Differences in the measurement of cognition for the assessment of dementia across geographic contexts: Recommendations for cross‐national research

Background Although 71% of individuals with dementia are projected to live and low‐ and middle‐income countries by 2050, most dementia research to‐date has been conducted in high‐income countries. Cross‐national studies are needed to understand how the causes and consequences of dementia may differ by country. However, factors such as urbanicity, language of administration or race/ethnicity, which vary across countries, affect the measurement of cognition. A better understanding of which cognitive items perform well across countries or within specific countries will allow researchers to optimize dementia measurement in future cross‐national and international studies. Method We used data from the Harmonized Cognitive Assessment Protocol surveys in the US, Mexico, India, England, and South Africa (combined N = 11,364). To define cognitive impairment consistently across countries, we compared participants’ cognitive performance with robust neuropsychological norms within each country. For each HCAP country, we assessed item performance by estimating associations between each cognitive item and cognitive impairment using logistic regression models, controlling for age and gender. We avoided circularity in the analysis by using an iterative quasi‐leave‐one‐out approach. We compared patterns of associations across countries using median odds ratios, and visually, using heatmaps. Result The associations between cognitive items and cognitive impairment were stronger in the US (Median Odds Ratio [OR] = 0.17) and England (Median OR = 0.19), in comparison to South Africa (Median OR = 0.23), India (Median OR = 0.29), and Mexico (Median OR = 0.28). Memory items, notably delayed recall tasks, had the most consistent associations of the largest magnitudes across countries. In comparison, there was variability in performance between settings for many language items (e.g. naming a hammer). Items requiring numeracy, including the Trail‐making test, did not perform well in countries with low educational attainment. Conclusion The performance of cognitive items for the classification of cognitive impairment was not consistent across countries. Although items performing well in a single country could be used to improve measurement precision in specific settings, transporting items between countries without prior validation warrants caution. However, we did identify items that performed well across a range of countries (i.e. delayed recall, animal naming); these items may be of