Development and Validation of Models to Estimate the Incident Risk of Cognitive Impairment and Atherosclerotic Cardiovascular Disease in Older Adults

Abstract

Background Guidelines emphasize using atherosclerotic cardiovascular disease (ASCVD) risk prediction models for treatment decisions, but risk of cognitive impairment is an equally important concern in older adults. Current ASCVD risk prediction models were derived in younger adults and do not include holistic measures of health or predict cognitive impairment. Methods We utilized data from the Framingham, Framingham Offspring, CHS (Cardiovascular Health Study), and ARIC (Atherosclerosis Risk in Communities) cohorts to derive and validate 2 Selective Functional Prediction models to estimate an older person's (aged ≥75 years) risk within 5 years of developing incident: (1) cognitive impairment; and (2) ASCVD, while accounting for the competing risk of death. Variable selection, including functional status, was based on the least absolute shrinkage and selection operator method. The cognitive impairment (N=3466) and ASCVD (N=4403) model populations were split into derivation and validation cohorts with external validation, then performed in MESA (Multi‐Ethnic Study of Atherosclerosis). Results In the derivation and external validation cohorts (median age, 79 years), 579 (16.7%) and 67 (15.3%) participants developed incident cognitive impairment, respectively; 748 (17.0%) and 80 (8.4%), respectively, experienced an ASCVD event. The cognitive impairment model (baseline Mini‐Mental State Examination (MMSE), atrial fibrillation, antidepressant use, mobility impairment, and dependence for grocery shopping) had good discrimination in the internal and external validation cohorts (C index 0.75 and 0.73, respectively). The ASCVD model (employment status, MMSE, aspirin, lipid‐lowering medications, blood pressure medications, systolic blood pressure, general health status, high‐density lipoprotein cholesterol, triglycerides, creatinine, and mobility impairment) had satisfactory discrimination (C index 0.67) on internal validation and outperformed the pooled cohort equations, but had modest discrimination (C index 0.59) on external validation. Although both models were well calibrated in the internal validation cohorts, they overpredicted risk in the external validation cohort. Conclusions Accurate prediction of an older person's risk of developing cognitive impairment is possible, but predicting future ASCVD events remains more challenging.

Keywords

• cardiovascular disease
• cognitive impairment
• geriatric
• older adults
• prevention
• risk prediction