Though symptom checking apps and websites have been touted by some as the future of clinical diagnostics, a recent study has found that they are far less accurate than actual physicians at identifying the correct diagnosis when given a clinical vignette.
Though symptom checking apps and websites have been touted by some as the future of clinical diagnostics, a recent study has found that they are far less accurate than actual physicians at identifying the correct diagnosis when given a clinical vignette. The results of the study, published in a research letter in JAMA, compared the performance of physicians and online symptom checkers in diagnosing hypothetical clinical situations.
Lead author Ateev Mehrotra, MD, MPH, of Harvard Medical School, and colleagues had previously assessed the accuracy of 23 web- and app-based symptom checkers in a 2015 BMJ study, finding that the computer algorithms identified the correct diagnosis just 34% of the time. The 45 clinical vignettes used to test these symptom checkers provided the patient’s medical history and symptom presentation, but did not include physical examination or test findings.
In this study, the researchers distributed the same 45 clinical vignettes to physicians participating in Human Dx, an online platform where doctors generate differential diagnoses based on such vignettes. The doctors did not know which cases were part of this research study.
The 234 physicians who assessed the study vignettes determined the correct diagnosis first 72.1% of the time, which was more than double the accuracy rate of the symptom checkers. They were also significantly more likely (84.3% vs 51.2%) to have listed the correct condition in their top 3 diagnoses. Diagnosis accuracy was similar whether the physician was an attending, a fellow or resident, or an intern.
Accuracy also varied based on the acuity and prevalence of the conditions described in the vignettes. Physicians had an accuracy rate of 79.1% when diagnosing high-acuity situations for which urgent care would be required, compared to the 24.3% of symptom checkers that listed the correct diagnosis for these emergent conditions. In addition, 75.5% of physicians correctly identified an uncommon condition (defined as diagnoses that accounted for less than 0.3% of ambulatory visits in the US in 2009 to 2010) as opposed to 28.1% of symptom checking apps or websites.
The study authors acknowledged that although the symptom checkers had much lower accuracy rates, future research might examine whether these technologies could help support the diagnosis of a physician instead of replacing it.
What Will AI in Health Care Look Like in the Future?
July 10th 2025Artificial intelligence (AI) transforms oncology by enhancing clinical decision-making, improving patient outcomes, and shaping future health care training, as discussed by experts at Tennessee Oncology's “Tech Innovations in Community Oncology."
Read More
Laundromats as a New Frontier in Community Health, Medicaid Outreach
May 29th 2025Lindsey Leininger, PhD, and Allister Chang, MPA, highlight the potential of laundromats as accessible, community-based settings to support Medicaid outreach, foster trust, and connect families with essential health and social services.
Listen
AI Tool Shows Promise for Faster, More Accurate Autism and ADHD Diagnoses
July 8th 2025Participants performed touchscreen-based reaching tasks while wearing wireless motion sensors, which recorded linear acceleration, angular velocity, and roll-pitch-yaw orientation at millisecond resolution.
Read More
Inside the Center's MDD Value Model and Its Use of Dynamic Pricing
May 13th 2025Larragem Raines, MS, of the Center for Innovation & Value Research, discusses the organization's major depressive disorder (MDD) open-source value model, dynamic pricing, and the future role of artificial intelligence in care.
Listen
Driving Value via Outcomes-Based Pricing and EHR Interoperability: Tyler Sandahl, PharmD
July 7th 2025Tyler Sandahl, PharmD, a clinical pharmacist at Mayo Clinic, discussed the complexities of alternative payment models for chimeric antigen receptor T-cell and bispecific therapies and the need for improved data sharing in cancer care.
Read More
How AI Can Address the Troubled Business Model in Oncology
July 7th 2025Artificial intelligence (AI) will transform community oncology, addressing workforce shortages and reimbursement challenges while enhancing patient care and operational efficiency, discussed speakers at Tennessee Oncology's “Tech Innovations in Community Oncology."
Read More