Ashley Rimay, Pharm.D., BCPS; Louis Palmisciano, BIT; Christine Collins, R.Ph., MBA
Lifespan/Rhode Island Hospital
Providence, Rhode Island
Detecting diversion and overuse of opioids is a major problem that few evidence-based tools address. Faced with the burden of compiling vast amounts of electronic health record data to detect potential diversion and overuse of opioids, a data scientist from the pharmacy department explored artificial intelligence and identified machine learning as a solution. This tool helped shape the controlled substance (CS) prescription stewardship program.
The CS prescription stewardship program is a multidisciplinary initiative designed to educate providers of the health system about regulatory requirements involved with CS prescribing, as well as create a tool to help identify inappropriate prescribing. This new tool uses a machine learning algorithm to predict the likelihood of prescribing an opioid prescription and compares this with the presence of an opioid prescription on any patient encounter. Providers writing opioid prescriptions that were not predicted are identified as “outliers.” Once these outlying providers are identified, an in-depth regulatory audit is completed by the controlled substance pharmacist (CSP) and distributed to the chief medical officer (CMO) of the affiliate. The audit is then further clinically reviewed by a physician peer evaluator selected by the chief of service. The final step of the auditing process includes targeted education to the provider and follow-up review of prescribing practices as necessary. This education supplements the system-wide education that the CSP provides at department level forums, such as grand rounds.
Over 50 in-depth provider audits of approximately 750 prescriptions were completed by the CSP. While no overtly inappropriate prescribing was detected, the algorithm enabled the CSP to create targeted education to ensure compliance with CS laws and decrease unnecessary opioid use. When examining department-wide trends, every department that completed targeted education, except for oncology/hematology, showed a decrease in average morphine equivalent daily dose (MEDD). System-wide increases in naloxone prescribing and decreases in opioid and benzodiazepine co-prescribing were noted. In addition to the targeted education, the CSP was able to provide regulatory education to more than 240 attending physicians and more than 300 residents and fellows.
The implementation of a CS prescription stewardship program based on machine learning was effective at reducing inappropriate opioid prescribing in a large academic health system based on metrics of decreased MEDD, increased naloxone prescribing, and decreased opioid and benzodiazepine co-prescribing.
