Statistical harmonization of versions of measures across studies using external data: self-rated health and self-rated memory

Purpose: Harmonizing variables for constructs measured differently across studies is essential for comparing, combining, and generalizing results. We developed and fielded a brief survey to harmonize Likert and continuous versions of measures for two constructs, self-rated health and self-rated memory, for use in studies of French older adults.

Methods: We recruited 300 participants from a French memory clinic in 2023 to answer both the Likert and continuous versions of self-rated health and self-rated memory questions. For each construct, we predicted responses to the Likert version with multinomial and ordinal logistic models, varying specifications of continuous version responses (linear or spline) and covariate sets (question order, age, sex/gender, and interactions between the continuous version and covariates). We also implemented a percentiles-based crosswalk sensitivity analysis. We compared Cohen's weighted kappa values to identify the best statistical harmonization approach.

Results: In the final models [multinomial models with continuous version spline, question order (self-rated memory model only), age, sex/gender, and interactions between the continuous version and covariates], weighted kappa values were 0.61 for self-rated health and 0.60 for self-rated memory, reflecting moderate agreement.

Conclusions: Primary data collection feasibly facilitates statistical harmonization of variables for constructs measured differently across studies.