Artificial intelligence research practices in primary healthcare institutions in China and the potential issue of data poverty: a scoping review

Artificial intelligence (AI) technology plays a significant role in enhancing the quality of primary healthcare and promoting health equity. China has introduced multiple key policies to encourage the development and application of AI in primary healthcare institutions. However, health data poverty, defined as the difficulty individuals or populations face in benefiting from technological innovation due to a lack of representative data, may exacerbate the gap in medical service quality between primary healthcare institutions and general hospitals. Currently, there is a lack of systematic review and research on the development status of AI in primary healthcare in China and its associated health data poverty challenges.

A systematic search was conducted in PubMed, Embase, Web of Science, CNKI, and Wanfang for studies published between 2009 and 2025 on AI models in primary healthcare in China. Data were extracted and analyzed across three dimensions: model characteristics (algorithm type, validation method, and results), database characteristics (data collection location, data acquisition, and content), and metadata distribution (population coverage, missing value handling, etc.).

A total of 57 studies were finally included. Existing primary healthcare AI models in China are primarily based on machine learning and deep learning, covering various application scenarios such as disease prediction and diagnosis. The diseases focused on are mainly endocrine disorders (e.g., diabetes), cardiovascular diseases (e.g., hypertension), and mental health conditions. However, databases used for developing or validating these primary healthcare AI models face several challenges: insufficient accessibility(no fully public datasets), poor reporting quality (average score: 6.53 out of 10; nearly one-third failed to report inclusion/exclusion criteria or missing value handling), and inadequate sample representativeness (only 13.5% of the population was under 45 years old, with a dominance of data from developed regions).

Facilitating health system data interoperability, developing high-quality research-ready datasets and establishing a comprehensive oversight system are effective strategies to address health data poverty in primary healthcare institutions in China. These measures will help provide data support for the high-quality development of primary healthcare services.