Identifying Patient-Reported Care Experiences in Free-Text Survey Comments: Topic Modeling Study

Area of Science:

• Health Services Research
• Natural Language Processing
• Machine Learning

Background:

• Patient-reported experience surveys (PRES) offer valuable patient feedback for healthcare improvement.
• Traditional analysis of free-text comments in PRES is resource-intensive and complex.
• Advances in machine learning (ML) and natural language processing (NLP) present new opportunities for analyzing this underutilized data.

Purpose of the Study:

• To apply NLP techniques for topic modeling of free-text comments from patient-reported experience surveys.
• To leverage ML for extracting meaningful insights from patient feedback.

Main Methods:

• Utilized Consumer Assessment of Healthcare Providers and Systems (CAHPS) survey data linked to inpatient records.
• Employed unsupervised topic modeling with automated labeling using BERTopic.
• Incorporated sentiment analysis to aid in topic interpretation.

Main Results:

• Over 43% of adult patients and 46% of pediatric caregivers provided free-text responses.
• Identified 86 topics in adult responses and 35 in pediatric responses, revealing aspects of care not covered by existing surveys.
• The majority of identified topics were generally positive.

Conclusions:

• BERTopic effectively identified interpretable topics in patient feedback with minimal tuning.
• Findings support the application of ML in understanding patient experiences for person-centered care, patient safety, and quality improvement.
• ML analysis of patient feedback can identify temporal and site-specific trends, highlighting areas for concern and improvement.