Keyword: Implementation Science
1 result found.
Review Article
Epidemiology and Health Data Insights, 2(5), 2026, ehdi063, https://doi.org/10.63946/ehdi/19706
ABSTRACT:
Background: Generative AI and large language models are changing how people seek health information, but their effect on maternal health in sub-Saharan Africa is unclear. This narrative review synthesizes evidence on generative AI and maternal health misinformation, examines community health workers (CHWs) as information intermediaries, and proposes a conceptual framework for safe AI implementation.
Methods: We ran a purposive search of PubMed/MEDLINE, Web of Science, Scopus, ACM Digital Library, and Google Scholar for publications from January 2014 to August 2026. We synthesized findings narratively and used the Consolidated Framework for Implementation Research (CFIR) to organize implementation determinants.
Results: We found no study of generative AI maternal health misinformation in sub-Saharan Africa. Evidence from high-income settings and non-maternal topics shows AI answers on reproductive health can be inaccurate or inconsistent, and fluent AI text can seem more compelling than human writing. Multi-specialty African benchmarks and a global benchmark review point to a gap between benchmark performance and clinical safety, but maternal-specific performance is unreported. CHWs have a well-documented record of improving maternal and child health, but their ability to identify AI errors is untested. African digital health implementations highlight user-centered design, interoperability, and sustained investment.
Conclusions: We propose, as hypotheses for evaluation, localized benchmarking, retrieval-augmented generation restricted to national guidelines with added safeguards, CHW curricula that include AI literacy, and decentralized research hubs. Unvalidated AI could bypass CHW authority and affect trust, though this is unstudied. Priorities include measuring AI use among pregnant women and prospectively evaluating CHW-facing tools.
Methods: We ran a purposive search of PubMed/MEDLINE, Web of Science, Scopus, ACM Digital Library, and Google Scholar for publications from January 2014 to August 2026. We synthesized findings narratively and used the Consolidated Framework for Implementation Research (CFIR) to organize implementation determinants.
Results: We found no study of generative AI maternal health misinformation in sub-Saharan Africa. Evidence from high-income settings and non-maternal topics shows AI answers on reproductive health can be inaccurate or inconsistent, and fluent AI text can seem more compelling than human writing. Multi-specialty African benchmarks and a global benchmark review point to a gap between benchmark performance and clinical safety, but maternal-specific performance is unreported. CHWs have a well-documented record of improving maternal and child health, but their ability to identify AI errors is untested. African digital health implementations highlight user-centered design, interoperability, and sustained investment.
Conclusions: We propose, as hypotheses for evaluation, localized benchmarking, retrieval-augmented generation restricted to national guidelines with added safeguards, CHW curricula that include AI literacy, and decentralized research hubs. Unvalidated AI could bypass CHW authority and affect trust, though this is unstudied. Priorities include measuring AI use among pregnant women and prospectively evaluating CHW-facing tools.
