Progress towards predicting 1-year mortality in older people living in residential long-term care.

Background: frail older people living in residential long-term care (LTC) have limited life expectancy. Identifying those with poor prognosis may improve management and facilitate transition to a palliative approach to care. Objective: to develop methods for predicting mortality in LTC. Design: a population-based cohort study. Setting: LTC facilities, Auckland, New Zealand. Subjects: five hundred randomly selected older people in a census-type survey of those living in LTC in 2008. Methods: mortality data were obtained from New Zealand Ministry of Health. Two methods for assessing mortality risk were developed using demographic, functional and health service information: (i) two geriatricians blinded to identifying data and to mortality, independently reviewed survey, medications and pre-survey hospitalisations data, and grouped residents according to perceived risk of death within 12 months; (ii) multivariate logistic regression model used the same survey and medication items as the geriatricians. Results: for the geriatricians' assessment, each quintile of perceived risk was associated with a significant increase in mortality (P < 0.001). Area under the curve (AUC) for both physicians was 0.64. The logistic regression model included age, gender, assistance with feeding and requiring night attention, all variables which are easily available from LTC records. AUC for the model was 0.70, but when validated against the entire OPAL cohort, it was 0.65. When either or both geriatrician and the model together predicted high risk of death, 1-year mortality was >50%. Conclusion: two methods with the potential to identify older people with limited prognosis are described. Use of these methods allowed identification of over half of those who died within 12 months.