Strengthening the patient voice in health service evaluation: machine learning on free text comments from surveys and online sources

Summary

The government has ordered the patients’ health service, where the patient's voice must be heard, including strengthening the involvement of patients in decisions and the development and evaluation of health services. An important patient-oriented tool at the national level is the national system for measurement of patient experiences. Free-text comments from these surveys are considered highly relevant and actionable by clinicians and managers aiming to improve quality, but are mostly unused due to the time and resources needed to analyse patient comments. There is thus a clear need for an innovative and highly efficient method for analysing large amounts of patient comments. The field of Natural Language Processing (NLP), a branch of Data Science and Artificial Intelligence, is concerned with automated analysis of human language. One application of particular relevance is sentiment analysis. The task of sentiment analysis is to (i) identify subjective opinions and attitudes expressed in text, (ii) detect whether the opinion has a positive or negative polarity, and (iii) identify who/what are the targets and holders of the opinion. Moreover, in so-called aspect-based sentiment analysis the task is additionally to (iv) map the identified targets to more general topic categories. While machine-based sentiment analysis has been introduced as a way of analysing comments from patients in health services research, these tools are both domain- and language-specific and we presently lack tools to analyse Norwegian text in the medical domain.

This project is an innovation project in collaboration between UiO and NIPH and is funded by the Research Council of Norway. This project will develop and test resources and tools for aspect-based sentiment analysis of comments in Norwegian language. To develop the model, we will use patient comments from NIPS national surveys on patient experiences with general practitioners and general practice offices, as well as patient experiences with inpatient mental health care. The goal is for the model to be used to analyze patient comments from all patient experience surveys.