By Drs. Mattina Davenport, Scott Ryals, Trung Le and Vidya Krishnan, on behalf of the AASM Synthetic Intelligence in Sleep Drugs Committee
Synthetic intelligence (AI) has been more and more highlighted as a possible instrument to boost population-level sleep well being and tackle gaps in sleep care. There’s rising curiosity within the potential for AI to lift group consciousness about sleep well being, assist the prevention of sleep problems, enhance sleep surveillance and develop entry to sleep care. Nevertheless, the empirical proof supporting these prospects stays restricted. As AI applied sciences proceed to evolve, cautious consideration will likely be wanted to make sure equitable entry and decrease potential disparities in digital well being.
A historic digital divide
The coronavirus pandemic highlighted the significance of AI well being care options, telehealth platforms and entry to real-time information for self-monitoring. There have been rising expectations that these digitalized processes will assist scale back well being care prices, facilitate sufferers’ entry, enhance the standard of care, pave the way in which for precision medication to advertise higher diagnostics and personalised remedies, and scale back population-level well being disparities. Regardless of the earnestness of those hopes, COVID-19 additionally make clear the historic digital divide, which has widened throughout our most up-to-date digital shift. McAuley outlined the digital divide as “a societal division between those that have the means and functionality to make full use of digital know-how and people who lack these means for causes regarding earnings, schooling or age.” The Nationwide Institute on Minority Well being and Well being Disparities designates that the next populations expertise well being disparities: Minoritized racial/ethnic teams (American Indian or Alaska Native, Asian, Black or African American, Hispanic or Latino American, Center Jap or North African, Native Hawaiian or Pacific Islander), folks with decrease socioeconomic standing, underserved rural communities, sexual and gender minority teams and people with disabilities.
In sleep medication, whereas there’s important enthusiasm surrounding AI developments, funding companies and researchers have but to totally tackle key questions equivalent to, “What populations presently profit from our AI improvements?” and “What populations might face challenges in accessing these developments?” Given the prevailing digital divide, addressing these questions stays an necessary space of focus.
Making use of a framework for digital well being fairness in sleep medication
As AI implementation expands in sleep medication, you will need to account for each social determinants of well being and digital determinants of well being. Digital determinants of well being might operate independently as limitations and facilitators of the digital divide, interacting with different social determinants of well being to affect sleep well being and information disparities. The NIMHD Analysis Framework was lately expanded for digital well being fairness. This framework outlines how digital setting determinants affect particular person well being, household/organizational well being, group well being, and inhabitants well being degree disparities. Some notable digital determinants to think about in sleep medication embody know-how entry, implicit tech bias, well being care infrastructure, group tech norms, information requirements, and algorithmic bias. In sleep medication, extra work is required to formally assess and tackle how digital determinants of well being work together with social determinants of well being to form entry to AI improvements and scale back sleep well being information disparities throughout NIMHD-designated well being disparate populations.
Sleep information disparities and the manifestation of bias in AI
More moderen frameworks for AI and sleep medication are highlighting the significance of leveraging a number of sources of information, equivalent to omics, digital medical data, goal sleep assessments, environmental information, epigenetics, and extra sleep metadata (e.g., geospatial, insurance coverage claims). With these a number of information sources being leveraged and harmonized, the sphere of sleep medication has infinite prospects to check a broader vary of sleep problems, higher predict sufferers’ danger, perceive how sleep problems happen and progress, and determine improved methods to boost detection, screening and therapy. Nevertheless, not all populations have the sleep well being information that’s essential to equitably pursue this endeavor. In truth, some populations might have systematic variations within the amount and/or high quality of their sleep well being information. These sleep well being information disparities might trigger sure populations to be unable to learn from the AI discoveries or improvements rising in sleep medication. To keep away from manifesting biased AI (Desk 1), it’s changing into more and more clear that the sphere of sleep medication must prioritize information assortment in real-world settings to extend the representativeness of sleep well being information sources.
Desk 1 outlines the kind of biases that manifest from information era to AI implementation.
Remaining ideas
In abstract, there’s unbelievable potential for AI to learn all areas of sleep well being. These advantages, nonetheless, will solely be nearly as good because the algorithms (and information) used to develop that AI. Explicit care to handle and mitigate biases and acquire information from numerous populations is vital to stopping additional widening of the health-related fairness hole. We encourage the adopters of rising AI improvements in sleep medication to think about this as we transfer towards enhancing sleep well being for all.
