Investigating predictors of incident cognitive impairment in women

Abstract

AbstractBackgroundMore women than men are affected by Alzheimer’s disease (AD). However, risk factors for cognitive impairment (CI) among older women by race/ethnicity are not completely understood. The well‐characterized Women’s Health Initiative Memory Study (WHIMS) has followed for more than 25 years postmenopausal women to assess cognitive and brain outcomes. WHIMS provides a unique opportunity to determine the extent to which risk factors for CI vary between White and Black women. Due to the high‐dimensional nature of this problem we based our analyses on machine learning methods.MethodTo investigate signatures of predictors of incident CI (either mild cognitive impairment or probable dementia) in women, we included 6778 women (503 Black and 6275 White) with up to 25 years of follow‐up including annual cognitive assessments. Mild cognitive impairment and probable dementia were adjudicated annually across all phases of the trial and subsequent observational study by a central, multidisciplinary adjudication panel of experts. We selected 52 potential predictors collected at baseline representing different domains–demographic, health, social, psycho‐social, biological, and medical. Survival Random Forests (SRF) were used to identify the most reliable predictors of incident CI. Here we use the SRF implementation available in the RandomForestSRC R library. SRF models were fitted independently using White and Black participants based on the default parameters of the software. The permutation index was used to determine predictors’ relevance. The concordance index was used as a measure of model performance.ResultThe most important predictors of incident CI in Black women were lower education, higher systolic blood pressure, lower income and higher high‐density lipoprotein cholesterol. The concordance index was 0.64. A different profile of predictors was observed among White women. Baseline age was the most relevant predictor of ICI followed by self‐reported forgetfulness, self‐reported difficulty concentrating, physical function and higher depression. The concordance index for this model was 0.71.ConclusionPhenotypic signatures predicting incident CI differed between older Black and White women. Education, markers of vascular disease and income were major predictors in Black women and age, self‐reported cognitive difficulties, physical function and mood were major predictors in White women.