MIGEE: A comprehensive package for missing data imputation and longitudinal analysis in large-scale datasets

TY - JOUR

T1 - MIGEE

T2 - A comprehensive package for missing data imputation and longitudinal analysis in large-scale datasets

AU - Kumar, Neelesh

AU - Bhattacharjee, Atanu

AU - Vishwakarma, Gajendra K.

AU - Majumdar, Tanmoy

N1 - Publisher Copyright: © 2026 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

PY - 2026/6/13

Y1 - 2026/6/13

N2 - This study introduces MIGEE, an R package that combines multiple imputation strategies, longitudinal data modelling, and visualization within a single streamlined workflow. The package implements six imputation methods (PMM, kNN, norm, RF, norm.nob, and sample) and supports both linear and mixed-effects models for downstream analysis. Unlike conventional workflows that require manual coordination across packages such as mice, lme4, and ggplot2, MIGEE reduces the full impute–model–visualize pipeline to a single function call, eliminating intermediate data reshaping steps (e.g., long-to-wide transformations) and ensuring structural consistency between imputation and modelling outputs. This results in a reduction of more than 80% in user-written code (from more than 50 lines across seven manual steps to approximately 10 lines in a single function call) and consolidates seven conceptually distinct pipeline steps into one, removing multiple preprocessing operations typically required in manual workflows. It was evaluated using a longitudinal clinical dataset comprising 11,761 rows and 12 variables from 2000 patients with repeated measurements over time. Across this dataset, all imputation methods successfully preserved the mean and variability of incomplete variables, with differences in mean below 0.2 units and standard deviation differences below 0.5 units, indicating negligible bias and stable treatment-effect estimates. MIGEE streamlines complex longitudinal workflows, reduces computational burden, and supports reproducible research in biomedical and population health applications. The MIGEE package is publicly available on CRAN at https://doi.org/10.32614/CRAN.package.MIGEE.

AB - This study introduces MIGEE, an R package that combines multiple imputation strategies, longitudinal data modelling, and visualization within a single streamlined workflow. The package implements six imputation methods (PMM, kNN, norm, RF, norm.nob, and sample) and supports both linear and mixed-effects models for downstream analysis. Unlike conventional workflows that require manual coordination across packages such as mice, lme4, and ggplot2, MIGEE reduces the full impute–model–visualize pipeline to a single function call, eliminating intermediate data reshaping steps (e.g., long-to-wide transformations) and ensuring structural consistency between imputation and modelling outputs. This results in a reduction of more than 80% in user-written code (from more than 50 lines across seven manual steps to approximately 10 lines in a single function call) and consolidates seven conceptually distinct pipeline steps into one, removing multiple preprocessing operations typically required in manual workflows. It was evaluated using a longitudinal clinical dataset comprising 11,761 rows and 12 variables from 2000 patients with repeated measurements over time. Across this dataset, all imputation methods successfully preserved the mean and variability of incomplete variables, with differences in mean below 0.2 units and standard deviation differences below 0.5 units, indicating negligible bias and stable treatment-effect estimates. MIGEE streamlines complex longitudinal workflows, reduces computational burden, and supports reproducible research in biomedical and population health applications. The MIGEE package is publicly available on CRAN at https://doi.org/10.32614/CRAN.package.MIGEE.

KW - Data transformation

KW - Data visualization

KW - Generalized linear model

KW - Linear mixed-effects models

KW - Longitudinal data analysis

KW - Missing data imputation

UR - https://www.scopus.com/pages/publications/105041552287

U2 - 10.1016/j.jocs.2026.102906

DO - 10.1016/j.jocs.2026.102906

M3 - Article

AN - SCOPUS:105041552287

SN - 1877-7503

VL - 99

JO - Journal of Computational Science

JF - Journal of Computational Science

M1 - 102906

ER -