Diagnostic Prediction Models for Primary Care, Based on AI and Electronic Health Records: Systematic Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Primary Care Research

Background:

• Artificial intelligence (AI) offers potential for enhancing diagnostic accuracy in primary care (PC) by leveraging electronic health record (EHR) data.
• Despite AI's potential, a systematic evaluation of AI-based diagnostic prediction models using PC EHR data is lacking.
• Existing research has explored various prediction models based on EHR data, but a comprehensive review is needed.

Purpose of the Study:

• To systematically evaluate AI-based diagnostic prediction models developed using PC EHR data.
• To assess the content, risk of bias, and applicability of these AI models.
• To identify gaps in the current research and clinical readiness of these tools.

Main Methods:

• A systematic review adhering to PRISMA guidelines was conducted.
• Searches were performed across major databases (MEDLINE, Embase, Web of Science, Cochrane).
• Studies developing or validating AI diagnostic prediction models using PC EHR data were included; risk of bias and applicability were assessed using PROBAST.

Main Results:

• Out of 10,657 records, 15 papers were selected, with most focusing on a single chronic condition.
• Only two studies externally validated models in a PC setting; thirteen developed models.
• A high risk of bias was found in 60% of studies, with unclear applicability in 67% due to reporting deficiencies.

Conclusions:

• Most AI-based diagnostic prediction models in PC focus on single chronic conditions and lack robust external validation in primary care settings.
• Significant methodological limitations and a high risk of bias were identified, hindering clinical implementation.
• Current AI diagnostic prediction models are not sufficiently developed or validated for routine use in primary care.