@article {bmjPosttill2026,
	author = {Postill, Gemma and Sedlakova, Jana and Tancredi, Stefano and Smit, Frerik and Ekongefeyin, Sintieh and Baumer, Andreas M and Chiolero, Arnaud and Bernard, J{\"u}rgen and Rosella, Laura C and Wyl, Viktor von},
	title = {P318 Integrating artificial intelligence tools in population health research},
	volume = {80},
	number = {Suppl 2},
	pages = {A187--A188},
	year = {2026},
	doi = {10.1136/jech-2026-SSMabstracts.377},
	publisher = {BMJ Publishing Group Ltd},
	abstract = {Background Artificial intelligence (AI)-enabled research tools are increasingly used to support population health research, including the analysis of tabular health data. Although combining human expertise with AI has the potential to improve efficiency, standardization, and research quality, these benefits have not been widely realized at scale. A key challenge is that disciplines such as epidemiology rely on structured, discipline-specific workflows that differ from iterative data science approaches commonly embedded in AI tools. Limited guidance exists on how to integrate AI-enabled tools without disrupting established research practices. As a result, researchers and trainees may adopt these tools in ways that inadvertently alter workflows and compromise methodological rigour.Methods We examined how differences between traditional epidemiologic workflows and AI-driven analytical processes may influence the use of AI-enabled research tools. Epidemiology was used as a case study to generate practical, discipline-specific insights. We reviewed the stages of the research workflow and identified points where AI integration may introduce risks or opportunities. We applied several different generative AI tools to conduct an epidemiology research question. Based on such analysis, we developed actionable guidance for aligning AI tool use with established research standards and for educating the next generation of health researchers.Results Differences between epidemiological research workflows and AI-enabled approaches can lead to misalignment in study design, data handling, analysis, and interpretation if not carefully managed. Our case study demonstrates that, without appropriate scaffolding, AI tools may inadvertently propagate poor scientific practices or reduce methodological transparency. Practical recommendations to support responsible adoption and workflow-aligned implementation include:Never assume shared meaning between the terminology of data science to ensure appropriate interpretation of AI-enabled research tools;Break down goals and processes into clearly documented tasks and steps;Define the purpose of an AI-enabled tool regarding the specific workflow step and degree of human involvement applied;Do not assume how AI-enabled tools work; andApproach AI-enabled workflow steps like a peer review (reject, revise, and accept).Conclusion AI-enabled research tools hold significant promise for advancing health research. Their benefits depend on thoughtful integration with discipline-specific workflows. Using epidemiology as a case study, we provide guidance for leveraging these tools in ways that support methodological integrity and high-quality outputs. Preparing the next generation of population health researchers will require targeted training that emphasizes both technical proficiency and adherence to established research practices, ensuring that AI augments scientific rigour.},
	issn = {0143-005X},
	URL = {https://jech.bmj.com/content/80/Suppl_2/A187.3},
	eprint = {https://jech.bmj.com/content/80/Suppl_2/A187.3.full.pdf},
	journal = {Journal of Epidemiology \& Community Health},
	pdf = {https://jech.bmj.com/content/jech/80/Suppl_2/A187.3.full.pdf},
	teaserPage = {1},
	topics = {Interactive Machine Learning, Explainable and Trustworthy AI, Visual Analytics in Health Care, Experiments and Empirical Work},
	code = {J043}
}