Multiple imputation for harmonizing longitudinal non-commensurate measures in individual participant data...

Area of Science:

• Biostatistics
• Psychiatry
• Epidemiology

Background:

• Individual participant data (IPD) meta-analysis offers advantages over published-results meta-analysis, including increased statistical power and sample heterogeneity.
• A key challenge in IPD meta-analysis is inconsistent measurement of variables across studies, hindering data combination.
• Adolescent depression research often involves multiple trials with varying depression assessment tools.

Purpose of the Study:

• To propose and evaluate a multiple imputation method for handling missing outcome data in IPD meta-analysis when measurement instruments differ across studies.
• To address the challenge of combining data from adolescent depression trials using distinct depression measures.

Main Methods:

• Framed the measurement inconsistency as a missing data problem solvable by multiple imputation.
• Applied multiple imputation using external calibration studies where both depression measures were available.
• Considered treatment group and study as factors in the imputation model and assessed imputation model fit.

Main Results:

• Successfully applied a novel multiple imputation strategy to integrate data from five longitudinal adolescent depression trials with differing depression measures.
• Demonstrated the utility of external calibration studies within the imputation framework to bridge measurement gaps.
• Provided diagnostics for evaluating the imputation model's performance and the incorporation of external information.

Conclusions:

• Multiple imputation, leveraging external calibration data, provides a viable solution for the missing data challenge in IPD meta-analysis with heterogeneous measures.
• This approach enhances the feasibility and robustness of combining data from diverse clinical trials, particularly in adolescent mental health research.
• The proposed method facilitates more sophisticated and reliable analyses by creating a unified dataset from disparate sources.