This is a fun little article presenting data relating to the Human Diagnosis Project, an online medical platform in which medical students and physicians create and solve teaching cases. Cases can be created by anyone, and are “solved” by submitting a ranked differential diagnosis to the system. Approximately 14,000 users have created or solved 230,000 cases in the few years it has been operational.
The article here, generally, highlights the diagnostic accuracy of respondents for 1,572 cases with 10 or more solve attempts. In their analysis, diagnostic performance, as measured by the likelihood for including the correct diagnosis in their top three, increased as additional physicians were added to the mix – effectively from 60-70% diagnostic accuracy up to a ceiling of about 90% when the collective diagnoses from 9 physicians were pooled.
While there are obvious limitations to using this platform to fully evaluate diagnostic performance and pooled diagnostic performance, my other takeaway: regardless of the actual number, even with the combined intelligence of multiple clinicians, accuracy is never 100%. While the expectation of our patients (and medicolegal systems) is perfect performance, it is not reasonable to expect perfection.
“Comparative Accuracy of Diagnosis by Collective Intelligence
of Multiple Physicians vs Individual Physicians”
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2726709