Exploring the Relationship Between Inquiry and Diagnosis Using Advanced Patient Simulators

Online medical consultation (OMC) restricts physicians to gathering patient information solely through targeted questions, making the already complex sequential decision-making process of diagnosis even more challenging. Recent rapid advancements in large language models (LLMs) have demonstrated significant potential for transforming OMC. However, most studies have primarily focused on improving diagnostic accuracy under conditions of relatively sufficient information, while paying little attention to the "inquiry" phase of the consultation process. This lack of focus has resulted in an inadequate exploration of the relationship between "inquiry" and "diagnosis."

A new research approach leverages realistic patient simulators to address this gap. By extracting real-world patient interaction strategies from authentic doctor-patient conversations and using them to guide the training of a patient simulator that accurately reflects real-world behavior, extensive experiments can be conducted. By inputting patient records into the simulator to simulate patient responses, the relationship between "inquiry" and "diagnosis" in the consultation process can be thoroughly investigated.

The Interplay of Inquiry and Diagnosis

The results of these experiments demonstrate that inquiry and diagnosis follow Liebig's Law: poor inquiry quality limits the effectiveness of diagnosis, regardless of diagnostic capability, and vice versa. Simply put, just as a plant's growth is limited by the nutrient present in the minimum amount, the accuracy of a diagnosis is limited by the quality of the preceding inquiry.

Furthermore, the experiments revealed significant differences in the inquiry performance of different models. To investigate this phenomenon, the inquiry process was divided into four categories:

- Obtaining the chief complaint - Specifying known symptoms - Inquiring about associated symptoms - Obtaining family or medical history

Analyzing the distribution of questions across the four categories for different models allows for the exploration of the causes for the significant performance differences. This categorization helps to identify the strengths and weaknesses of various AI models in terms of information gathering and to pinpoint areas for improvement.

Outlook and Open-Source Initiative

The developers of the patient simulator plan to make the weights and associated code open source. This will allow other researchers to build upon this work, conduct their own experiments, and contribute to the further development of AI-assisted diagnostic systems. The open-source initiative promotes transparency and collaboration in this important research field and accelerates progress toward improved patient care.

Exploring the relationship between inquiry and diagnosis using advanced patient simulators is a promising approach to addressing the challenges of online medical consultation. By combining realistic patient simulations with the analysis of AI models, valuable insights into the diagnostic process can be gained and the accuracy of diagnoses improved in the future.

Bibliography

- Liu, Z., Tu, Q., Ye, W., Xiao, Y., Zhang, Z., Cui, H., Zhu, Y., Ju, Q., Li, S., & Xie, J. (2025). Exploring the Inquiry-Diagnosis Relationship with Advanced Patient Simulators. arXiv preprint arXiv:2501.09484. - Paans, W., Sermeus, W., Nieweg, R. M. B., Krijnen, W. P., & van der Schans, C. P. (2012). Do knowledge, knowledge sources and reasoning skills affect the accuracy of nursing diagnoses? a randomised study. BMC nursing, 11(1), 11. - Liu, G., Chimowitz, H., & Isbell, L. M. (2022). Affective influences on clinical reasoning and diagnosis: insights from social psychology and new research opportunities. Diagnosis, 9(3), 295-305. - Alsentzer, E., Li, M. M., Kobren, S. N., Noori, A., Undiagnosed Diseases Network, Kohane, I. S., & Zitnik, M. (2022). Few shot learning for phenotype-driven diagnosis of patients with rare genetic diseases. bioRxiv, 2022-12-07.