Raise your hand if you’ve rushed to your primary care physician after diagnosing yourself via Dr. Google. Of course, your doctor didn’t rely on that diagnosis and instead probably asked a bunch of questions to begin your clinical assessment anew—there’s a reason why doctors spend so many years in training.
Symptom checkers, which have become quite popular among patients, collect more information before presenting a possible diagnosis when compared to a Google search. Their output, however, is still not suitable to serve doctors. That’s partly because they were intended for patients to self-diagnose themselves at home. They are built for consumers, not physicians
A 2021 study from the peer-reviewed scientific journal PLOS One concluded that the overall performance of symptom checkers resulted in “significantly below what would be accepted in any other medical field.” A correct diagnosis is displayed in the top five diagnoses in just 51 percent of cases.
Where symptom checkers fall short in serving physicians
Most symptom checkers today were originally built for patient use, and as such, they don’t provide physicians with the information they need to make the visit more efficient. The AI behind them simply doesn’t consult with medical sources of the same magnitude or employ the thought process on which human healthcare professionals base their clinical thinking and decision making.
Standard symptom checker tools are built using a combination of three different methods:
- Leveraging big data by comparing a patient’s symptoms with other patients’ data,
- Using static decision trees, where questions are pre-defined based on answers patients provide.
- Using “expert systems,” which base their algorithmic decisions on data contributed by physicians.
By relying on other patients’ medical records, pre-built decision trees, or individual contributions, these tools struggle to handle edge cases, explain their diagnostic process, and gather data the same way a physician would.
Existing symptom checkers essentially function to guess what the diagnosis might be. That’s not relevant for doctors, who are more interested in a viable tool that can perform clinical interviews that include all the relevant questions to help them optimize time with patients. The missing ingredient responsible for making today’s symptom checkers used in medical settings is their ability to replicate the way a physician actually thinks and performs a clinical assessment.
And how do physicians actually think? For one thing, they go through many years of education that expose them to medical knowledge through peer-reviewed articles, expert opinions, and clinical trials. They are trained to interview patients based on a working diagnosis list they build given the patient’s symptoms.
The knowledge together with the clinical know-how provides them with the foundation they need to come up with a differential diagnosis, refer to the right tests, and provide professional care. Without factoring in the vast amounts of medical knowledge available and really replicating clinical thinking, symptom checkers designed to guess a diagnosis can’t generate the right questions ensuring pertinent findings are uncovered. They simply aren’t built to rule out the rare presentations of common diseases or the common presentation of rare diseases.
In order to be used by the medical community and help address current healthcare challenges—overuse of medical services and lack of accessible care in isolated regions—these tools must base their decisions on the same knowledge physicians use and reason the same way they do. This includes building a differential diagnosis list, ruling out and ruling in diseases, and relying on originating sources from the literature.
Designing a tool that mimics clinical reasoning and that is based on referenced and peer-reviewed medical knowledge, instead of the other methodologies, eliminates the guessing game and allows for more relevant, dynamically tailored questions to be asked to the patient.
This would result in more thorough, accurate, and medically-referenced summaries and differential diagnoses. With this type of approach, a new category of tools will be able to produce drastically better quality, gaining the trust of healthcare providers.
Since 2007, primary care physicians have faced an annual increase in the workload of 2.5 percent. In order to assist physicians and the greater healthcare industry in overcoming this workload, new ways to gather data from patients must be adopted.
With this new category of tools that aims to eliminate the guessing game through leveraging pre-existing medical knowledge, physicians could have a tool that helps them streamline the patient diagnosis process. By introducing methods that provide referenced and trusted medical knowledge, these new tools will be able to perform clinical assessments to the benefit of the doctor, patient, and the entire healthcare system at large.
Written By Eitan Ron, Co-Founder and CEO of Kahun Medical