I recently evaluated one of my patients who had a stroke 2 years ago. After his stroke, we did an extensive workup, including prolonged cardiac monitoring, but could not determine what caused it. Then, a few months ago, his smart watch detected that he was in atrial fibrillation. Subsequent testing confirmed that diagnosis, and he was placed on anticoagulation to prevent additional strokes. Believe it or not, this patient’s diagnosis was made with the use of a neural network algorithm, a form of artificial intelligence (AI), that was embedded in his watch. This diagnosis significantly changed his management, and very well may have averted a large, disabling stroke. Despite the great impact this form of AI can have on patient outcomes, not all patients have access.
Stroke is the second leading cause of death worldwide and more than 795,000 people have strokes annually in the U.S., which averages to a stroke every 40 seconds. That means as many as 10 people will have had a stroke by the time you finish reading this article, and more than half of them may experience chronic disability. At the same time, the Association of American Medical Colleges estimates that in the next 11 years, we could have a nationwide shortfall of as many as 124,000 physicians. This shortfall is expected to disproportionately impact rural, low-income regions as well as communities that are predominantly Black or Hispanic. This could translate to more death and disability related to stroke in communities of color. AI presents an opportunity to better serve precisely those communities — and I’ve personally seen how it can happen.
AI enables data driven approaches to provide patients more personalized and precise care pertinent to their individual situation. Machine learning (ML), a subset of AI, can analyze a large volume of data and identify possible factors contributing to health inequities. For example, if a hospital notes that certain demographics are less likely to receive certain interventions, AI can be used to identify non-obvious factors — such as zip code, physical activity level, frequency of doctor’s visits, and distance between home and the nearest first responder stations — that may be stroke contributors or affect timely access to care.
Strokes disproportionately leave Black patients with severe disability. Data demonstrate that Black patients are less likely to get specialized treatment that could be used to treat or reverse their stroke, even when receiving care at a comparable facility to their white counterparts. This includes access to clot-busting medication as well as mechanical thrombectomy, a procedure used to extract the stroke from the occluded vessel. Black patients are more likely to be treated in hospitals lacking advanced technology and specialists who can provide more sophisticated emergency treatment. AI can assist these underserved hospitals in detection, triage, and outcome prediction after a stroke. Proper triage may result in the patient being transferred to an appropriate facility, where they can receive appropriate, timely treatment by a specialist.
When someone has a stroke, we want to act fast so that we can preserve function and potentially reverse their deficits. In addition to predicting an individual’s risk of stroke, AI can be utilized to rapidly detect a stroke as well as large vessel occlusion. After patients arrive in the emergency room, whether they are candidates for various therapies depends on the results of their initial CT scan of the brain. Incorporation of AI into stroke protocols has resulted in more rapid decision making, more rapid treatment of acute stroke, and more rapid decision to transfer patients for definitive care when transfer is needed. This technology can also send the images directly to all members of the stroke team so they have the opportunity to review the films and make a determination of whether or not the patient requires emergency treatment.
Looking forward, AI may also be used to determine the last time a patient was known to be well — information needed to decide whether a patient can get certain time-sensitive treatments. This would be particularly useful for patients who live alone, who often do not have someone who can reliably provide this critical piece of information.
To be clear, in these instances, AI is not replacing the medical provider, nor is the AI making a clinical decision. AI is allowing the medical provider to make a more rapid diagnosis so the patient can receive timely, definitive treatment. As in the case of my patient, AI provided information that the medical team did not have. Unfortunately, as helpful as this may be, not all patients have the financial means to purchase a “smart” watch. This adds to my concern that access to high quality AI may become a new predictor of health outcomes, which will disproportionately impact underserved communities. These communities are historically the last to gain access to or benefit from new and emerging technologies. Even when they gain access, they may be relegated to bad or dangerous AI technologies instead of being seen by an actual medical provider.
Furthermore, even if these underserved communities have access to this technology, there may be biased data due to algorithms anchored in systems of oppression. This is why it is important to ensure AI models are trained on diverse communities, that members from underserved communities are involved in creating the algorithms, and that there is oversight promoting equitable algorithms. Additional challenges include validating the models, preserving patient privacy, and securing data properly. By ensuring diverse representatives are involved in creating and monitoring the AI models, AI can be better poised to assist in narrowing the gap in access to care faced by Black and Brown communities. Done wrong, there’s a great risk of AI exacerbating existing inequities — but done right, AI can be integrated into healthcare in a productive and just manner.
AI is certainly not a panacea when it comes to eliminating health disparities, however given the significant challenges faced by underserved communities, we absolutely must not miss the opportunity to leverage this technology to improve the health of our most vulnerable communities.
Mill Etienne, MD, MPH, is associate professor of neurology and medicine at New York Medical College, where he is the vice chancellor for Diversity and Inclusion. He is chief of neurology at Good Samaritan Hospital, a PD Soros Fellow, and a Public Voices Fellow of The OpEd Project.
Please enable JavaScript to view the