A machine learning program has successfully reduced polypharmacy, improved medication adherence and lowered costs in patients with behavioral health conditions, presenters said at AMCP Nexus 2023, in Orlando, Fla.
Behavioral health issues are a significant and growing burden on the healthcare system; a 2020 Milliman study of 2017 commercial healthcare claims from 21 million people found that while only 27% had a behavioral health condition, those patients accounted for 56.5% of total healthcare expenditures.
Polypharmacy is another challenge in these patients: 60% of adults with a behavioral health condition are on two or more psychotropic medications, the presenters said. Polypharmacy increases the risk for drug–drug interactions and adverse events, and contributes to high healthcare costs.
“A medication may be started and there’s a partial response, [but] instead of maximizing that medication or stopping it, another one just gets added to it, and another,” said Caroline Carney, MD, the chief medical officer for Magellan Health. “Or a person with similar conditions may be treated for both of those conditions separately.” For example, a single medication may effectively treat many components of both depression and anxiety. “But I commonly see an antidepressant started and then a medication for anxiety, and then something else for sleep on top of that,” she said.
Another factor is that patients frequently receive medications from multiple prescribers, such as a primary care doctor, specialist and clinician in the inpatient setting. “Nothing ever gets reconciled; I used to refer to that as the Ziploc bag phenomenon of someone walking into your office carrying that bag of medications that they have no idea what to do with,” Dr. Carney said.
To address these problems, Magellan Health partnered with medication management tech startup Arine to develop inforMED (previously Navigate Whole Health; bit.ly/3vCWqnV). This program uses artificial intelligence to identify targetable prescribers, then generates optimized care plans with treatment recommendations and patient education. New data from clinical outcomes using these interventions are fed back into the system to continuously improve it.
“It takes into account hundreds of parameters,” Dr. Carney explained. “We go beyond just telling providers to change a medication and instead provide them with recommendations explaining what the problem is, why it’s a problem and the evidence-based literature supporting the change.”
Identifying Prescribing Patterns
Machine learning algorithms identify which prescribers to target based on their longitudinal prescribing patterns and the number of patients in their panel with prescribing outliers, said Yoona Kim, PharmD, PhD, the co-founder and CEO of Arine. “We also look at the social picture,” she added. “Even when other social determinants of health data are not available, we utilize ZIP code information for what percentage of that area has low income, or is on Medicaid or doesn’t have vehicle access, to get a good view of what the barriers to care access may be in those geographies.”
To date, Dr. Kim said, the program has reduced behavioral health polypharmacy by 45% to 55%, increased medication adherence by 20%, lowered average daily morphine milligram equivalents by 20% and saved $360 to $840 in pharmaceutical costs per enrolled member per year.
“The program is effective because we give providers data, and then we walk them through what to do with it, how to take care of that patient and what the next best alternatives are,” Dr. Carney said. “That kind of support really lends itself to better outcomes and to longer, better relationships over time.”
Dr. Carney reported no relevant financial disclosures. Dr. Kim reported that she owns stock in Gilead.
This article is from the April 2024 print issue.