Key points are not available for this paper at this time.
Background: Shortage of Rheumatologists worldwide results in long delays in diagnosis and poor outcomes for patients. At the same time the world is seeing an unprecedented rise in Artificial intelligence language models which have been able to diagnose complex medical cases outperforming human readers 1. Objectives: To evaluate consecutive new referrals to a Rheumatology clinic both by Rheumatologists and by Proprietary Rule Engine and AI GPT4. Methods: Symptoms and available lab tests of consecutive new patients were fed into the engine developed by Algorithm Health Ltd (AHL) who have developed a proprietary medical rule engine that combines pattern recognition and algorithmic assessment of the parameters used for the classification. The rule engine has included symptoms and lab test results and each were given specific weightages linked to individual diseases. The model also looks at the input data to assess from the overall pattern of the parameters, the appropriate diagnosis which the data would best fit. The system is capable of handling large volumes of data near instantly and providing the required output. In addition the data was fed into Chat GPT4 AI models to generate a diagnosis. Patient data was anonymized and the data scientists were blinded to the Rheumatologists diagnosis. We also obtained diagnosis from a second Rheumatologist who was similarly blinded to initial diagnosis. Results: Results A total of 100 new patients presented to the clinic between November 20th 2023 to January 5, 2024. The AHL Rule Engine and AI Chat GPT 4 presented 3 differential diagnosis based on the symptoms and lab tests fed into the engine. The Rule engine and AI correlated with the first diagnosis of Rheumatologist 80 % of the time and in the top 2 differential diagnosis 17 percent of the time. The Rule engine and AI GPT4 missed the diagnosis in one cases (inflammatory mono arthritis of shoulder although an appropriate differential diagnosis was given). In a second case of knee bursitis the Rule engine was unable to diagnose but GPT4 was able to provide an appropriate differential. The Rule Engine also presented an inappropriate differential diagnosis of Addison's disease although the Primary diagnosis of Sarcoidosis was correct. In 1 case GPT4 missed diagnosis of Reactive arthritis although AHL Rule was able to pick it up. AHL engine appropriately prioritized Rheumatoid arthritis over other diagnosis in 2 cases which GPT4 could not. Conclusion: AI powered engines were 98% accurate in initial diagnosis or differential diagnosis of new cases presenting to a Rheumatology clinic. The Proprietary AHL rule engine has the potential to be more accurate than GPT4 as minor symptoms can be tailor made for subtle symptoms and complex scenarios. These technologies can be powerful tools for initial differential diagnosis and prompt triage for Rheumatology referrals. We see a near future were most preliminary assessments and early treatment can be commenced at a Primary care level while awaiting Rheumatology appointments. In addition, in countries where access to Specialists is through self referral it can be used as a self screening tool by patients. REFERENCES: 1 Use of GPT-4 to Diagnose Complex Clinical Cases Alexander V. Eriksen, M.D. et al November 9, 2023 NEJM AI 2023;1(1) DOI: 10.1056/AIp2300031 VOL. 1 NO. 1. PRIMARY DIAGNOSIS BY RHEUMATOLOGIST Acknowledgements: NIL. Disclosure of Interests: None declared.
Badsha et al. (Sat,) studied this question.