Using a Machine Learning System to Identify and Prevent Medication Prescribing Errors: A Clinical and Cost Analysis...

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Patient Safety

Background:

• Traditional clinical decision support (CDS) alerting tools are rule-based and limited to detecting pre-programmed medication errors.
• Machine learning (ML) offers a promising approach to enhance medication error detection and reduce adverse event costs.

Purpose of the Study:

• To evaluate the efficacy of a machine learning system (MedAware) in generating clinically valid alerts for medication errors.
• To estimate the potential cost savings from preventing adverse events using ML-based alerts.

Main Methods:

• Retrospective analysis of outpatient data from two academic medical centers (2009-2013).
• Comparison of MedAware alerts with alerts from an existing CDS system.
• Medical record review of 300 randomly selected MedAware alerts to assess accuracy and clinical validity.
• Estimation of potential adverse event outcomes, severity, and associated healthcare costs.

Main Results:

• MedAware generated 10,668 alerts, with 68.2% not identified by the existing CDS system.
• 92% of reviewed alerts were accurate, and 79.7% were clinically valid.
• Potential cost savings estimated at over $60 per drug alert and $1.3 million for the patient population.

Conclusions:

• Machine learning systems can identify clinically valid medication error alerts missed by conventional CDS systems.
• ML-based alerting demonstrates significant potential for cost reduction through the prevention of adverse drug events.