Machine learning-based clinical decision support can be an effective way to review the accuracy of medical prescription orders, a study published this week in the Journal of the American Medical Informatics Association shows.

According to the Paris-based research team, a CDS tool trained on data from 10,716 patients was more accurate than existing techniques at intercepting potential prescription errors.

The technology, developed by Lumio Medical, is a hybrid AI decision-support software that combines machine learning and a rule-based expert system. It makes predictions at the patient level rather than focusing on individual prescription orders.

"Our findings confirm that the algorithm outperformed classic systems in its capacity to limit the number of false alerts without overlooking patients with prescription order errors," wrote the research team.

WHY IT MATTERS

Medical errors are the third-leading cause of death in the United States, according to a 2016 study by a team at the Johns Hopkins School of Medicine. 

Reducing medication errors is a crucial way to address that issue, wrote the team in the JAMIA study. Although computerized prescriber order-entry and clinical decision-support systems have been developed as ways to improve the prescription process, the Paris researchers argued that CPOE and CDS can lead to their own issues – namely, additional errors and alert fatigue.

Meanwhile, clinical pharmacist medication review has been recognized as a critical step in the prevention of adverse drug events, but it is time-consuming and not always reproducible.

Researchers trained a binary classifier to identify patients who were likely to have at least one drug-related error in their prescription order. Some 133,179 prescription orders, along with each patient's individual information, composed the dataset used for model development. 

In an independent validation dataset, the Lumio Medication hybrid algorithm intercepted 74% of prescription orders requiring pharmacist intervention. Of the remaining 26%, none were life threatening, researchers said. 

But the team noted the study has limitations. It was only conducted in a single hospital setting and did not include ICU or neonatology patients, because they were managed by a different medical software.

Still, researchers said, "Our hybrid decision support system combining machine learning with a rule-based expert system was notably more accurate at detecting medication errors compared with other tools described in the literature."

THE LARGER TREND

Last year, Frost & Sullivan predicted that clinical decision-support systems would eventually supplant electronic health record systems as the primary health IT interface point for clinicians. And such systems are increasingly leveraging AI and ML to improve prescription and diagnostic accuracy.

In a HIMSS20 Digital presentation earlier this year, Mayo Clinic Platform president Dr. John Halamka said that, although it's still relatively early days for AI and CDS, systems powered by machine learning hold immense potential.

"Imagine the power to an AI algorithm if you could make available every pathology slide that has ever been created in the history of the Mayo Clinic," said Halamka. "That's something we're certainly working on."

That said, experts caution that successful tools depend on high-quality data.

"The data sets have to be better," said journalist and researcher Paul Cerrato, who coauthored with Halamka the new HIMSS Book Series edition, Reinventing Clinical Decision Support: Data Analytics, Artificial Intelligence, and Diagnostic Reasoning. "They have to be more representative. Some of the better algorithms are using two or three different data sets." 

ON THE RECORD

"A hybrid machine learning-based decision support system has been developed to intercept prescription orders with a high risk of containing at least 1 medication error," wrote the researchers in the JAMIA study. "Given that it is based on machine learning- and rule-based alerts, this decision support system has the advantage of not overfitting errors, of decreasing alert fatigue, and also of addressing infrequent but nevertheless potentially critical errors."

Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: kjercich@himss.org
Healthcare IT News is a HIMSS Media publication.