Combining Item Response Theory with Multiple Imputation to Crosswalk Between Health Assessment Questionnaire
*Chenyang Gu, Brown University
Keywords: Item nonresponse, missing data, multiple imputation, crosswalk, generalized partial credit model
The assessment of patients' functional status across the continuum of care requires a common patient assessment tool. However, assessment tools used in various health care settings differ and cannot be easily contrasted. The Inpatient Rehabilitation Facility Patient Assessment Instrument (IRF-PAI) is used to measure activities of daily living when patients stay in a rehabilitation facility; after discharge, the Minimum Data Set (MDS) is collected on patients staying in a nursing home, and the Outcome and Assessment Information Set (OASIS) is used for patients using a home health facility. To compare patients’ functional status across facilities, we assume that all patients have observed IRF-PAI and treat the unmeasured MDS or OASIS items as missing. We propose a variant of the predictive mean matching method that relies on the generalized partial credit model to impute the missing measurement items. Using real data sets, we simulated missing measurements and compared our proposed approach to existing methods for missing data imputation. For all of the estimands examined, the proposed approach had better coverages, smaller biases, and shorter interval estimates.