The researchers investigated five years of EHR-provided clinical data for 747,985 lives at the Brigham and Women’s Hospital and the Center for Patient Safety and Practice, plus retrospective data from Partners HealthCare BWH and Massachusetts General Hospital. The study was published in the Journal of American Medical Informatics Association. With algorithmic software from Israel-based startup MedAware, researchers screened the data to detect extremes suggestive of potential medication errors and compared with an existing clinical decision support screening system.
Using MedAware, researchers got alerts for over 15,000 charts, and they then looked at a sample of 300 charts, finding that 75 percent of alerts were valid in identifying potential medication errors. Unlike existing CDS systems, MedAware generates alerts based on screening for outliers.
Of the valid alerts, 75 percent of them were for potentially life-threatening prescription errors, giving researchers a validation of MedAware’s probabilistic, machine-learning approach. “There is a need for innovative solutions to address the current prescription errors that result in substantial morbidity, mortality, and wasteful healthcare cost,” so stated Dr. Ronen Rozenblum, a professor at Harvard Medical and a co-investigator on the study. “MedAware’s medication error detection system appears to have the ability to generate novel alerts that might otherwise be missed with existing clinical decision support systems.”
Fake alerts
The researchers stated that existing CDS screening systems can only detect a small number of actual errors, because they are not patient-specific or sufficiently adaptable. Besides that, they often result in high false alarm rates. Since doctors get so many fake alerts, they begin disregarding them altogether, and the result is more than $20 billion in healthcare losses every year in the United States.Using MedAware, researchers got alerts for over 15,000 charts, and they then looked at a sample of 300 charts, finding that 75 percent of alerts were valid in identifying potential medication errors. Unlike existing CDS systems, MedAware generates alerts based on screening for outliers.
Need for innovative solutions
“It has been hard to find medication errors which come completely out of the blue -- like a medication used only in pregnant women which is ordered for an elderly male -- but this approach detects orders which appear to be anomalous in some way, and it represents a very exciting new way to pick these errors up before they get to the patient,” so said Dr. David Bates, a professor at Harvard and expert on patient safety.Of the valid alerts, 75 percent of them were for potentially life-threatening prescription errors, giving researchers a validation of MedAware’s probabilistic, machine-learning approach. “There is a need for innovative solutions to address the current prescription errors that result in substantial morbidity, mortality, and wasteful healthcare cost,” so stated Dr. Ronen Rozenblum, a professor at Harvard Medical and a co-investigator on the study. “MedAware’s medication error detection system appears to have the ability to generate novel alerts that might otherwise be missed with existing clinical decision support systems.”
