Can AI Improve Diagnostic Accuracy in Primary Care?

By Rasit Dinc

The integration of artificial intelligence (AI) into healthcare has been a topic of intense discussion and research, with the potential to revolutionize many aspects of medical practice. One of the most promising areas for AI application is in diagnostics, particularly within the demanding environment of primary care. Primary care physicians are tasked with diagnosing a vast spectrum of conditions, often with limited time and resources. This has led to a growing interest in whether AI can serve as a reliable tool to enhance diagnostic accuracy, reduce errors, and ultimately improve patient outcomes.

The Current Landscape: AI vs. Human Expertise

Recent research has sought to quantify the diagnostic capabilities of AI in comparison to human physicians. A systematic review and meta-analysis published in npj Digital Medicine provides a comprehensive overview of this comparison. The study, which analyzed 83 individual studies, found that while there was no statistically significant overall difference in diagnostic accuracy between generative AI models and physicians in general, AI models were found to be inferior to expert physicians. The overall diagnostic accuracy for AI was reported to be 52.1%, which, while promising, underscores the complexity of medical diagnosis and the nuanced expertise that experienced clinicians bring to the table [1]. This suggests that while AI can be a valuable assistant, it is not yet ready to replace the seasoned judgment of a medical expert.

AI in Specialized Areas of Primary Care

The potential of AI in primary care diagnostics becomes even more apparent when looking at specialized domains. A scoping review in the Journal of Medical Internet Research highlights the significant progress of AI in areas such as dermatology and ophthalmology. The review notes that AI tools have demonstrated high diagnostic accuracy, particularly when trained on structured data like ECGs, dermoscopic images, and electronic health records (EHRs). For example, in dermatology, AI-powered assistance has been shown to improve the diagnostic agreement between general practitioners and dermatologists. In ophthalmology, an AI-assisted telemedicine platform achieved an impressive 97% sensitivity and 99% specificity in detecting urgent retinal diseases, demonstrating the potential for AI to triage patients effectively and reduce the workload on specialists [2]. These findings suggest that AI can be a powerful tool for augmenting the diagnostic capabilities of primary care physicians, especially in image-based specialties.

The Rise of Large Language Models in Diagnostics

The advent of powerful large language models (LLMs) like GPT-3 has introduced a new dimension to the discussion of AI in medical diagnostics. A study published in The Lancet Digital Health evaluated the diagnostic and triage accuracy of GPT-3 and found that it performed remarkably well. The model achieved an 88% accuracy in providing a correct diagnosis within its top three suggestions, a figure that is significantly higher than that of lay individuals and close to the 96% accuracy of physicians. However, the study also highlighted a key limitation: while GPT-3 excelled at diagnosis, its triage accuracy was lower than that of both physicians and lay individuals [3]. This suggests that while LLMs can be a valuable source of diagnostic information, their ability to assess the urgency of a situation and provide appropriate triage advice requires further refinement.

The Path Forward: Collaboration and Integration

The evidence to date suggests that AI is not a replacement for human clinicians but rather a powerful collaborator. The most effective use of AI in primary care diagnostics will likely involve a human-in-the-loop approach, where AI provides decision support and highlights potential diagnoses, while the final judgment remains with the physician. This collaborative model leverages the strengths of both AI—its ability to process vast amounts of data and identify patterns—and human clinicians—their contextual understanding, empathy, and ability to navigate the complexities of patient care.

For AI to be successfully integrated into primary care workflows, several challenges must be addressed. These include ensuring data privacy and security, developing user-friendly interfaces that align with clinical workflows, and establishing clear guidelines for the ethical and responsible use of AI in medicine. Furthermore, ongoing research and validation are crucial to ensure that AI tools are both safe and effective in real-world clinical settings.

In conclusion, AI holds immense promise for improving diagnostic accuracy in primary care. By augmenting the capabilities of physicians and providing valuable decision support, AI has the potential to reduce diagnostic errors, improve patient outcomes, and enhance the efficiency of healthcare delivery. However, the path to widespread adoption requires a thoughtful and collaborative approach that prioritizes patient safety and the central role of the clinician in the diagnostic process.

References

[1] Takita, H., Kabata, D., Walston, S. L., Tatekawa, H., Saito, K., Tsujimoto, Y., Miki, Y., & Ueda, D. (2025). A systematic review and meta-analysis of diagnostic performance comparison between generative AI and physicians. npj Digital Medicine, 8(1), 175. https://doi.org/10.1038/s41746-025-01543-z

[2] Katonai, G., Arvai, N., & Mesko, B. (2025). AI and Primary Care: Scoping Review. Journal of Medical Internet Research, 27, e65950. https://doi.org/10.2196/65950

[3] Levine, D. M., Tuwani, R., Kompa, B., Varma, A., Finlayson, S. G., Mehrotra, A., & Beam, A. (2024). The diagnostic and triage accuracy of the GPT-3 artificial intelligence model: an observational study. The Lancet Digital Health, 6(8), e555-e561. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(24)00097-9/fulltext