Exploration of Story Arcs in Palliative Care Conversations Using Natural Language Processing

Palliative care is an approach to improving the quality of life for patients with a serious, most likely terminal, condition. Conversations between palliative care doctors and patients are essential to ensuring optimal care for patients with a life-threatening illness. Understanding what constitutes effective communication among clinicians, seriously ill patients, and family members is a current national priority, yet there exists very little empirical knowledge of what goes into a conversation of this type. Palliative care consultations are often referred to by palliative care professionals as narratives, as the conversation is guided to best fit patient needs. Using transcribed text conversations of 354 palliative care consultations, from 225 patients, we explore the storyline of a palliative care conversation by investigating trends in word usage over narrative time. In the analysis, we see a decrease in the usage of terms describing the illness itself as the conversation progresses, with treatment talk following symptom talk. Additionally, we investigate trends in the sentiment of a conversation, using crowd sourced sentiment rankings of the most common terms in the English language. We observe that the overall positivity of the sentiment increases over the course of the conversation, largely to a reduction in the use of negative terms describing the illness. Finally, we explore temporal reference overtime by looking at the variation of verb tense, and the usage of the temporal nouns yesterday, today, and tomorrow, suggesting that discussion of the past decreases throughout the conversation, while discussion about the future increases. These trends found provide valuable, clinically-relevant insight into the storyline of a palliative care conversation, helping professionals to better understand these critical discussions