Machine Learning May Bolster Opioid Stewardship Efforts

Source –

Opioid stewardship is most often a multidisciplinary effort, but at Lifespan health system in Rhode Island, an unusual team “member”—artificial intelligence—is being recognized for its contribution.

Employing physician education and machine learning led to significant reductions in prescribed opioid doses, decreases in benzodiazepine co-prescribing, and increased naloxone co-prescribing, as Ashley Rimay, PharmD, a controlled substance pharmacist at Lifespan, said during a virtual poster presentation at the ASHP 2020 Midyear Clinical Meeting and Exposition (Best Practice poster 1).

“The collaboration between the pharmacists, data scientists, physicians and leadership was essential to making this program a success,” she said.

Rimay and a team of hospital and pharmacy leaders, a pharmacy data scientist, a pharmacy informatics coordinator, a senior clinical pharmacist specialist, and the chief medical officer set out to develop an opioid stewardship program. They wanted to capitalize on the wealth of electronic health record (EHR) data at their disposal and did so by providing a machine-learning model with two years of EHR-based opioid prescribing data from their institution, which it has used to identify outlying prescribers.

Each day, the software generates a scatterplot image plotting out that day’s prescribing practices, positioning prescribers on the graph according to whether or not their opioid prescribing falls in line with appropriate opioid prescribing.

Once an outlier is identified, the controlled substance pharmacist selects 15 opioid prescriptions at random and audits them for compliance with controlled substance laws. The findings are then passed on to physician leadership for peer clinical evaluation, provider education as well as possible follow-up audits, the researcher said.

In addition to providing outlier physicians with information on appropriate opioid prescribing, Rimay and her team educated them on the consequences of inappropriate opioid prescribing and opioid diversion, presented them with license reprimands issued by Rhode Island’s Medical Board, and detailed the state’s controlled substance law on prescribing for acute and chronic pain.

Between January and December 2019, the machine-learning software identified 25 outlying prescribers, Rimay reported. Additionally, more than 240 attending physicians and 900 residents participated in the system-wide opioid education initiatives.

In 2019, after the opioid stewardship program was rolled out, the health system saw a 14.4% decrease in average morphine equivalent daily doses (MEDD) among those provided with opioid education, including both outliers and the general physician population. In contrast, average MEDD increased by 0.39% during the same period among a group of Lifespan physicians not provided with opioid prescribing education.

During the same time, among physicians who received opioid prescribing education, there was a 9.7% decrease in benzodiazepine co-prescribing and a 15.7% increase in naloxone co-prescribing—both of which are in line with practices recommended by the CDC for those receiving opioids for chronic pain. In contrast, there was only a 4.35% drop in benzodiazepine co-prescribing and a 5% increase in naloxone co-prescribing among those who did not receive education on appropriate opioid prescribing, Rimay said.

“Health systems should foster collaboration between pharmacists, data scientists, physicians and leadership to develop a controlled substance prescription stewardship program,” the researcher concluded.

Artificial Intelligence Universe