Detection of overdose and underdose prescriptions-An unsupervised machine learning approach

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Pharmacovigilance

Background:

• Prescription errors, including overdoses and underdoses, can lead to severe adverse drug events or reduced therapeutic efficacy.
• Accurate detection and prevention of these dosing errors are crucial for patient safety and effective treatment.

Purpose of the Study:

• To evaluate the efficacy of one-class support vector machine (OCSVM), an unsupervised machine learning algorithm, in identifying overdose and underdose prescriptions.
• To assess the performance of OCSVM models using clinical and synthetic prescription data.

Main Methods:

• Prescription data from Kyushu University Hospital (2014-2019) were analyzed.
• One-class support vector machine (OCSVM) models were developed for 21 drugs using age, weight, and dose as features.
• Clinical and synthetically generated overdose/underdose prescriptions were used to test model performance.

Main Results:

• OCSVM models successfully detected 87.1% of clinical overdose and underdose prescriptions.
• High precision, recall, and F-measure scores were achieved for detecting synthetic overdose and underdose prescriptions.
• OCSVM demonstrated superior performance compared to other unsupervised outlier detection algorithms.

Conclusions:

• OCSVM models utilizing age, weight, and dose are effective tools for detecting prescription errors.
• This machine learning approach shows significant potential for improving medication safety and preventing adverse drug events in clinical practice.