Machine-Learning to Identify Moments of Connectional Silence in Serious Illness Conversations

In order to understand, disseminate, support and incentivize high quality communication in serious illness, we need measurement methods that can be scaled for larger clinical studies and routine health service quality reporting. This project develops a machine learning pipeline to develop such measures using existing cohort data that includes more than 12,000 minutes of audio-recorded palliative care conversations from the Palliative Care Communication Research Initiative. We focus initially on conversational pauses that are indicative of human connection in serious illness conversations. 

Funders:  Holly & Bob Miller Endowed Chair in Palliative Medicine, University of Vermont College of Engineering, University of Vermont Department of Family Medicine

Project Team: Viktoria Manukyan, Brigitte Durieux, Cailin Gramling, Lindsay Ross, Larry Clarfeld, Michelle Nyland, Aidan Ryan, Donna Rizzo, Maggie Eppstein, Stewart Alexander (Purdue University), Bob Gramling

Project Publications:

  • Manukyan V, Durieux B, Gramling C, Clarfeld L, Rizzo D, Eppstein M, Gramling R. Automated Detection of Conversational Pauses from Audio Recordings of Serious Illness Conversations in Natural Hospital Settings. Journal of Palliative Medicine. 2018. 21 (12).
  • Durieux B, Gramling C, Manukyan V, Clarfeld L, Ryan A, Ross L, Niland M, Rizzo D, Eppstein M, Alexander SC, Gramling R. Tandem Machine Learning-Human Coding Method for Identification of Connectional Silence in Serious Illness Conversations. Journal of Palliative Medicine. 2018. 21 (12).

 

 

Self Organizing Map

Natural Language Processing for Characterizing Palliative Care Conversations

Clinical conversations are information-dense sources of clinical data. This project leverages audio-recordings of more than 350 palliative care consultations from the Palliative Care Communication Research Initiative to develop automated Natural Language Processing methods for measuring fundamental features of palliative care conversations, beginning with expressions of uncertainty, prognosis, and emotion.

Funders: Holly & Bob Miller Endowed Chair in Palliative Medicine, University of Vermont College of Engineering, University of Vermont Department of Family Medicine

Project Team: Lindsay Ross, Laura Hirsch, Bennet Cazayoux, Larry Clarfeld, Bridger Banco, Brigitte Durieux, Cailin Gramling, Christopher Danforth, Donna Rizzo, Maggie Eppstein, Bob Gramling

Student Theses and Dissertations:

  • "Exploration of StoryArcs in Palliative Care Conversations using Natural Language Processing" Lindsay Ross (Honor's Thesis, University of Vermont Department of Computer Science)

brain.jpg