The implementation of a controlled substance prescription stewardship program based on education and machine learning was found to be effective at reducing inappropriate opioid prescribing in a large academic health system. According to a poster presented at the American Society of Health-System Pharmacists (ASHP) virtual 2020 ASHP Midyear Clinical Meeting and Exhibition, the impact of the program’s implementation was determined by metrics of decreased Morphine Equivalent Daily Doses (MEDD), increased Narcan prescribing, and decreased opioid and benzodiazepine co-prescribing.

According to the poster’s authors, there have been 350,000 opioid overdose deaths in the United States between 1999 and 2016, with a 200% increase in death rates since 2000. Further, provider diversion and a lack of accountability with controlled substances has led to multimillion dollar fines for health systems. Monitoring outpatient prescribing patterns has been a major challenge for the Lifespan health system in Rhode Island (RI), since it has over 1.5 million electronic prescriptions written annually, including over 360,000 controlled substances scheduled 2 through 4.

The objective of the study was to develop a controlled substance prescriptions stewardship program to monitor vast amounts of electronic health record data to detect potential diversion and over utilization of opioids. This leads to utilizing machine learning to identify outlying prescribers, auditing prescribers on outlying prescriptions, educating providers on controlled substance laws and guidelines, and improving prescribing practices based on metrics of decreased MEDD, benzodiazepine co-prescribing and naloxone co-prescribing.

The study utilized Lifespan’s The RI Hospital, The Miriam Hospital, Bradley Hospital, and Newport Hospital, which are all primary teaching facilities for the Warren Alpert Medical School of Brown University in Providence, RI. The study team was assembled with hospital and pharmacy leadership representing both inpatient and outpatient pharmacy, a pharmacy data scientist, a controlled substance pharmacist, a pharmacist informatics coordinator, a senior clinical pharmacist specialist, and physician chief medical officer.

A model was created using a trained, supervised XgBoost classification model. The results were grouped by the provider to visualize the entire organization for quick identification of uncommon prescribing practices. Information about the encounter was stored in a data warehouse along with the model’s prediction. Additionally, a web-based dashboard is refreshed daily as a scatterplot that aggregates patient-level predictions by the provider.

Implementing the auditing process involved:
  • Monitoring the dashboard for an outlying provider
  • Selecting 15 random prescriptions that were not predicted to be written by the model
  • Assessing for compliance to controlled substance laws
  • Communicating results to appropriate physician leadership
  • Physician peer clinical evaluation and follow-up audits

Provider education includes as follows:
  • Background on national incidences related to opioid prescribing
  • Implications of diversion for organizations and physicians
  • Data from the RI Medical Board on license reprimands related to opioid prescribing
  • RI Controlled Substance Law on prescribing acute versus chronic pain
  • Proper electronic prescribing in the electronic medical record

From January 1, 2019 to December 31, 2019, the metrics to determine prescribing improvements found reductions in MEDD, increased naloxone and opioid co-prescribing, and decreased co-prescribing of opioids and benzodiazepines. The results of audits and targeted education showed no overly inappropriate prescribing was detected.

The study authors concluded that health systems should foster collaboration between pharmacists, data scientists, physicians, and leadership to develop a controlled substance prescription stewardship program.


REFERENCE

Rimay A, Palmisciano L, Collins C. Establishing an opioid prescription stewardship program utilizing education and machine learning. Poster presented at: 2020 ASHP Midyear Clinical Meeting and Exhibition; virtual: December 6-10, 2020.