Prescription medication adherence is a problem. And an expensive one. The Centers for Disease Control and Prevention (CDC) finds that “the costs of nonadherence to prescribed medications are high and place significant financial strains on the health care system as a whole.” For example, the CDC cites a study showing that higher adherence to prescribed medications lowered healthcare costs among people taking medications for congestive heart failure by an estimated $7,800 per person annually. For individuals taking blood pressure medication, the cost is around $3,900 annually, and about $1,250 each year for those taking medication to manage high cholesterol.

Prescription medication adherence is especially a problem in underserved areas, and it becomes even more of a problem when a public health disease such as tuberculosis (TB) is involved. TB treatment often requires Directly Observed Therapy (DOT), in which healthcare professionals watch patients take their medications and monitor for responses to treatments. And TB hits underserved communities the hardest. CDC data shows that about 88% of the TB cases reported in America occur among ethnic and racial minority groups.

To reduce the number of healthcare professionals needed for in-person DOT, the CDC recommends telehealth technology with video-enabled devices that facilitate remote interactions between patients and healthcare workers. And AI may help reduce that burden on healthcare workers even further. A new study led by researchers at the University of Georgia shows that artificial intelligence evaluates patient-submitted videos to ensure medication adherence with accuracy.

The study looked at TB treatment in low-resource communities in Uganda, where a program dubbed DOT Selfie generates thousands of videos of people taking their medications. But instead of healthcare workers watching all the videos, the study tasked AI-enabled programs with reviewing the videos. The results? The best of the four AI models used identified patients taking their medication 85% of the time, a percentage comparable to humans performing the same task.

Though humans aren’t left out of the process entirely. University of Georgia researcher Juliet Sekandi says, “AI is really an accelerator of that process because then a nurse will not be worried that they have to watch all the 10,000 videos, but maybe watch only a few that need verification, say 100 out of 10,000.” While the study only considered AI-assisted DOT among TB patients, it’s easy to see how this technology could apply to a wide range of treatments that may benefit from observed medication adherence.