The emerging patient-AI relationship is largely a result of the rising adoption of wearable devices and the improved sensor capabilities of smart devices (mobile phones, watches, fitness bands, etc.). More than 300,000 healthcare applications are available in app stores.13
Some apps use AI/ML technology and are available without the need for a prescription from a licensed healthcare provider. This direct-to-consumer marketing of healthcare technology is a growing industry that will likely impact the patient-clinician relationship.14
Some early adopters of wearables and healthcare applications have been healthy individuals with an interest in quantifying their physiological signals (referred to as the “quantified self”). But more recent consumer-oriented wearables are targeting use cases related to medical diagnosis. For example, Apple has received FDA approval for wearable (Apple Watch) AI-based technology that notifies users of the presence of atrial fibrillation and how frequently they are in atrial fibrillation.15,16
These types of technologies not only offer the opportunity to monitor patients and gather data outside of scheduled office visits (home, work, etc.) but also may permit early detection and treatment to avoid negative patient outcomes (e.g., thromboembolic stroke). This out-of-office monitoring could provide reassurance to some patients but may cause unnecessary anxiety for others. Each individual patient responds in a way that will lead to more or less concern and subsequent contact with the health system related to AI output.
While technologies such as the Apple Watch can provide accurate diagnoses and useful data (e.g., burden of atrial fibrillation), they have drawbacks. In many cases, they do not individualize a patient’s risk, consider patient-specific contextual factors, or address patients’ specific concerns about diagnosis and treatment.
For example, atrial fibrillation can be asymptomatic or symptomatic, can be chronic or paroxysmal, and can occur in the presence or absence of valvular heart disease. For a young male patient with no other risk factors, nonvalvular atrial fibrillation confers a 0.2 percent annual risk of stroke.17,18 For an older female patient with all possible risk factors, the annual risk of stroke is more than 10 percent in nonvalvular atrial fibrillation.
Although two such patients may be given the same advice of “talk to your healthcare provider” in response to AI-suspected atrial fibrillation, the urgency of this advice differs depending on the medical context. These examples also highlight the potential for patients to experience unwarranted reassurance or alarm due to AI output. More specifically, it allows an appreciation of the role of competent clinicians in interpreting AI output with consideration of patients’ full context.
The patient-AI dyad offers benefits, but gaps remain that for the time being are most effectively addressed by a clinician capable of serving as a liaison in the patient-AI relationship.19
