Griffith et al., 2013. Harmonization of Cognitive Measures in Individual Participant Data and Aggregate Data Meta-Analysis


Objectives: The aim of this study was to identify approaches to statistical harmonization which could be used in the context of summary data and/or individual participant data meta-analysis of cognitive measures and to apply and evaluate these different approaches to cognitive measures from three studies.

Data Sources: MEDLINE®, Embase, Web of Science and MathSciNet with a supplemental search using the Google search engine. The references of relevant articles were also checked and a search for more recent articles that cited the articles already identified as being of interest was undertaken.

Review methods: A two-pronged approach was taken for this environmental scan. First, a search of studies that quantitatively combined data on cognition was conducted. The second component was to identify general literature on statistical methods for data harmonization. Standard environmental scan methods were used to conduct these reviews. The search results were rapidly screened to identify articles of relevance to this review. The references of relevant articles were checked and a search for more recent articles that cited the articles already identified as being of interest was undertaken.

Results: Three general classes of statistical harmonization models were identified: (1) standardization methods (e.g., simple linear-, Z-transformations, T-scores, and C-scores); (2) latent variable models; and (3) multiple imputationmodels. Cross-sectional data from three studies including 9,269 participants were included in the applied analyses to examine the relationship between physical activity and cognition. A harmonization process was undertaken to determine the combinability of data across studies. The latent variable analysis underscored the difficulty harmonizing these cognition data. In general consistency was found among the statistical harmonization methods; however, there was some evidence that heterogeneity can be masked when specific standardization methods were used.

Conclusions: This study provides empirical evidence to inform methods of combining complex constructs using aggregate data (AD) or individual participant data meta-analysis. The results underscore that very careful consideration of inferential equivalence needs to be undertaken prior to combining cognition data across studies. Of the three methods of statistical harmonization for cognition data, T-score standardization is the least desirable compared with the centered score method or latent variable methods. Finally, assessment of the assumptions underlying statistical harmonization is not possible without some individual-level data which are required to assess the potential for bias in combining complex outcomes using AD meta-analysis.