Developing a predictive model for mortality in patients with cognitive impairment

Abstract

Objectives: We developed a predictive model for all-cause mortality and examined the risk factors for cause-specific mortality among people with cognitive impairment in a Japanese memory clinic-based cohort (2010-2018).

Methods: This retrospective cohort study included people aged ≥65 years with mild cognitive impairment or dementia. The survival status was assessed based on the response of participants or their close relatives via a postal survey. Potential predictors including demographic and lifestyle-related factors, functional status, and behavioral and psychological status were assessed at the first visit at the memory clinic. A backward stepwise Cox regression model was used to select predictors, and a predictive model was developed using a regression coefficient-based scoring approach. The discrimination and calibration were assessed via Harrell's C-statistic and a calibration plot, respectively.

Results: A total of 2610 patients aged ≥65 years (men, 38.3%) were analyzed. Over a mean follow-up of 4.1 years, 544 patients (20.8%) died. Nine predictors were selected from the sociodemographic and clinical variables: age, sex, body mass index, gait performance, physical activity, and ability for instrumental activities of daily living, cognitive function, and self-reported comorbidities (pulmonary disease and diabetes). The model showed good discrimination and calibration for 1-5-year mortality (Harrell's C-statistic, 0.739-0.779). Some predictors were specifically associated with cause-specific mortality.

Conclusions: This predictive model has good discriminative ability for 1- to 5-year mortality and can be easily implemented for people with mild cognitive impairment and all stages of dementia referred to a memory clinic.

Keywords: death; dementia; memory clinic; mild cognitive impairment; mortality; older adults; predictive model; prognosis.