Date Presented 4/18/2015
In this study, we extracted and linked two activities of daily living (ADL) scales from a national database: the National Health and Nutrition Examination Survey (NHANES). We challenge the notion that measurement across the continuum of health care can only be accomplished by all health care facilities using the same instrument.
SIGNIFICANCE: A critical problem in health care is to monitor patients across the continuum of care. The Classical Test Theory (CTT) promotes all health care providers to use a single instrument across facilities. However, Item Response Theory (IRT) provides an alternative approach to monitor patients across the continuum of care using existing instruments. Using the IRT method, a common metric across different instruments can be established. In this study, we aim to compare (1) item-level psychometric properties, (2) estimated mean person measures, and (3) measurement precisions of two linked scales.
INNOVATION: In this study, we challenge the common belief that using a single measure is necessary to monitor patients across the continuum of care. We used IRT common person equating to link two activities of daily living (ADL) scales generated from the 2005 to 2012 National Health and Nutrition Examination Survey (NHANES) database. This is the first study to use a national database to demonstrate IRT common person equating and to compare the measurement precision of two linked instruments.
APPROACH: Hypotheses included the following: (1) ADL instruments generated from the same national database will have similar item-level psychometrics, (2) linked ADL measures will generate similar person measures for community-dwelling individuals poststroke, and (3) linked ADL measures established equivalently will show similar precision.
Although developing a single instrument would involve considerable costs and resources, an IRT approach is cost efficient to translate instruments presently used across facilities. The viability of the IRT method is dependent on showing psychometrics across instruments. We propose to accomplish this by generating and linking two ADL scales from the NHANES database and by comparing item-level psychometrics, person measures, and measurement precisions of two scales.
METHOD: We applied a secondary and cross-sectional study design, and we used IRT common person equating to link the scales.
The study took place in a community-dwelling setting. Participants included 465 individuals who reported having a stroke. Twenty ADL items were chosen from the NHANES data set and were divided into two 10-item ADL scales equivalently. Factor loading and residual correlations were examined with Mplus. We achieved the model fit, person strata, and cocalibration using Winsteps Rasch analysis. We examined scale standard errors using Item Response Theory for Patient-Reported Outcomes (IRTPRO).
RESULTS: Hypothesis 1 was partially supported, with unidimensionality of both scales but different difficulty of the scales. Hypothesis 2 was supported with similar person ability means (1.76 ± 1.84, 1.79 ± 1.81). Hypothesis 3 was partially supported with similar person strata (2.71, 2.04). Scale 1 was more precise for persons with lower ability, and Scale 2 was more precise for persons with higher ability.
CONCLUSION: The majority of the findings supported cocalibration as a method of linking different ADL scales. Previous studies also supported feasibility and validity of linking health care measures. Although instruments can be linked, the individual instruments may have different levels of precision dependent on the ability level of the individual, which supports facilities maintaining the use of existing instruments. Because measurement precision can be different across the breadth of an instrument that can cover the range of person ability, practitioners should be cautious when reporting clinical minimal changes.
