Trends in the Risk of Cognitive Impairment in the United States, 1996-2014

Abstract

Background: Accumulating evidence suggests risk of cognitive impairment is declining in high-income countries. Much of this research uses longitudinal surveys in which learning over repeated tests may bias results. We analyze trends in cognitive impairment in the United States, accounting for prior test experience and selective mortality.

Methods: We use the Health and Retirement Study, a population-based, nationally representative panel dataset and include individuals ages 50 years and older in 1996-2014 (n = 32,784). We measure cognitive impairment and dementia using standard cutpoints of the modified Telephone Interview for Cognitive Status. We estimate logistic regression models for any impairment and dementia over time, adjusting for age, sex, and race/ethnicity, comparing models with and without adjustment for practice effects and education. We examine heterogeneity in trends by age, sex, race/ethnicity, and education.

Results: Models not controlling for test experience suggest that risk of cognitive impairment and dementia decreased over the study period. Controlling for test experience reverses the trend. In our primary models, prevalence of any cognitive impairment increased for women from 18.7% to 21.2% (annual change 0.7%, 95% confidence interval [CI], 0.1%, 1.3%) and for men from 17.6% to 21.0% (annual change 1.0%, CI, 0.5%, 1.4%). For dementia, women's annual increase was 1.7% (CI, 0.8%, 2.6%) and men's 2.0% (CI, 1.0%, 2.9%). If not for education, the increase would have been stronger. Increased risk was particularly rapid for Latinas, the least educated, and older ages.

Conclusions: Risk of cognitive impairment increased from 1996 to 2014. Uncovering determinants of increasing cognitive impairment risk should become a research priority. See video abstract: http://links.lww.com/EDE/B702.

FIGURE 1.

Odds ratios and 95% confidence intervals for time from logistic regression models of any cognitive impairment or dementia (Model 1 includes pairs of survey waves, age, age squared, race/ethnicity, and the interaction between gender and interview mode. Model 2 adds test number. Model 3 adds educational attainment).

FIGURE 2.

Odds ratios and 95% confidence intervals for time from logistic regression models for any impairment (A) and dementia (B), estimated from subsamples restricted to first, second, fourth, or sixth test (The reference category is 1996–1998 for tests 1 and 2, 2002–2004 for test 4, and 2004–2006 for test 6).