Missing data and imputation in the international studies following the Harmonized Cognitive Assessment Protocol.

OBJECTIVES: Missing data poses special challenges in aging research when health and functional declines of older adults limit data collection. Large-scale, cross-national studies further suffer from cultural, linguistic, and educational differences across populations. Little is known about such challenges and mitigation strategies.

METHODS: We leveraged international studies following the Harmonized Cognitive Assessment Protocol (HCAP), which was designed to ascertain the prevalence of and characterize relevant risks for mild cognitive impairment and dementia globally. By describing the current practice in studies under six diverse geographic contexts (United States, England, Chile, Mexico, India, and China), we documented the ways in which missing data were harmonized and handled to facilitate cross-national research.

RESULTS: Various missing codes were used to document reasons for missingness. Overall, the rates of missingness were low, around 15% of observations in studies for India and China, and up to 7% for other countries. Item-level missingness and missing subinterviews (neurocognitive assessments on the older adults or their informants' reports) were imputed using chained equations. Dementia classification was also imputed for individuals in the parent studies from the Health and Retirement Study-International Family Studies (HRS-IFS), of which the HCAPs are substudies.

DISCUSSION: The current practice worked well in the HCAP studies. We include concrete recommendations on best practices for study design, data collection, and post-hoc treatments to mitigate challenges due to missing data, and for use and reporting of the imputed HCAP data.