Integrating multiple imputation with harmonization in multisite neuroimaging data: implications for site effects and biological inference
Year: 2026
Type: preprint
Abstract: Multisite neuroimaging harmonization in Alzheimer's disease research with missing data has received limited attention in the literature. Here, we evaluated the performance of several harmonization methods in conjunction with multiple imputation (MI) to address missing data, using baseline [18F]flortaucipir tau positron emission tomography (PET) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). We compared complete case and MICE-imputed datasets (m = 5) across the original ADNI-preprocessed data and four statistical harmonization approaches: ComBat with and without covariates, quantile normalization, and quantile regression. Weighted average standardized uptake value ratios (SUVRs) across six Braak regions were analyzed for each dataset-method combination. We assessed harmonization effectiveness using covariate-adjusted intraclass correlation coefficients and site-level pairwise differences, and evaluated biological effect preservation using Hedges' g, regression coefficients, and statistical significance for comparisons between cognitively impaired and unimpaired participants. For imputed analyses, effect sizes and regression coefficients were pooled using Rubin's rules, with statistical inference based on the pooled estimates. Post-processing harmonization reduced site-level variability compared to the original data, with method performance varying across evaluation metrics, while overall harmonization quality remained comparable after MI. Disease group differences remained statistically significant across all combinations, exhibiting a consistent regional pattern, strongest in Braak I and weakest in Braak VI. MI modestly increased effect size and coefficient estimates, with improved precision reflected by median standard error ratios below 1 (0.89 for effect sizes and 0.93 for coefficients). These findings indicate that integrating MI with harmonization preserves disease group differences and improves statistical efficiency without compromising harmonization performance, supporting a practical framework for multisite neuroimaging studies.
Authors: Yuxin Ren(Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Writing – original draft, Writing – review & editing), Yiwen Dong(Methodology, Writing – review & editing), Davneet Minhas(Data curation, Methodology, Writing – review & editing), Ann D Cohen(Writing – review & editing), Qiong Wu(Conceptualization, Methodology, Supervision, Writing – review & editing), Dana Larisa Tudorascu(Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing)
Institution: University of Pittsburgh
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funders: Foundation for the National Institutes of Health, Alzheimer's Association, Bristol-Myers Squibb, Eli Lilly and Company, Pfizer, Roche, Biogen, Meso Scale Diagnostics, Alzheimer's Disease Neuroimaging Initiative, BioClinica, Novartis Pharmaceuticals Corporation, Northern California Institute for Research and Education, Eisai, Servier, H. Lundbeck A/S, IXICO, National Institutes of Health, Genentech, Canadian Institutes of Health Research, National Institute on Aging, National Institute of Mental Health
OpenAlex ID: W7235445992
