Skip to main content

Languages

Measurement Harmonization in Individual-Participant-Data Meta-analysis

Date and Location

Session: 

P3.039

Date

Sunday 22 September 2013 - 10:30 - 12:00

Location

Presenting author and contact person

Presenting author

Tania B. Huedo-Me...

Contact person

Tania B. Huedo-Me...
Abstract text
Background: Our project stands to advance measurement harmonization a great deal by carefully developing statistical methods to support the practice of individual participant data meta-analysis. Some statistical harmonization approaches have been already developed to integrate different measures but there is not a compromised solution in a meta-analytic context with complex health behavior constructs. Objectives: To review the methodological approaches to combine different measures of the same construct, and, finally, compare one of the most common ones, standardization with a moderated nonlinear factor analysis (MNLFA). Statistical performance of each one will be evaluated under different circumstances to, and finally, apply to a real dataset. Methods: For this study we review the statistical harmonization techniques that have been used in the literature and we compared simple standardization to a method can be applied to the dichotomous, ordinal, and sometimes continuous measures, MNLFA. We used Monte Carlo simulations and real archival data from seven HIV prevention intervention trials. Different measurement scales and distributions were created using parameters and scales derivate from the real HIV prevention interventions individual data. In order to evaluate the robustness of the measurement scales under different conditions the percentage bias of the estimate was calculated as well as the efficiency as the variability of the estimate across replications. Results: The MNLFA was more efficient and unbiased in most of the circumstances and it did not make a difference from a simple standardization when the measures were all continuous and normal regardless the sample sizes and the number of studies. Conclusions: The moderate nonlinear factor analysis is more generalizable harmonization technique when individual data needs to be integrated than just standardizing the metrics. The most important limitations are that requires at least some items to overlap within studies and the data needs to be independent within study.