Predictors of hospital mortality in older adults with and without dementia: a necessary step towards personalized prognostication

Abstract

AbstractBackgroundDementia has been associated with increased risk of hospital mortality (1). However, predictors of hospital mortality for patients with dementia have not been widely addressed. We aimed to investigate predictors of hospital mortality according to preexisting clinical dementia in older adults.MethodsWe analyzed data from a prospective cohort study in Sao Paulo, Brazil. We included acutely ill patients aged ≥60 years admitted to a tertiary university hospital between 2009 and 2020. Candidates underwent comprehensive geriatric assessments in the first 24 hours of hospitalization. Preexisting clinical dementia (after this, referred to as dementia) was determined by a Clinical Dementia Rating (CDR) ≥1 (2). We investigated independent predictors of hospital mortality according to dementia using multivariable Poisson regressions. We further analyzed our multivariable models' accuracy in predicting hospital mortality using area under the ROC curves (AUC).ResultsWe included 3,863 participants (mean age = 80; female = 62%). Overall, 1,761 participants (46%) had dementia. Mortality was 11% in participants without dementia and 21% in those with dementia (p<0.001). In participants without dementia, mortality was independently predicted by age, multimorbidity, functional status, delirium, nutritional status, and serum albumin (Table 1). However, in participants with dementia, mortality was also predicted by illness acuity and creatinine. Finally, we observed that our multivariable model had an AUC = 0.88 (95%CI = 0.86‐0.90) in participants without dementia and 0.80 (0.78‐0.83) in those with dementia (Figure 1).ConclusionOlder adults with dementia were twice as likely to die in the hospital than those without dementia. Moreover, independent predictors of hospital mortality were not necessarily shared between patients with and without dementia. Given the importance of prognostic assessments for decision‐making processes, researchers should consider exploring new algorithms that differ according to baseline cognition.