Identifying Deviations from Usual Medical Care using a Statistical Approach.

Area of Science:

• Clinical Informatics
• Pharmacovigilance
• Machine Learning in Healthcare

Background:

• Automated clinical alerting systems can improve patient care by flagging unusual treatment decisions.
• Detecting deviations from standard medical care is crucial for enhancing patient safety and optimizing treatment protocols.

Purpose of the Study:

• To develop and evaluate a novel method for identifying deviations in medication administration within the intensive care unit (ICU).
• To assess the clinical utility and potential disruptiveness of an automated system for flagging statistically unusual treatment decisions.

Main Methods:

• Utilized logistic regression models trained on historical patient data to identify statistically unusual medication administration patterns.
• Applied predictive models to a dataset of 3000 patient cases, identifying 53 potential deviations across 6 medications.
• Conducted a physician-led evaluation of predicted deviations and non-deviations for clinical relevance and alert utility.

Main Results:

• The developed method successfully predicted deviations in medication administration.
• Intensive care physicians deemed the predicted deviations to often warrant alerts and to be clinically useful.
• The projected frequency of alerts was considered unlikely to be disruptive in a clinical setting.

Conclusions:

• The proposed method demonstrates potential for developing effective automated clinical alerting systems in the intensive care unit.
• Machine learning-based detection of medication administration deviations can provide valuable clinical decision support.
• The system offers a promising approach to enhance patient safety without causing significant workflow disruption.