Analytics in Action News

Machine Learning System Accurately Identifies Medication Errors

A machine learning tool was able to identify medication errors better than traditional clinical decision support systems.

Machine learning accurately identifies medication errors

Source: Thinkstock

By Jessica Kent

- An alert system driven by machine learning could identify medication errors that traditional clinical decision support systems might otherwise miss, according to a study published in the Joint Commission Journal.

Prescription drug errors can lead to significant patient harm, resulting in high rates of mortality, morbidity, and increased healthcare costs, researchers noted. While clinical decision support (CDS) alerting tools are widely used to identify and reduce medication errors, the team pointed out that these tools have several limitations.

“Current CDS systems are rule-based and can thus identify only the medication errors that have been previously identified and programmed into their alerting logic. Further, most have high alerting rates with many false positives, resulting in alert fatigue,” researchers said.

Alert fatigue is one of the root causes of physician burnout, a phenomenon that can negatively affect patient satisfaction and lead to more errors in care.

The team set out to develop a CDS system driven by machine learning to help improve medication error identification. Researchers generated alerts retrospectively using outpatient data from two academic medical centers, and then compared these alerts to alerts in an existing CDS system.

Of the 10,668 alerts generated by the machine learning system, 68.2 percent would not have been generated by the traditional CDS system. Researchers also selected a random sample of 300 alerts for an in-depth review of the clinical records. Ninety-two percent of these 300 randomly selected alerts were accurate, and the team judged 79.7 percent to be clinically valid.

Additionally, researchers rated half of the 300 alerts assessed in the chart review as medium or high value.

“These data suggest that many of these errors had the potential for patient harm. This kind of approach can complement traditional rule-based decision support, because it is likely to find additional errors that would not be identified by usual rule-based approaches,” the team stated.

The machine learning tool could also have a substantial impact on healthcare spending. The average cost of an adverse event potentially prevented by an alert was $60.67. The prevention of ADEs could result in savings of $60.63 per alert, representing 99.93 percent of the total potential savings.

This extrapolates to a total of $1.3 million for a patient cohort with outpatient encounters in a two-year period when they are followed over five years.

The researchers said that their results are consistent with previous studies showing that systems combining clinical knowledge with machine learning monitoring and alerting can identify medication errors that might be missed by existing CDS systems.

Machine learning and artificial intelligence have emerged as potential solutions for curbing adverse patient events and boosting patient safety. Recently, MedStar Health and the Pennsylvania Patient Safety Authority partnered to use machine learning tools and generate more actionable insights about how to improve patient safety.

“There's huge promise for machine learning in healthcare,” Raj Ratwani, PhD, director of the MedStar Health National Center for Human Factors in Healthcare, told HealthITAnalytics.com in an interview.

“If clinicians use the technology on the front lines, it could lead to improved clinical decision support and new information at the point of care. But in order for that to happen, the insights generated through machine learning have to be presented to the clinician in a way that's intuitive. We need to provide information to the clinician that's going to effectively guide their decision-making.”

Researchers on the machine learning study echoed these words, noting that the potential of the technology hinges on providers’ ability to recognize and respond to alerts.

“Machine learning capabilities are increasingly applied in healthcare. Examples of machine learning applications in the health domain include drug discovery and development, diagnosis, disease and outcomes prognosis, and patient management. Machine learning programs may also enhance CDS systems and have the potential to improve the identification and prevention of medication errors and thus, patient safety,” the team said.

“The true value of such alerts is highly contingent on whether and how clinicians respond to such alerts and their potential to prevent actual patient harm.”

The study was limited in that cost analysis included only the direct healthcare costs of adverse events and associated pharmacist/prescriber interactions. Including other costs, like indirect medical and nonmedical costs, might result in different estimates of savings. The study also used retrospective data, so further research using real-time prospective data is necessary before the system can be deployed in clinical settings.

Despite these limitations, the results of the study show that machine learning-driven CDS tools can be a viable approach to improving medication error detection and preventing patient harm.