Predictors of long-term nursing home placement under competing risk: evidence from the Health and Retirement Study

To examine predictors of long-term nursing home placement (LTNHP) while controlling for mortality as a competing risk event.

Longitudinal.

Health and Retirement Study, 1998-2010.

Nationally representative sample of community-living older adults (N = 10,385).

Longitudinal data were used with a maximum follow-up of 12 years. First, a traditional Cox proportional hazards model was estimated treating death as an uninformative censoring event. A joint cause-specific hazards model that accounts for the competing risk of mortality in estimating the risk of LTNHP was then estimated.

The effect of adjusting for competing risk of mortality is evident for nearly all predictors of LTNHP. Predictors were over- or underestimated in the traditional Cox model, and several predictors changed in the direction of the association, whereas others changed in magnitude. For example, after controlling for mortality, women aged 85 and older had more than twice the risk (hazard ratio = 7.23, 95% confidence interval = 5.18-10.10) of LTNHP than evidenced in the traditional Cox model.

Whenever possible, the competing risk of mortality should be recognized and adjusted in developing screening tools and predictive risk models for LTNHP.