Accuracy of the WebMD Symptom Checker for Ophthalmic Diagnoses
By Lynda Seminara
Selected and Reviewed By: Neil M. Bressler, MD, and Deputy Editors
Journal Highlights
JAMA Ophthalmology, June 2019
Download PDF
As the accessibility of internet-based resources continues to grow, more patients are conducting self-guided research of symptoms, making it important to determine the accuracy of online symptom checkers. In a cross-sectional study using validated ophthalmic clinical vignettes, Shen et al. looked at the accuracy of the popular WebMD symptom checker. They found that the primary diagnosis generated by the symptom checker was correct for only 26% of the clinical scenarios. Moreover, the correct diagnosis was not on the list of possibilities for nearly half of the vignettes.
This cross-sectional descriptive study involved generating 42 validated clinical vignettes of ophthalmic symptoms and distilling them to their core presenting signs. The “cases” were entered into the WebMD symptom checker by two people who were masked to the diagnoses (one of whom was medically trained).
Output from the symptom checker was documented, including triage urgency and the list of diagnoses ranked from most to least likely. The main outcome was diagnostic accuracy of the symptom checker.
The mean number of symptoms entered for each case was 3.6 (range, 1-8). The median number of generated diagnoses per case was 26.8 (range, 1-99). The checker’s primary diagnosis was correct for only 11 (26%) of the 42 vignettes. The correct diagnosis was among the top three entries in 16 (38%) of cases. However, for 18 cases (43%), the correct diagnosis was not on the list.
The triage urgency for the top-listed diagnosis was appropriate in seven (39%) of 18 emergency cases and in 21 (88%) of 24 nonurgent cases. Interuser variability for the correct diagnosis being among the top three was at least moderate.
The clinical vignettes generally were devoid of comorbid and distractor symptoms, which may indicate that the checker’s accuracy was overestimated. However, the vignettes devised for this study resemble those used for training physicians in pattern recognition.
The authors emphasized that although the WebMD symptom checker may pinpoint an ophthalmic diagnosis, its overall accuracy is low. As a result, patients should exercise caution when using such online resources and should understand that symptom-checker output “is not a substitute for professional medical advice, diagnosis, or treatment,” as stated on the main page of the WebMD tool.
After this study was completed, the WebMD interface and output scheme were revised. (Also see related commentary by Rahul N. Khurana, MD, in the same issue.)
The original article can be found here.