296P - Item response theory and factor analysis as mean to characterize occurrence of response shift for longitudinal quality of life study in breast cance...

Date 01 October 2012
Event ESMO Congress 2012
Session Poster presentation III
Topics Breast Cancer, Early Stage
Psychosocial Aspects of Cancer
Presenter Amélie Anota
Authors A. Anota1, C. Bascoul-Mollevi2, F. Guillemin3, T. Conroy4, M. Velten5, D. Jolly6, M. Mercier7, S. Causeret8, T.S. Dabakuyo9, F. Bonnetain9
  • 1Unité Méthodologique Et De Qualité De Vie En Cancérologie, University hospital of Besançon, 25000 - Besançon/FR
  • 2Biostatistic Unit, Centre Val-d’Aurelle - Paul Lamarque, Montpellier/FR
  • 3Clinical Epidemiology And Evaluation Department, CIC-EC, 54035 - Nancy/FR
  • 4Medical Oncology, Centre Alexis Vautrin, FR-54511 - Vandoeuvre Les Nancy CEDEX/FR
  • 5Epidemiology And Public Health Laboratory, College of Medicine, 67085 - Strasbourg/FR
  • 6Création Du Pôle Recherche Et Innovations, University Hospital, 51092 - Reims/FR
  • 7Clinical Research Unit In Quality Of Life, CHU Jean Minjoz, 25030 - BESANCON/FR
  • 8Surgery, Centre Georges François Leclerc, 21000 - Dijon/FR
  • 9Biostatistic And Epidemiological Unit(ea 4184), Centre Georges François Leclerc, 21000 - Dijon/FR



Health-related quality of life (HRQoL) is a dynamic process which depends on the adaptation of the patient and reflected by a Response Shift (RS) effect. RS results in a recalibration, a reprioritization and a reconceptualization of key HRQoL domains. Longitudinal analyses of HRQoL have to take into account the possible occurrence of RS. However, there is no standard of statistical analysis to characterize RS. Two complementary methods are investigated to characterize RS.


This work builds on data of a prospective multicenter study including all primitive breast cancer patients or suspicion. HRQoL was evaluated using the EORTC QLQ-C30 and QLQ-BR23 at baseline, after surgery, at 3 months and 6 months, according to the « then-test/post-test » design: the retrospective assessments done after surgery and at 3 months refer to baseline HRQoL; the retrospective measurement done at 6 months refers to HRQoL at 3 months. The order then-test and post-test of HRQoL questionnaires was randomized. Recalibration was explored by Multiple Correspondence Analyses (MCA) and the Linear Logistic Model with Relaxed Assumptions (LLRA) of Item Response Theory (IRT). LLRA gives trend of item easiness parameters. Reprioritization and reconceptualization were explored by Principal Component Analyses (PCA).


Between February 2006 and February 2008, 381 patients were included, 90% had a confirmed breast cancer. PCA show a secondary reprioritization of the QLQ-C30's dimensions. Fatigue and pain remain priority symptoms. Secondary symptoms are insomnia at baseline, diarrhea after surgery, nausea and vomiting at 3 months and 6 months. A stronger and stronger link between functional scales reflects a reconceptualization. Main recalibration's profiles reflected by MCA are from one modality to an adjacent one. A lower or upward recalibration of each dimension is reflected by the IRT model. Based on retrospective assessment at six months of HRQoL at three months, arm and breast symptoms were overestimated with trend parameters equal to -0.59 and -1.05 (p < 0.001).


IRT models have mainly been used to validate HRQoL questionnaires. This work shows their interest to characterize occurrence of RS. Further analyses should be lead to validate their abilities to characterize all RS components.


All authors have declared no conflicts of interest.