47 Operationalising a friends and family test natural language processing pipeline

Abstract

<h3>Background</h3> To manage the workload of data entry, manual sentence splitting and redaction of patient feedback that comes in form of 2000 comments per month, the patient experience team have worked with multiple data teams at GOSH to automate the process through a pipeline that incorporates Artificial Intelligence techniques. <h3>Methodology</h3> Part of this workflow involves a routine utilising artificial intelligence techniques, that handles sentence splitting, automatic data redaction and prediction. It was implemented in Python (Python Software Foundation. Python Language Reference, version 3.8. Available at www.python.org). It runs as a scheduled task every midnight of each day, to process all feedback captured in the previous day. On our secure digital environment, patient feedback is pulled from a database and identifiable data are masked on doctor, nurse, patient names as well as phone numbers and email addresses. Using natural language processing techniques, comments are split into grammatical sentences for further understanding of what sentiment is expressed in each sentence. This also enables the tool to assign topics from a predefined set of categories in the NHS Patient Experience Framework Themes 2022. Sentences from the same comment sharing the same sentiment and theme are joined by preserving the order of their occurrence. We are working with the Patient Experience, Quality and Information Services teams to utilise this output for downstream operations like manual validation using an in-house built webapp and measuring and monitoring the overall patient experience performance through a dashboard across wards using the Qlik Sense platform. <h3>Conclusion</h3> This pipeline demonstrates a use case of deploying an analytics project to be used near real time, it also shows collaboration amongst various departments with GOSH. <h3>Acknowledgements for funding or support</h3> This work is supported by the Health Foundation.

Ähnliche Arbeiten