Using Text Mining Techniques to Identify Health Care Providers With Patient Safety Problems: Exploratory Study

Area of Science:

• Health Informatics
• Natural Language Processing
• Patient Safety

Background:

• Regulatory bodies analyze patient complaints to identify health care safety issues.
• Analyzing large volumes of complaints manually is challenging.
• Text mining offers a potential solution for identifying patient safety signals.

Purpose of the Study:

• To explore text mining for identifying patient safety problems from health care complaints.
• To automatically predict the severity of patient complaints.

Main Methods:

• Exploratory study on over 22,000 Dutch Health and Youth Care Inspectorate complaints.
• Utilized supervised machine learning for complaint severity prediction.
• Evaluated features including sentiment analysis and bag-of-words models.

Main Results:

• A bag-of-words text classification achieved 87%-93% accuracy in severity prediction.
• F1 scores for severe complaints ranged from 45%-51% due to class imbalance.
• Combined severity prediction with keyword analysis identified potential patient safety issues at specific providers.

Conclusions:

• Text mining effectively supports inspectorates in analyzing civilian complaints.
• Automated severity prediction and keyword analysis aid in detecting patient safety problems.
• This approach helps prioritize supervision by sorting complaints by severity per organization or sector.