Researchers from the University of Vermont Conversation Lab used machine learning and natural language processing to understand what a conversation about the treatment of life-threatening medical illnesses looks like. This research could eventually help healthcare providers improve the end-of-life conversations they have with their patients.
Robert Gramling is the Holly and Bob Miller Chair in Palliative Medicine at the University of Vermont Larner College of Medicine. In a new paper, Gramling and his colleagues show how machine learning and natural language processing can be used to better understand what end-of-life conversations look like (Source: Andy Duback)
The team wanted to understand the types of conversations people have when talking about serious illnesses. They wanted to identify common features and storylines in various types of conversations. During the study, researchers borrowed some techniques from a previous study on fiction writing. Their machine learning algorithm was previously used to analyze language in fiction manuscripts to identify different kinds of stories.
The algorithm was adapted to analyze 354 transcripts of palliative care conversations. The conversations were gathered by the Palliative Care Communication Research Institute and broken down into 10 parts with an equal number of words. The 231 patients in the study were from New York and California. The algorithm examined the conversations for the frequency of words that refer to time, illness terminology, sentiment, possibility and desirability between the groups.
The results of the study showed that conversations progressed from symptoms at the beginning to treatment options in the middle and prognosis by the end. Participant use of modal verbs increased as the conversation continued. By the end, the conversations had turned into more of an evaluation rather than a description of the ailment. People tended to make meaning out of stories, so using narrative in medicine is important to a patient’s understanding.
The researchers focused on using their tool to identify different types of conversations that happen in healthcare. Eventually, the team hopes their tool can help healthcare practitioners understand what makes a good conversation and how types of conversations may require different responses. Ultimately they will create interventions that will match what the conversation status indicates what the patient needs.
A paper on this study was published in the journal Patient Education and Counselling.
