Leveraging AI to Drive Timely Improvements in Patient Experience Feedback: Algorithm Validation.

Area of Science:

• Natural Language Processing
• Machine Learning
• Health Informatics

Background:

• Manual analysis of Friends and Family Test (FFT) data is challenging due to increasing volumes.
• Automated text analysis is crucial for real-time sentiment and theme prediction in patient feedback.

Purpose of the Study:

• To test and refine a supervised algorithm for analyzing patient feedback across diverse healthcare settings.
• To improve the cross-contextual performance of text analytics in varied care environments.

Main Methods:

• A supervised text analytics algorithm, initially developed in London, was tested across 9 diverse English healthcare organizations.
• The algorithm was iteratively trained using bag-of-words from anonymized FFT data, with manual coding for validation.
• A deployment framework and pipeline were developed for standardized implementation.

Main Results:

• The algorithm achieved over 75% accuracy in predicting themes and sentiments across various settings.
• Initial lower accuracy in pediatrics and mental health improved significantly with trust-specific coding templates.
• Thematic saturation was reached after five organizations, indicating efficient learning.

Conclusions:

• The developed supervised learning algorithm effectively leverages free-text FFT data for insights in diverse healthcare settings.
• Addressing contextual variations through further coding enhanced algorithm accuracy and reliability.
• The approach promotes collaboration and shared learning in analyzing patient feedback.