Article Text

Download PDFPDF
186 Predicting patient perceptions in clinical interactions: an NLP approach to enhanced understanding of patient-doctor communication
  1. Lisa A Oh1,
  2. Weicheng Ma2,
  3. Paul J Barr1,
  4. Soroush Vosoughi2
  1. 1Center for Technology and Behavioral Health, Dartmouth College, Hanover, USA
  2. 2Department of Computer Science, Dartmouth College, Hanover, USA
  3. 3The Dartmouth Institute for Health Policy and Clinical Practice, Dartmouth College, Hanover, USA

Abstract

Introduction Effective patient-doctor communication has historically been challenging to measure and improve, largely relying on time-consuming and subjective patient feedback via surveys. The integration of Natural Language Processing (NLP) into such interactions has the potential to transform healthcare communication. Through NLP, we can uncover objective markers of effective communication and provide real-time, scalable quality assessments to clinicians, unlocking new possibilities for personalized healthcare.

Methods Thirteen primary care clinic visits with multimorbid White older adults (mean age: 76.2 years, 54% female) were recorded, transcribed, and paired with the Inter Personal Processes of Care, a patient reported survey of communication quality. We trained a Longformer model on this data to predict the responses to each item in the survey.

Results On certain questions such as ‘How often did doctors let you say what you thought was important?,’ our model achieved generalizable predictions with a validation loss below 0.007. However, it struggled to learn generalizable patterns for questions such as ‘How often were doctors concerned about your feelings?,’ suggesting a need for further data, model refinement, or additional information such as non-verbal cues to make more accurate predictions.

Discussion This research demonstrates the potential of NLP in predicting aspects of patient perception of care from the verbal content of clinic visits. However, further exploration is needed, such as training the models on larger, more diverse datasets and investigating multi-modal approaches to predicting complex constructs with non-verbal information.

Conclusion This research highlights how patient perception of care can be predicted and thus enhance patient-doctor communication. It lays the groundwork for developing tools that offer clinicians real- time, scalable feedback via the identification and modification of problem language to better accommodate for individual patient needs. This advancement promises significant improvements in healthcare communication, an important yet often overlooked aspect of patient care.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.