Predicting Mortality and Costs After Emergency Department Visits by People With Dementia: Timing and Location Matter.

BACKGROUND

People with dementia have high rates of emergency department (ED) and hospital utilization, high mortality and costs, and other poor outcomes. To successfully impact the care trajectories of these patients, health care systems must pragmatically identify the correct target population. This study described patterns of ED utilization by people with dementia and explored the accuracy of administrative data models to predict mortality and costs.

METHODS

Retrospective cohort study of a 20% random sample of Traditional Medicare (TM) beneficiaries with dementia, age ≥ 66 years, and an index ED visit in 2018. One-year mortality and high costs were described, and associations with the timing of prior hospitalizations examined. As a preliminary step to evaluate models based on administrative data only, C-statistics were used to examine the accuracy of eight multivariate models, stratified by the setting of care before and after an ED visit.

RESULTS

The majority of the 250,343 person cohort of individuals with dementia resided in the community before their index ED encounter (83.9%) rather than in a nursing home (NH, 16.1%), and 34.4% required hospitalization. One-year mortality by location varied from 18.4% (community before and after ED visit) to 47% (admitted NH residents). One-year mortality was 40.3% for those hospitalized within the past month, and 26.2% for those hospitalized 6+ months before the ED encounter. C-statistics were less than or equal to 0.72 for seven multivariate models, and 0.81 for the model examining high costs for NH residents discharged back to their NH.

CONCLUSIONS

Mortality and costs for people with dementia vary by location of care before and after ED encounters, as well as by timing of prior hospitalizations. However, multivariate models using only administrative data lack accuracy, suggesting the need to add pragmatically selected clinical data and/or other measures to better identify the "right patients, at the right time".