Speaker: Ian Duncan
The cost of dying in is significant; in the United States, about one-third of all Medicare expenses are incurred in the last 6 months of a person's life with 8 out of 10 deaths being covered under the Medicare benefit. We discuss the causes, process, place and cost of death, looking at differences between inpatient, skilled-nursing and hospice care. Death is frequently over-medicalized, with patients spending their last days away from loved ones undergoing painful surgeries and therapy of low value mis-aligned to patient/family preferences. There are alternatives that are more dignified and consistent with patient and family wishes, including hospice care. In the U.S. patients are admitted to hospice with a life expectancy of less than 6 months; the median duration of stay is 14 days.
Much of the (considerable) literature on patient survival rates in hospice addresses either the accuracy of physician prognosis or applies Kaplan-Meier models to available admission or clinical data [1-16]. Chiang et al  point to a shortcoming of the static statistical models: "..none of these methods can give us the full picture of the changes in survival probability for a given patient, over time, until death with the best available statistical accuracy." We have access to a large hospice dataset that includes detailed prescription drug data that allows us to assess changes in drug, dose, strength and route of administration over time. We apply time-dependent Cox proportional hazard models to predict survival of hospice patients from admission to death.