DDC-Outlier: Preventing Medication Errors Using Unsupervised Learning.

Area of Science:

• Health Informatics
• Computational Medicine
• Data Science

Background:

• Electronic health records (EHRs) integrate patient data, enhancing hospital practices.
• Understanding EHR data is crucial for preventing life-threatening medical errors.
• Automatic detection of outlier prescriptions (dosage, frequency) remains an underexplored area.

Purpose of the Study:

• To propose an unsupervised method for automatically detecting outlier prescriptions.
• To address the gap in research concerning the identification of abnormal medication dosages and frequencies.
• To evaluate the proposed method's effectiveness in a real-world dataset.

Main Methods:

• Development of a novel unsupervised outlier detection algorithm: density-distance-centrality (DDC).
• Utilized a large dataset comprising 563,000 prescribed medications for evaluation.
• Compared the DDC method against various state-of-the-art outlier detection techniques.

Main Results:

• The DDC method demonstrated superior performance in detecting overdose and underdose prescriptions.
• Achieved better results compared to existing outlier detection methods on the medical prescription dataset.
• A significant portion of false positives identified by DDC corresponded to actual potential prescription errors.

Conclusions:

• The proposed DDC method is effective for identifying outlier prescriptions in EHR data.
• This approach can enhance patient safety by flagging potential medication errors.
• Further research can leverage DDC for improving prescription accuracy and reducing adverse drug events.