Temporal Reference: Automatically Identifiable References to the Future Are Associated with Both Expected and Observed Survival Time in Palliative Care Consultations

CONTEXT: The rise in routine ambient visit recording enables the scalable study of communication, and identifying clinically important, equitable and epidemiologically useful markers of serious illness conversation is critical to meet these research and quality-improvement opportunities. PURPOSE: The distribution of time that people spend talking about the past, present and future is a foundational feature of computational narrative analysis, and one with promise for large-scale study of serious illness conversations. METHODS: We used a validated natural language processing algorithm to classify the temporal orientation (past, present, and future) of each verb/verb phrase in verbatim transcripts of 231 palliative care inpatient consultations. From these references, we generated measures of prevalence and conversational trajectory. Unsupervised clustering methods and standard epidemiological methods were used to evaluate association with outcomes. RESULTS: < 0.05), and with decisions to limit the acceptable types of interventions (OR: 1.82, 95% CI: 1.07, 3.07) and higher rate of mortality (HR: 1.36, 95% CI: 1.02, 1.82). We observed interaction between the amount of future-oriented language and the trajectory over conversation time on both decisional and survival outcomes. We did not find similar relationships with references to the past or present. CONCLUSIONS: Future-oriented temporal reference is a conceptually grounded and epidemiologically useful marker of decision-making and survival outcomes in palliative care consultations. The overall proportion of conversation time spent talking about the future and the trajectory of how future-oriented talk happens over the course of conversation provide synergistic scientific information.