Is there a Doctor in the Crowd ?

Medical diagnosis is a central component of physician competence, with mastery as a major objective of medical education (Elstein, 1978; Norman, 2005). It is widely recognized that diagnosing is highly individual, dependent on physicians’ perceived difficulty of a particular case and their relevant domain knowledge (Elstein, 2002). As a result, health consumers in practice often choose to seek additional opinions from other providers. Health 2.0 (or participatory health) is a growing movement toward use of Internet technologies for personalized health-care (Swan, 2012). Specialized social networks for consumer health have become particularly popular, allowing people to ask questions about their health problems and receive feedback from others with similar symptoms or medical specialists. As a result, the pool from which consumers can seek “second opinions” has been drastically expanded. At present, 35% of adults in the U.S.A. have gone online to self-diagnose a medical condition for themselves or for someone else. Among those seeking online diagnoses, 35% did not visit a clinician to get professional opinions, 41% received confirmations from their doctors, and 19% reported disagreements from or inconclusive conversations with their doctors (Fox & Duggan, 2013). With the advent of crowdsourcing and human computation (Quinn & Bederson, 2011), a landscape of new and innovative approaches to healthcare are emerging to tackle open problems via crowd-based Internet services. For example, community question answering (CQA), social networks, and specialized medical sites, such as PatientsLikeMe, offer new ways to poll the social “village” or friendsource one’s social network (Horowitz & Kamvar, 2010; Oeldorf-Hirsch, Hecht, Morris, Teevan, & Gergle, 2014; Rzeszotarski & Morris, 2014). Prediction markets (Quinn & Bederson, 2011) such as the Iowa Electronic Markets aggregate crowd knowledge to make collective predictions. A study on use of Facebook to diagnose illnesses, the correct diagnosis was found for 5 of 6 cases within 10 minutes (Folkestad, Brodersen, Hallas, & Brabrand, 2011). A recent startup CrowdMed.com offers an interesting prediction market approach to medical diagnosis. People seeking a diagnosis may fill out a questionnaire detailing their symptoms, history and background for analysis by CrowdMed’s “Medical Detectives”, who suggest and bet on diagnoses. CrowdMed’s proprietary algorithm uses these bets to estimate a probability for each diagnosis. While we are not familiar with any clinical evaluations of CrowdMed’s effectiveness, the New Scientist reported that, “In 20 initial test cases, around 700 participants identified each of the mystery diseases as one of their top three suggestions” (Nuwer, 2013). CrowdMed reports finishing over 100 cases with < 10% of patients reporting inaccurate diagnosis. While such success is seemingly achieved through specialized, proprietary methods, is such complexity truly needed for accurate diagnosis, or might simpler methods suffice in some cases? In fact, due to the above Cheng, J., Manoharan, M., Zhang, Y., Lease, M. (2015). Is there a Doctor in the Crowd? Diagnosis Needed! (for less than $5). In iConference 2015 Proceedings.