Virtually all healthcare executives (98%) agree that drug diversion occurs in hospitals, and nearly four in five believe that most drug diversion goes undetected, according to a new survey from Invistics, now part of Wolters Kluwer.
The health-system leaders who were most confident about their own drug diversion programs were those whose systems used machine learning/artificial intelligence (ML/AI) tools to detect diversion. More than half (53%) of the respondents using ML/AI tools reported that they are very confident in the efficacy of their diversion detection efforts, versus 23% of those who do not use such tools, according to the survey.
Since a similar Invistics survey in 2019, hospitals that report using machine learning to detect patterns of diversion and automatically flag potential cases have nearly doubled, from 29% to 56%. That increased uptake makes sense, given the ability of ML/AI tools to “do the hard work of sifting through mountains of data to find suspect cases of diversion so resource-strapped hospitals can run an effective program and ensure diversion is detected,” said Karen Kobelski, the vice president and general manager of clinical surveillance compliance and data solutions for Wolters Kluwer Health.
In a three-year study of 10 hospitals funded by the National Institutes of Health, Invistics reported that its machine learning model had 96.3% accuracy, 95.9% specificity and 96.6% sensitivity in detecting transactions involving a high risk for diversion (Am J Health Syst Pharm 2022;79[16]:1345-1354).
The study also underscored the speed of this detection technology. In subsequent testing using a much larger historical data set, the analytics detected known diversion cases in blinded data significantly faster than existing detection methods (a mean of 160 days and a median of 74 days faster; range, 7-579 days faster).
One participating health system, Atlanta-based Piedmont Healthcare, provided Invistics with data on nine known diversion incidents that occurred over 24 months. Using the machine learning model, the software detected all nine incidents, in all cases significantly faster (mean, 288 days), with detection time improved by a range of 25 to 579 days. The health system used machine learning to audit automated dispensing cabinet (ADC) and electronic health record data. The audits required 10 to 30 minutes, versus four to 20 hours of manual reconciliation using existing methods.
The software was developed based on patterns of known diversion behavior, Ms. Kobelski explained, including:
- healthcare workers dispensing or wasting more than their peers for a given drug or group of drugs;
- individuals with multiple canceled ADC transactions;
- documenting medication administrations at the end of the day instead of when they were administered; and
- removing medications when not on the clock.
“These patterns, along with data that could identify them—such as time sheets, waste amounts, administrations and cabinet dispensing timing—were all fed into the machine learning model and pulled in an overnight, near real-time basis,” she said. “Doing the investigation in this way, as compared with month-end reconciliation, also allows investigators to go to the floor and follow up on the machine learning findings when memories were fresher.”
Legacy Technology a Liability
Legacy technology for detecting drug diversion is not as nimble or complete as machine learning, Ms. Kobelski said. Such systems “often have built-in anomalous usage reports, but those typically do not use AI; they simply employ reconciliation. ‘This much was dispensed, this much was administered, here is a discrepancy,’ etc.,” she explained. “But there are so many other ways that people divert, including falsely documenting administrations so that there are no numerical discrepancies.”
Greg Burger, MS, RPh, a managing director at Visante, agreed that it’s easy for drug diversion to go undetected. To fill that oversight, one good place to start is to examine how controlled substances are handled by your inpatient and outpatient pharmacies. “When you look at the detailed steps most acute care inpatient hospitals take to secure their controlled substances, and then see all of the loose practices in that same hospital’s on-site ambulatory pharmacy, the difference can be pretty striking,” Mr. Burger said.
Liability also should drive anti-diversion efforts. The Drug Enforcement Administration has levied multimillion-dollar fines on health systems that have had major drug diversion incidents. Avoiding the financial and reputational hit that results from these events has brought on “a huge shift toward AI drug diversion solutions,” Ms. Kobelski said.
Ms. Kobelski and Mr. Burger reported no relevant disclosures beyond their stated employment.
This article is from the April 2024 print issue.