Myriam Blanchin1, Véronique Sébille2, Alice Guilleux2, Jean-Benoit Hardouin2. 1. EA 4275, Biostatistics, Pharmacoepidemiology and Subjective Measures in Health Sciences, University of Nantes, Nantes, France. myriam.blanchin@univ-nantes.fr. 2. EA 4275, Biostatistics, Pharmacoepidemiology and Subjective Measures in Health Sciences, University of Nantes, Nantes, France.
Abstract
PURPOSE: Statistical methods for identifying response shift (RS) at the individual level could be of great practical value in interpreting change in PRO data. Guttman errors (GE) may help to identify discrepancies in respondent's answers to items compared to an expected response pattern and to identify subgroups of patients that are more likely to present response shift. This study explores the benefits of using a GE-based method for RS detection at the subgroup and item levels. METHODS: The analysis was performed on the SatisQoL study. The number of GE was determined for each individual at each time of measurement (at baseline T0 and 6 months after discharge M6). Individuals showing discrepancies (with many GE) were suspected to interpret the items differently from the majority of the sample. Patients having a large number of GE at M6 only and not at T0 were assumed to present RS. Patients having a small number of GE at T0 and M6 were assumed to present no RS. The RespOnse Shift ALgorithm in Item response theory (ROSALI) was then applied on the whole sample and on both groups. RESULTS: Different types of RS (non-uniform recalibration, reprioritization) were more prevalent in the group composed of patients assumed to present RS based on GE. On the opposite, no RS was detected on patients having few GE. CONCLUSIONS: Guttman errors and item response theory models seem to be relevant tools to discriminate individuals affected by RS from the others at the item level.
PURPOSE: Statistical methods for identifying response shift (RS) at the individual level could be of great practical value in interpreting change in PRO data. Guttman errors (GE) may help to identify discrepancies in respondent's answers to items compared to an expected response pattern and to identify subgroups of patients that are more likely to present response shift. This study explores the benefits of using a GE-based method for RS detection at the subgroup and item levels. METHODS: The analysis was performed on the SatisQoL study. The number of GE was determined for each individual at each time of measurement (at baseline T0 and 6 months after discharge M6). Individuals showing discrepancies (with many GE) were suspected to interpret the items differently from the majority of the sample. Patients having a large number of GE at M6 only and not at T0 were assumed to present RS. Patients having a small number of GE at T0 and M6 were assumed to present no RS. The RespOnse Shift ALgorithm in Item response theory (ROSALI) was then applied on the whole sample and on both groups. RESULTS: Different types of RS (non-uniform recalibration, reprioritization) were more prevalent in the group composed of patients assumed to present RS based on GE. On the opposite, no RS was detected on patients having few GE. CONCLUSIONS: Guttman errors and item response theory models seem to be relevant tools to discriminate individuals affected by RS from the others at the item level.
Authors: Alice Guilleux; Myriam Blanchin; Antoine Vanier; Francis Guillemin; Bruno Falissard; Carolyn E Schwartz; Jean-Benoit Hardouin; Véronique Sébille Journal: Qual Life Res Date: 2014-12-05 Impact factor: 4.147
Authors: Christel M C van Leeuwen; Marcel W M Post; Lucas H V van der Woude; Sonja de Groot; Christof Smit; Dirk van Kuppevelt; Eline Lindeman Journal: Qual Life Res Date: 2011-11-30 Impact factor: 4.147
Authors: Estelina Ortega-Gómez; Purificación Vicente-Galindo; Helena Martín-Rodero; Purificación Galindo-Villardón Journal: Health Qual Life Outcomes Date: 2022-02-05 Impact factor: 3.186
Authors: Maxime Salmon; Myriam Blanchin; Christine Rotonda; Francis Guillemin; Véronique Sébille Journal: Cancer Med Date: 2017-10-10 Impact factor: 4.452