Carl F Falk1, Felix Fischer2,3. 1. Department of Psychology, McGill University, 2001 McGill College, 7th Floor, Montreal, QC, H3A 1G1, Canada. carl.falk@mcgill.ca. 2. Charité - Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin and Humboldt-Universität zu Berlin, Department for Psychosomatic Medicine, Charitéplatz 1, 10117, Berlin, Germany. 3. Berlin Institute of Health at Charité - Universitätsmedizin Berlin, Clinical Study Center, German PROMIS® National Center, Charitéplatz 1, 10117, Berlin, Germany.
Abstract
PURPOSE: In developing item banks for patient reported outcomes (PROs), nonparametric techniques are often used for investigating empirical item response curves, whereas final banks usually use parsimonious parametric models. A flexible approach based on monotonic polynomials (MP) provides a compromise by modeling items with both complex and simpler response curves. This paper investigates the suitability of MPs to PRO data. METHOD: Using PROMIS Wave 1 data (N = 15,725) for Physical Function, we fitted an MP model and the graded response model (GRM). We compared both models in terms of overall model fit, latent trait estimates, and item/test information. We quantified possible GRM item misfit using approaches that compute discrepancies with the MP. Through simulations, we investigated the ability of the MP to perform well versus the GRM under identical data collection conditions. RESULTS: A likelihood ratio test (p < 0.001) and AIC (but not BIC) indicated better fit for the MP. Latent trait estimates and expected test scores were comparable between models, but we observed higher information for the MP in the lower range of physical functioning. Many items were flagged as possibly misfitting and simulations supported the performance of the MP. Yet discrepancies between the MP and GRM were small. CONCLUSION: The MP approach allows inclusion of items with complex response curves into PRO item banks. Information for the physical functioning item bank may be greater than originally thought for low levels of physical functioning. This may translate into small improvements if an MP approach is used.
PURPOSE: In developing item banks for patient reported outcomes (PROs), nonparametric techniques are often used for investigating empirical item response curves, whereas final banks usually use parsimonious parametric models. A flexible approach based on monotonic polynomials (MP) provides a compromise by modeling items with both complex and simpler response curves. This paper investigates the suitability of MPs to PRO data. METHOD: Using PROMIS Wave 1 data (N = 15,725) for Physical Function, we fitted an MP model and the graded response model (GRM). We compared both models in terms of overall model fit, latent trait estimates, and item/test information. We quantified possible GRM item misfit using approaches that compute discrepancies with the MP. Through simulations, we investigated the ability of the MP to perform well versus the GRM under identical data collection conditions. RESULTS: A likelihood ratio test (p < 0.001) and AIC (but not BIC) indicated better fit for the MP. Latent trait estimates and expected test scores were comparable between models, but we observed higher information for the MP in the lower range of physical functioning. Many items were flagged as possibly misfitting and simulations supported the performance of the MP. Yet discrepancies between the MP and GRM were small. CONCLUSION: The MP approach allows inclusion of items with complex response curves into PRO item banks. Information for the physical functioning item bank may be greater than originally thought for low levels of physical functioning. This may translate into small improvements if an MP approach is used.
Authors: Honghu Liu; David Cella; Richard Gershon; Jie Shen; Leo S Morales; William Riley; Ron D Hays Journal: J Clin Epidemiol Date: 2010-08-05 Impact factor: 6.437
Authors: Matthias Rose; Jakob B Bjorner; Barbara Gandek; Bonnie Bruce; James F Fries; John E Ware Journal: J Clin Epidemiol Date: 2014-05 Impact factor: 6.437