A machine learning program can generate clinically legitimate alerts for medication errors that might be missed with existing clinical decision support (CDS) techniques, in response to a study released in the January issue of The Joint Commission Journal on Quality and Patient Safety.
Ronen Rozenblum, Ph.D., from Harvard Medical School in Boston, and associates examined the ability of a machine learning system (MedAware) to produce clinically valid alerts. They projected the associated cost savings with potentially prevented antagonistic events.
Alerts had been generated on outpatient data from two academic medical centers between 2009 and 2013. MedAware signals were in contrast with those in a present CDS system.
The researchers found that 10,668 alerts were produced. Utilizing the existing CDS system, 68.2% of MedAware signals wouldn’t have been generated.
Based on structured data available in the document, 92% of a random sample of 300 chart-reviewed alerts had been correct; 79.7% had been clinically valid.
Potentially prevented adverse events had an estimated cost of over $60 per drug alert in an outpatient setting and $1.3 million when extending conclusions to the entire patient population.
Becton Dickinson funded the study. Several authors revealed financial links to the medical device and medical technology markets.