges. Artificial Intelligence Review, 57(10), 259.
- Bowen, K. J., Ebi, K. L., Woodward, A., McIver, L., Tukuitonga, C., & Nayna Schwerdtle, P. (2024). Human health and climate change in the Pacific: A review of current knowledge. Climate and Development, 16(2), 119–133. https://doi.org/10.1080/17565529.2023.2185479
- Ezanno P, Picault S, Beaunée G, Bailly X, Muñoz F, Duboz R, Monod H, Guégan JF. (2021) Research perspectives on animal health in the era of artificial intelligence. Vet Res. 2021 Mar 6;52(1):40. doi: 10.1186/s13567-021-00902-4. PMID: 33676570; PMCID: PMC7936489.
- Thompson, R. N., & Brooks-Pollock, E. (2019). Detection, forecasting and control of infectious disease epidemics: Modelling outbreaks in humans, animals and plants. Philosophical Transactions of the Royal Society B: Biological Sciences, 374(1775), 20190038.
- Sujau, M., Wada, M., Vallée, E., Hillis, N., & Sušnjak, T. (2025). Accelerating Disease Model Parameter Extraction: An LLM-Based Ranking Approach to Select Initial Studies for Literature Review Automation. Machine Learning and Knowledge Extraction, 7(2), 28. https://doi.org/10.3390/make7020028
