AI Powers Advancements in Sleep Research
The fusion of artificial intelligence with sleep science, as showcased at the Ken Kennedy Institute’s 2023 AI in Health Conference, heralds a transformative era in healthcare.
The conference kicked off its “Data Science of Sleep’’ sessions by converging experts from across the country and the Texas Medical Center, including Drs. Ashura Buckley, Justin Baker, Michelle Patriquin, and Mya Shiess. Their research and clinical experiences provided a comprehensive, data-driven deep-dive into the impact of sleep on disorder development, mental illness, brain behavior, and personal health.
According to keynote speaker Dr. Ashura Buckley, director of the Sleep Disorders and Neurodevelopment Consult Service at the National Institute of Mental Health, sleep health has a significant impact not only on the health of individuals, but also on societal and economic levels — sleep deprivation is estimated to cost $411 billion in lost productivity in the US alone. The National Institute of Health dedicated over $500 million to sleep research in 2022, demonstrating the public and scientific appetite for advances in sleep science. The pursuit of understanding sleep through the lens of AI is not merely academic but a response to an immediate call for public health innovation.
“The way to think about sleep is akin to exercise or diet,” said Dr. Buckley. “It is an essential, fundamental pillar of health.”
AI, we can dissect the enigma of sleep at an unprecedented granular level, parsing through the intricate web of biological signals to tailor personalized health interventions.
One key use case is in the interpretation of large, complex data sets, such as those generated from polysomnography (PSG) studies. PSG tests integrate a large swathe of distinct but related data sets, including recordings of brain activity, eye movement, respiration, cardiac activity, blood oxygenation, and muscle activity at various sites across the body. This combination of recording devices results in substantial datasets associated with each individual study.
PSG studies are typically performed to diagnose sleep disorders, such as obstructive sleep apnea (OSA). Once a diagnosis is made, however, the data generated from the tests is generally stored and forgotten.
However, the path is not without its challenges. The ethical stewardship of vast data pools generated from PSG studies and wearable devices is paramount. As we harness this technology to distill actionable insights from ‘data lakes’, we must also protect patient privacy and ensure data integrity.
“AI presents an opportunity to make use of all of this data that is just sitting on shelves all over the country,” said Dr. Buckley. “We’ve always collected more data than we know what to do with. Now, with the help of AI, we have exciting options to explore, so long as we know what questions to ask.”
However, the path is not without its challenges. The ethical stewardship of vast data pools generated from PSG studies and wearable devices is paramount. As we harness this technology to distill actionable insights from ‘data lakes’, we must also protect patient privacy and ensure data integrity.
Another avenue by which AI enables advancements in sleep medicine is through the expansion of single point acquisition devices. Rather than the multitude of sensors and equipment required by PSG studies, certain AI algorithms may be targeted to extrapolate such data from a single point of recording, such as an optical sensor placed on the finger. These recording devices may be minimally invasive and even wearable, akin to smart watches or wearable glucose monitors. When paired with activity data, such as that recorded by smartphone apps or other wearables, enormous possibilities for data generation are enabled while simultaneously reducing the financial and logistical strain placed on patients and providers.
Data from these kinds of wearable devices is a particular area of interest for keynote Dr. Justin Baker, Scientific Director of the Institute for Technology in Psychiatry at McLean Hospital. Dr. Baker presented results from a study in which 250 patients wore watch-like devices to continuously monitor their activity levels. This data was coupled with medication history and data from patient-doctor interactions to characterize patient behavior and guide treatment.
“Sleep is a behavior; our research is interested in using activity data to look at a range of behaviors,” said Dr. Baker. “When your activities of daily living, like teeth brushing, eating, or drinking change in measurable ways, sleep behavior changes as well.”
Remote monitoring data from such studies enables correlations to be made between sleep patterns and other aspects of patient health. In one patient with cyclical depressive episodes, activity data revealed that the patient woke up later in the day and was generally less active — a pattern which reversed when the episodes abated. This data led to what Dr. Baker calls the “battery model”.
“If we think of a patient who didn’t get enough sleep for the amount of activity that day, they didn’t charge their battery enough. Over a long period of time, a consequence of this behavior is that they enter into a more energy-conservative activity pattern [ie. depressive episodes] to recharge that battery.”
While data from all manner of daily activities were recorded during the study, Dr. Baker noted that sleep-related data was of particular interest due to growing clinical interest in sleep metrics.
As interest and excitement grows at the intersection of artificial intelligence and sleep science, experts speaking at the Ken Kennedy Institute’s AI in Health conference cautioned that care must be taken to apply these algorithms thoughtfully and with caution.
“For clinicians on the ground, we really try to think of solutions that are hypothesis-driven,” explained panelist Dr. Michelle Patriquin, who serves as the director of research at the Menninger Clinic. “We don’t want to go into the data thinking ‘I wonder what might be in here?’ and use unsupervised models. There may be space for that kind of work, but we try to think of concrete solutions with support from broader literature, with clear paths to tangible solutions.”
In the broader context, this technological leap could redefine the paradigm of patient engagement and disease management. Imagine a future where preemptive health strategies are crafted from the rhythmic patterns of our sleep, where ‘charging our batteries’ becomes a science-informed directive, not just a colloquialism.
This is where strategic collaborations with institutions like Rice University become pivotal. Partnerships bridging academia, industry, and healthcare entities can accelerate the translation of AI-driven sleep research from bench to bedside. Together, we can craft a holistic approach to health that places sleep at its core, leveraging AI to illuminate the unseen patterns that signal imbalance and disease.
The call to action is clear: we must thoughtfully integrate AI into sleep medicine, ensuring that every byte of data serves the patient’s health and well-being. As we continue to explore this frontier, it’s the symbiosis of human-centered care and machine intelligence that will guide us toward a future where every individual has the knowledge and means to achieve restorative sleep, and by extension, a healthier life.
Watch the recorded sessions from the 2023 AI in Health Conference on the Ken Kennedy Institute YouTube channel.
View the conference website at aihealthconference.com.
Save the date for the 2024 AI in Health Conference, taking place September 23–27 at Rice University’s BRC in Houston, TX.
