Multiple imputation was an efficient method for harmonizing the Mini-mental State Examination with missing item-level data

Richard Burns, Peter Butterworth, Kim Kiely, Allison Bielak, Mary Luszcz, Paul Mitchell, Helen Christensen, Chwee von Sanden, Kaarin Anstey

    Research output: Contribution to journalArticle

    16 Citations (Scopus)

    Abstract

    Objective: The Mini-Mental State Examination (MMSE) is used to estimate current cognitive status and as a screen for possible dementia. Missing item-level data are commonly reported. Attention to missing data is particularly important. However, there are concerns that common procedures for dealing with missing data, for example, listwise deletion and mean item substitution, are inadequate. Study Design and Setting: We used multiple imputation (MI) to estimate missing MMSE data in 17,303 participants who were drawn from the Dynamic Analyses to Optimize Aging project, a harmonization project of nine Australian longitudinal studies of aging. Results: Our results indicated differences in mean MMSE scores between those participants with and without missing data, a pattern consistent over age and gender levels. MI inflated MMSE scores, but differences between those imputed and those without missing data still existed. A simulation model supported the efficacy of MI to estimate missing item level, although serious decrements in estimation occurred when 50% or more of item-level data were missing, particularly for the oldest participants. Conclusions: Our adaptation of MI to obtain a probable estimate for missing MMSE item level data provides a suitable method when the proportion of missing item-level data is not excessive.

    Original languageEnglish
    Pages (from-to)787-793
    Number of pages7
    JournalJournal of Clinical Epidemiology
    Volume64
    Issue number7
    DOIs
    Publication statusPublished - 2011

    Fingerprint Dive into the research topics of 'Multiple imputation was an efficient method for harmonizing the Mini-mental State Examination with missing item-level data'. Together they form a unique fingerprint.

  • Cite this