Co-calibration of cognitive performance in the National Health and Aging Trends Study with the Health and Retirement Study's Harmonized Cognitive Assessment Protocol: Implications for dementia classification.

BACKGROUND: Large population-based studies are crucial for dementia research; yet variation in cognitive tests and dementia classification approaches can lead to inconsistent findings. We harmonized cognitive data from two nationally representative US studies of aging to facilitate comparisons.

METHODS: We examined 2016 data for individuals aged ≥70 years from the National Health and Aging Trends Study (NHATS) (n = 5696) and the Harmonized Cognitive Assessment Protocol (HCAP) of the Health and Retirement Study (HRS) (n = 2731). We derived factor scores for general cognitive performance in the NHATS cognitive test battery that were co-calibrated to the HRS-HCAP battery and identified cutpoints for dementia that returned the expected prevalence in each study. We evaluated diagnostic characteristics of the cutpoints against study-specific dementia algorithms with Area Under the Curve analysis.

RESULTS: Study-specific algorithms yielded comparable dementia prevalence estimates: 10.8 % in NHATS and 11.1 % in HRS-HCAP. Co-calibrated scores showed similar distributions and had acceptable reliability, with similar dementia cutpoints. In both studies, sensitivity was higher among lower-educated (vs. higher educated) and non-White (vs. non-Hispanic White) groups. Co-calibrated populations with dementia in both studies had similar age and gender distributions but differed somewhat in education levels and race/ethnicity profiles.

CONCLUSIONS: NHATS and HRS-HCAP both provide reliable cognitive function measures and dementia prevalence estimates for older Americans. Co-calibrated scores based on each study's cognitive test battery provide a valid and feasible approach for comparative US research. Better aligned algorithmic approaches across studies could strengthen opportunities for comparative studies of disparities in the US context using the co-calibration approach.