Screening Electronic Health Record-Related Patient Safety Reports Using Machine Learning.

Area of Science:

• Health Informatics
• Patient Safety Research
• Machine Learning Applications

Background:

• Patient safety reporting systems are crucial for identifying healthcare hazards.
• Electronic Health Records (EHRs) introduce new safety challenges that require effective monitoring.
• Manual screening of safety reports is time-consuming and may miss critical information.

Purpose of the Study:

• To develop a semiautomated method for screening patient safety reports.
• To identify hazards specifically related to Electronic Health Records (EHRs).
• To enhance the analysis of mandatory, population-based patient safety reporting system data.

Main Methods:

• A query identified potentially relevant cases from the Pennsylvania Patient Safety Reporting System.
• A machine learning model was developed using manually screened training, testing, and validation data sets.
• The developed model was applied to automate the screening of remaining cases.

Main Results:

• A naive Bayes kernel algorithm demonstrated superior performance among four tested algorithms.
• The best-performing model achieved an area under the receiver operating characteristic curve of 0.927.
• High accuracy (0.855) and F score (0.877) were recorded, indicating effective classification.

Conclusions:

• Machine learning and text mining offer valuable tools for patient safety.
• These methods can semiautomate the screening and analysis of unstructured text in safety reports.
• The approach facilitates the identification of EHR-related events, even in legacy data or when not explicitly categorized.