Generative AI in healthcare Outperformed Professionals in diagnosis

In a recent study, GPT-4, an AI program developed by OpenAI, showcased superior diagnostic accuracy compared to clinicians, particularly in diagnosing elderly patients with extensive medical backgrounds and prolonged hospital stays. Research conducted by Yat-Fung Shea, MBBS, and co-authors from Queen Mary Hospital and the University of Hong Kong unveiled that the AI accurately diagnosed four out of six patients, while clinicians diagnosed only two out of the same six individuals.   

The AI's accuracy further improved to five out of six patient diagnoses when differential diagnoses were taken into account, surpassing clinicians who achieved three out of six accurate diagnoses. The study, published in JAMA Network Open, also utilized the medical diagnostic decision tool Isabel DDx Companion, which accurately diagnosed none of the patients initially and only two out of six with differential diagnosis information.   

According to Shea, the GPT-4 AI tool could potentially assist doctors when encountering elderly patients with prolonged hospital stays and inconclusive diagnoses. This AI technology proved adept at identifying missed diagnoses and alerting clinicians to potential underlying malignancies or drug side effects.   

The success of GPT-4 in diagnosing patients stemmed from comprehensive medical histories, including radiological and pharmacological information, for each case. This AI tool could help clinicians overcome challenges in diagnosing patients with complex conditions, particularly in scenarios involving multiple comorbidities.   

While GPT-4 excelled in diagnosing conditions like polymyalgia rheumatica, mycobacterium tuberculosis-related hemophagocytic lymphohistiocytosis, metronidazole-induced encephalopathy, and lymphoma, it encountered limitations in diagnosing multifocal infections. Nevertheless, the AI serves as a valuable tool to supplement clinician diagnoses and could aid in areas where specialized expertise is limited, especially in regions with fewer available specialists.   

It's crucial to note that the study's limited sample size included the medical histories of six patients at Queen Mary Hospital's Division of Geriatrics. Shea highlighted the potential of GPT-4 to enhance clinician responses, suggesting that this AI can bring overlooked aspects of medical history to clinicians' attention. However, the authors cautioned that GPT-4 may regurgitate incorrect information based on inaccurate medical histories.   

This study emphasizes the promise of using generative AI, such as GPT-4, in diagnosis when comprehensive medical histories are accessible. It presents an opportunity for improved diagnostic accuracy but also underscores new challenges for clinicians.   

-------------------------------------------

Explore the cutting-edge advancement in generating automatic medical reports, providing smart patient-record search and summarization with the latest feature of an innovative GPT-4 for healthcare – DrAid™ for EMR Smart Analytics. This innovative functionality is a paradigm-shifting module automating Electronic Medical Records (EMR) analysis that offers valuable insights to healthcare providers, enabling doctors to be assisted significantly in treatment recommendations, clinical decision support, and real-time alerts for timely interventions. 

Learn more about how DrAid™ for EMR Smart Analytics transforms the way doctors work!