The evolution of symptom checkers

Despite skepticism, AI symptom checkers improve healthcare.

Patient symptom checking has been a common occurrence since the early days of the internet. Over time, this universal experience of “Googling” one’s symptoms has led to the development and improvement of clinically-validated symptom checking tools. Despite their broad acceptance and use by our patients, many of my physician colleagues remain hesitant toward their formal integration into the healthcare system.

Evolution of Symptom Checkers

To understand physicians’ perspectives on symptom checkers, we must first understand their evolution. The earliest automated symptom checkers used cumbersome and often inaccurate decision tree models that took excessive time asking irrelevant and redundant questions. Over time, as technology has advanced, so too have these symptom checking tools.

Over the last several years, the most competitive symptom checkers have begun to use some form of artificial intelligence to enhance their performance — i.e. make more accurate diagnoses and a more efficient user experience for patients and providers. As specific types of AI models like adversarial, convolutional, recurrent and transformer-based neural networks continue to progress, AI will increasingly play a role in enhancing the efficiency and accuracy of diagnostic technology.

Clinician Apprehension to Symptom Checkers

There are many reasons why clinicians are concerned about the use of symptom checkers. The first is that inadequate symptom checking tools may steer patients toward incorrect diagnoses, potentially over or underestimating the level of care they need. This can often result in friction and skepticism towards clinicians when their time comes to assess and diagnose.

In addition, some clinicians may take umbrage at the use of symptom checkers because they feel that they contribute to the lack of trust between clinicians and patients. For example, if a patient is not confident in their physician's diagnostic skills, they will use symptom checkers as a crutch to make up for this self-perceived deficiency. Worse yet, some clinicians fear that their widespread adoption is an attempt to use technology to replace them. Although some of these concerns are valid, no matter how advanced symptom checkers become, they will always lack the fundamentally human element required to deliver complex care.

Potential For Improving Healthcare

Despite skepticism, symptom checkers hold enormous value for various reasons. Symptom checkers can be an incredibly useful tool for physicians in assessing what patients are concerned about and what they are feeling. When physical evaluations are not immediately possible, symptom checkers provide an approximation of underlying conditions and deliver a high-quality estimate of acuity.

Additionally, as symptom checkers advance and become more interoperable with other aspects of the digital front door and greater healthcare IT ecosystem, they can play an important role in triaging and directing patients to the right care, faster, before their condition worsens. This limits the burden on primary care physicians and their staff, and improves the financial and operational efficiency of the entire healthcare system. For example, emergency rooms are often overburdened with patients experiencing less severe symptoms. With the use of symptom checkers, some of these patients can, instead, be guided to telemedicine or other less urgent care settings.

The Future of Symptom Checkers

Although AI-powered symptom checkers will never fully replace physicians, healthcare technologists will (and should) continue to integrate these tools into the patient journey. This, along with an increasing acceptance of digital healthcare, has resulted in a growing number of EHR, patient engagement, and data companies attempting to create comprehensive, accurate, and easy-to-use symptom checkers.

What separates “the best” symptom checkers from “the rest” are the AI being deployed, and the algorithms and training sets used to build their knowledge base. One of the most promising models within diagnostic AI is the use of Sophisticated Bayesian Networks, built to assess probabilities, and mimic physician thought processes in taking a patient’s medical history.

As medical school students often do when taking their first history on a real patient — using a great deal of time to ask every question they ever learned germane to the patient’s complaint — so does the Bayesian model. With more experience, or patient data in the case of symptom checkers, this process becomes increasingly streamlined. Over time, both learn what follow-up questions are the most valuable in winnowing down the list of potential and probable diagnoses. Like the clinicians themselves, the more clinical encounters these models have, the faster and more accurate they will be.

What this means for clinicians moving forward

Symptom checkers will never replace clinicians. The proper physician-patient relationship involves a level of empathy that symptom checkers cannot provide. The ideal role of symptom checkers will likely always be to provide support to physicians in streamlining information flow, minimizing errors, and maximizing efficiency. When done well, symptom checkers will seamlessly blend into the clinical workflow to make the point-of-care encounter as productive as possible, allowing the physician and patient to devote more time to the human experience so integral to healthcare.

Irving Kent Loh, MD FACC FAHA FCCP FACP, is a cardiologist, clinical researcher, and the Chief Medical Officer/Co-Founder of Infermedica.