Machine learning in healthcare: Software detects drug theft

Piedmont benefits from machine learning in healthcare

Before implementing the Invistics drug diversion software at Piedmont Athens, a 360-bed nonprofit hospital and regional referral center in northeast Georgia, Nix built Piedmont Athens' drug diversion program, which involved watching surveillance, monitoring all transactions and getting hands-on with education and awareness within the facility.

Russ Nix

Nix said one of the first things he noticed when he started Piedmont Athens' drug diversion program was the significant time gap between when a potential diversion occurred and when the incident was discovered.

Though Piedmont Athens was seeing good results as far as minimizing or mitigating risk within the institution, Nix said the organization wanted to make the window smaller between an incident and its discovery, which is when Invistics came into play. Nix said he receives a daily report from the Invistics drug diversion software system, which uses machine learning to improve its drug theft detection capabilities, and he can also access the program and use it as a full-time monitoring or surveillance program.

Nix said Piedmont Athens tested the software before implementing it by providing Invistics with data that contained blinded diversion cases to see if it would discover the same diversion incidents he found through his investigations.

Based on the scores and indicators Piedmont Athens put into the program, the Invistics drug diversion software had a 100% success rate in detecting proven cases of diversion. Additionally, Nix said the Invistics software detected each of the diversion cases much earlier than he had detected it.

"The way Invistics helped us improve is it basically takes the technologies that are already available, automatic dispensing cabinet, analytical software that goes along with that, whichever charting and prescribing software you have, and it helps all of those communicate together and give you a more real-time alert saying that there's a problem," Nix said.

Nix said the Invistics drug diversion software is user-friendly and convenient, as it compiles records and information from different systems and departments in the hospital into one place.

"Not only is it easy, there's a convenience to it where the way it works is you can literally just click from link to link to drill down further into an incident," Nix said. "It's astoundingly simple to use."

Details on using machine learning in healthcare investigations

Tom Knight, Invistics CEO, said the company took advantage of machine learning in healthcare to train its software to recognize patterns of behavior by healthcare workers consistent with known diversion cases.

Tom Knight

Knight said in the software's initial phase of development, the company identified diversion cases by healthcare workers that occurred in the past that were detected by other means, such as a tip from a co-worker. After collecting data entered by those healthcare workers into systems like the EHR and automated dispensing cabinet, Invistics trained its machine-learning-in-healthcare model to recognize those behavior patterns, he said.

"Each time a healthcare worker touched medication, the computer checks that data for dozens of potential 'alerts' that could be consistent with known diversion cases," Knight said.

Knight said Invistics then tested the machine learning model by scanning data entered by other healthcare workers and examining the employees the software suspected of theft. Invistics' early research involved the software scanning data containing blinded diversion cases and testing to see if the programming could detect the blinded cases.

This problem is a perfect problem in need of a technology solution. Invistics was the technology solution we felt was the best.

Now in its second phase, Knight said the Invistics software scans all data entered in hospital systems by all its healthcare workers. When the software detects suspected diversion by a healthcare worker, the hospital is asked to investigate that healthcare worker to determine if diversion did or did not occur, Knight said. After the investigation ends, Knight said the conclusion is fed back into the Invistics software machine learning model so it can learn from the new result.

"For example, if the computer was successful in detecting diversion, that 'true positive' result tells it to strengthen its confidence in future findings," Knight said. "If, however, the computer was not successful in detecting diversion, that 'false positive' result tells it to adjust its scanning parameters to avoid making a similar mistake in the future. In both cases, this additional data makes the machine learning more accurate, since it learns from both its successes and failures."

Invistics drug diversion software program grows

Based on the success the program has had so far, Piedmont Athens is looking to expand the drug diversion software throughout the entire system comprised of 11 hospitals, said Charles Peck, M.D., president and CEO at Piedmont Athens.

Prevention of medication theft offers a good avenue to test machine learning in healthcare, Peck said.

Charles Peck

"This problem is a perfect problem in need of a technology solution," he said of drug diversion. "Invistics was the technology solution we felt was the best."

The National Institutes of Health (NIH) recently awarded Invistics a supplementary grant to expand its drug diversion product. The company was originally awarded a NIH grant in 2017 to fund research and expansion of the drug diversion software into healthcare. Using NIH grant funding, the Invistics drug diversion software will be deployed in seven health systems covering three main departments: nursing, pharmacy and anesthesia.

Though the grant period involving research and expansion of the drug diversion software doesn't end until next year, Knight said Invistics has already launched the product, which has been available since April.

"It's really detecting diversion earlier -- and diversion that would've gone undetected," Knight said.