More than a third of all adults in the United States use the Internet to self-diagnose everything from hangnails to heart attacks. Most start out by typing symptoms into an Internet search engine, which often leads to a "symptom checker" website or app.
A recent study by Harvard Medical School found that these symptom-checker sites and apps often provided inaccurate information for both triage and diagnosis.
Symptom checkers attempt to provide a potential diagnosis and direct patients to the most appropriate care setting. These symptom checkers ask users a series of questions about their symptoms, then use sophisticated computer algorithms to formulate a diagnosis or triage recommendation.
A number of private companies and organizations, such as the U.K.'s National Health Service and the American Academy of Pediatrics, have developed their own symptom checkers. One company, iTriage, reports 50 million users annually.
About Symptom Checkers
Symptom checkers are beneficial in that they can facilitate self-diagnosis and provide triage recommendation. They can also encourage patients suffering heart attacks, stroke or other life-threatening problems to seek emergency care.
This software can also reduce the number of patients showing up at hospitals with non-emergent problems not requiring immediate care. Reducing emergency department visits saves time and money for both patients and medical care institutions, and deters unnecessary use of antibiotics or prescription pain medications.
Lowered patient load relieves pressure on critical care providers. Symptom checkers also provide reassurance for those deciding whether to seek care.
The Study
In this groundbreaking audit, the researchers tested 23 online symptom checkers run by WebMD, Mayo Clinic, and the American Academy of Family Physicians. The researchers categorized the sites according to two main outcome measures — diagnosis and triage.
For sites that provided a diagnosis, the scientists wanted to find out whether the symptom checker listed the correct diagnosis first or within the first 20 potential answers. The main outcome goal for symptom checkers that provided triage recommendations was to determine whether the site made the correct recommendation for self-care, non-emergency care or emergency care.
To measure the diagnosis and triage advice, the researchers employed 45 standardized patient vignettes to audit the 23 symptom checkers. The vignettes reflected a range of conditions, ranging from the most common to least common conditions, and acuity varying from minor to life threatening.
The researchers searched for symptoms between June and November 2014 and selected sites and apps that were in English, free, publicly available, for humans rather than for veterinary use and that did not focus on a specific condition.
The Results
The symptom checkers listed the correct diagnosis first a mere 34 percent of the time on standardized patient evaluations. The sites named the correct diagnosis in the first 20 in 58 percent of cases and offered the appropriate triage advice in 57 percent of the cases.
Triage performance varied across symptom checkers and by condition urgency. Performance in triage recommendation ranged from 33 percent to 78 percent. The symptom-checker sites offered the appropriate triage advice in 80 percent of emergent cases, 55 percent of non-emergency cases but only 33 percent of self-care cases.
The Harvard audit demonstrates deficits in both diagnosis and triage in symptom checkers. These sites may not list the most likely diagnosis as the primary result, which can lead patients further from a correct diagnosis.
The triage advice provided by symptom checkers is typically risk-adverse in that it often suggests users to seek professional medical care when self-care would have been reasonable.
These websites and apps can supplement or someday replace the telephone triage lines common in today’s primary care, but only if they can improve accuracy of diagnosis and triage recommendations.
