Multiple Imputation of Missing Values Using the Response Function Method Based on a Data Set of the Health Assessment Questionnaire Disability Index

摘要

Objectives: This study aims to investigate how imputingmissing values in data obtained from the HealthAssessment Questionnaire Disability Index (HAQ-DI)influences the bias and precision of patient disabilitymeasurements.Patients and methods: Hypothetical missing data sets werecreated by deleting item responses completely at randomfrom the original data set with three missingness proportions(0.10, 0.30 and 0.50). Multiple imputation was carried outusing the response function method for each hypotheticaldata set containing the missing values. The Rasch modelwas used to estimate the patients’ latent trait levels for theoriginal data, the hypothetical incomplete data sets, andthe multiple imputed data sets. Then the estimates from thehypothetical missing data sets and the multiple imputed datasets were compared with those of the original data set.Results: A bias in disability estimates was observed,particularly as the missingness proportion increasedfor both the incomplete and imputed data; however, thisbias was indiscernible even for the 0.50 proportion ofmissingness. In terms of the uncertainty of the disabilityestimates, the imputed data had a higher precision ofestimates than the incomplete data.Conclusion: When researchers encounter missingnessin data collected with the HAQ-DI, the response functionimputation could be a convenient approach to imputemissing values in order to improve the precision of thepatient disability level estimates.