Multiple Imputation Methods for Missing Multilevel Ordinal Outcomes

TL;DR

This study compares multiple imputation methods for handling missing multilevel ordinal data, highlighting the importance of including cluster size to improve accuracy, with FCS outperforming other approaches in simulations and real data.

Contribution

It introduces and evaluates strategies for imputing multilevel ordinal outcomes with informative cluster size, an area previously underexplored.

Findings

Including cluster size improves imputation accuracy.

FCS provides more accurate estimates than JM.

Methods are validated through simulations and a real dental study.

Abstract

Multiple imputation (MI) is an established technique to handle missing data in observational studies. Joint modeling (JM) and fully conditional specification (FCS) are commonly used methods for imputing multilevel clustered data. However, MI approaches for ordinal clustered outcome variables have not been well studied, especially when there is informative cluster size (ICS). The purpose of this study is to describe different imputation and analysis strategies for the multilevel ordinal outcome when ICS exists. We conducted comprehensive Monte Carlo simulation studies to compare five different methods: complete case analysis (CCA), FCS, FCS+CS (include cluster size (CS) when performing the imputation), JM, and JM+CS under different scenarios. We evaluated their performances using an proportional odds logistic regression model estimated with cluster weighted generalized estimating equations (CWGEE). The simulation results show that including cluster size in imputation can significantly improve imputation accuracy when ICS exists. FCS provides more accurate and robust estimation than JM, followed by CCA for multilevel ordinal outcomes. We further applied those methods to a real dental study.