Skip to content

Projects | In Progress

Systematic Identification of Nursing Home Residents with 6-month Mortality Risk for Hospice Informational Referral

Research Areas


3.3 million Americans are living with moderate-to-late-stage Alzheimer’s Disease or related dementias (ADRD). They suffer from cognitive, physical, and neuro-psychiatric symptoms along with increasing dependency. Nursing home (NH) care is common for these individuals. Their deterioration results in frequent, yet burdensome hospitalizations, but 23-47% of ADRD hospitalizations are potentially avoidable. It may be beneficial to identify NH residents living with late-stage ADRD who are at high mortality risk. Then, we may prompt referrals to hospice informational visits — in order to reduce aggressive, hospital-based treatments near the end of life.

The Medicare Hospice Benefit pays for 86% of hospice in the US, and, to be eligible, it requires an estimated 6-month prognosis. Although hospice use for NH residents living with ADRD is increasing, the benefit remains under-utilized. This is, in part, because outcomes for ADRD are notoriously difficult to predict. Although clinical prediction models have been developed for ADRD, many predict 12-month (rather than 6-month) mortality, and their predictive performance are insufficient to be used to determine hospice eligibility.

In our study, we sought to (1) develop an algorithm to identify 6-month mortality among NH residents living with ADRD using Minimum Data Set (MDS) data from NH data; (2) assess the added value of combining electronic health record (EHR) data to MDS data; and (3) evaluate whether model predictions are equitably calibrated across groups such as race, ethnicity, and sex.

We developed and compared two models: i) elastic net regression and ii) XGBoost, a boosted tree-based model. Models included variables from the MDS only or MDS plus EHR data. We estimated performance for all candidate models across demographic subgroups to ensure equitable performance. We developed and tested the model using data from the first 100 days to predict mortality over the following six months among NH residents with ADRD.

Partnering Health System

Genesis HealthCare


National Institute on Aging