Generalizability of clinical predictors for mortality from lower respiratory infection (LRI) in nursing home residents has not been assessed for residents with dementia.

In prospective cohort studies of LRI in 61 nursing homes in the Netherlands (n = 541) and 36 nursing homes in Missouri, USA (n = 564), we examined 14-day and 1- and 3-month mortality in residents with dementia who were treated with antibiotics.

A logistic model predicting 14-day mortality derived from Dutch data included eating dependency, elevated pulse, decreased alertness, respiratory difficulty, insufficient fluid intake, high respiratory rate, male gender, and pressure sores. After adjusting coefficients with the heuristic shrinkage factor, the 14-day model showed good discrimination and calibration in both datasets. The apparent c-statistic for the original Dutch model was 0.80 (after correction for optimism, it was 0.75); the c-statistic was 0.74 in the U.S. validation population. The models predicting 1- and 3-month mortality showed moderate performance. A scoring system for estimating 14-day mortality performed equally well as the original model.

We identified a set of credible clinical predictors that are easily assessed and demonstrated validity in identifying residents at low risk of dying from LRI across different nursing home populations. This tool should inform decision-making for families and doctors.