JMIR AI · 2026
AI Model Reporting Standards in Health Care: Systematic Review and Consolidation.
СтатьяОбзорНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Akhmad Ekaterina, Slob Daniël, Gomes Aiara Lobo, Mingels Simone, Dekker Andre, Soest Johan van
Аннотация
BACKGROUND: AI technologies are currently experiencing significant growth in development and use in health care. The AI model lifecycle includes several stages, all of which should be adequately described. Reporting standards are commonly used for this purpose. However, the plain-text structure of these standards prevents transparent machine-readability of this information.
OBJECTIVE: This systematic review aims to investigate AI model reporting standards from clinical databases and technical documents published since 2016.
METHODS: The methodological systematic review was conducted according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. The inclusion criteria of this systematic review were papers published from 2016 to April 2026 in English and in a peer-reviewed journal with open access. We aimed to include standards for different data types (structured, images, and text-based). The exclusion criteria were specific medical domain, incomplete reports, and reviews. The following libraries have been consulted: Scopus and Web of Science, MEDLINE and Embase, the IEEE Xplore digital library, and the ACM Digital Library. We applied an Automated Systematic Review software tool (Utrecht University) to find relevant papers after records retrieval. Two reviewers (EA, DS) screened relevant papers independently and invited a third reviewer (JvS) to reach a consensus. All included papers were analyzed to extract relevant lifecycle stage, publication type, data type, and reporting questions. The results were presented and synthesized in a consolidated list by grouping questions and ordering them in a logical sequence according to the AI model lifecycle.
RESULTS: The total number of included studies was 25 from clinical and technical databases (22 and 3, respectively). The consolidated list includes 134 questions ordered in 7 groups: "Problem identification," "Data collection and processing," "Model development," "Technical validation," "Functional validation," "Operation & Monitoring," and one additional group with "Statements." Groups according to the lifecycle stages contain all relevant information for each stage. Both clinical and technical papers reflect information for "Problem identification" (s=5), "Data collection and processing" (s=25), "Model development" (s=13), "Technical validation" (s=14). While "Functional validation" (s=30) was represented mostly in clinical papers and "Operation & Monitoring" (s=18) in technical ones. In the "Statements" (s=29), all general questions were summarized, describing reusability and reproducibility of the model, and highlighting ethical considerations, such as model fairness, explainability, interpretability, and related risk. It was shown that only 10 out of 15 FAIR (findable, accessible, interoperable, reusable) principles were addressed in the consolidated list, with focus on providing rich metadata (F2) and accurate and relevant descriptions (R1).
CONCLUSIONS: We furthermore considered the mapping of the consolidated questions with FAIR principles. In future, we aim to incorporate the results of our work into a metadata schema for AI models developed according to FAIR principles.
OBJECTIVE: This systematic review aims to investigate AI model reporting standards from clinical databases and technical documents published since 2016.
METHODS: The methodological systematic review was conducted according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. The inclusion criteria of this systematic review were papers published from 2016 to April 2026 in English and in a peer-reviewed journal with open access. We aimed to include standards for different data types (structured, images, and text-based). The exclusion criteria were specific medical domain, incomplete reports, and reviews. The following libraries have been consulted: Scopus and Web of Science, MEDLINE and Embase, the IEEE Xplore digital library, and the ACM Digital Library. We applied an Automated Systematic Review software tool (Utrecht University) to find relevant papers after records retrieval. Two reviewers (EA, DS) screened relevant papers independently and invited a third reviewer (JvS) to reach a consensus. All included papers were analyzed to extract relevant lifecycle stage, publication type, data type, and reporting questions. The results were presented and synthesized in a consolidated list by grouping questions and ordering them in a logical sequence according to the AI model lifecycle.
RESULTS: The total number of included studies was 25 from clinical and technical databases (22 and 3, respectively). The consolidated list includes 134 questions ordered in 7 groups: "Problem identification," "Data collection and processing," "Model development," "Technical validation," "Functional validation," "Operation & Monitoring," and one additional group with "Statements." Groups according to the lifecycle stages contain all relevant information for each stage. Both clinical and technical papers reflect information for "Problem identification" (s=5), "Data collection and processing" (s=25), "Model development" (s=13), "Technical validation" (s=14). While "Functional validation" (s=30) was represented mostly in clinical papers and "Operation & Monitoring" (s=18) in technical ones. In the "Statements" (s=29), all general questions were summarized, describing reusability and reproducibility of the model, and highlighting ethical considerations, such as model fairness, explainability, interpretability, and related risk. It was shown that only 10 out of 15 FAIR (findable, accessible, interoperable, reusable) principles were addressed in the consolidated list, with focus on providing rich metadata (F2) and accurate and relevant descriptions (R1).
CONCLUSIONS: We furthermore considered the mapping of the consolidated questions with FAIR principles. In future, we aim to incorporate the results of our work into a metadata schema for AI models developed according to FAIR principles.