We analyzed data from the patient population enrolled in the SELECT BC study that completed both the EQ-5D-3L and EORTC QLQ-C30 questionnaires at baseline. Because AEs were not assessed after the study treatment, we included only the EQ-5D-3L and EORTC QLQ-C30 measured before the end of the last course of the study treatment. We summarized profiles of health utility and HRQOL, and calculated completion rates, defined as the number of completed questionnaires divided by the number of expected responses, excluding patients after the study treatment. We examined predictive factors of incomplete assessments using a multivariable logistic generalized estimating equation.
For processing AE data, we chose 15 non-hematological AEs relevant to the treatment of metastatic breast cancer: febrile neutropenia, fever, fatigue, alopecia, allergy, diarrhea, oral mucositis, nausea, vomiting, anorexia, edema, motor neuropathy, sensory neuropathy, arthralgia, and myalgia. We then identified the last grade of each AE immediately before EQ-5D-3L and EORTC QLQ-C30 assessment (i.e. 3, 6, and 12 months). We counted the total number of incidences for each AE at three time points, and calculated the difference between incidence date of AEs and assessment date of the EQ-5D-3L and EORTC QLQ-C30. Finally, we selected AEs with ten or more incidences for subsequent analysis, meaning that we did not analyze AEs with fewer incidences because those estimates that essentially reflect the small number of assessments would be unstable and imprecise.
We used linear marginal mean models with time-dependent AEs to quantify the impact of each AE on health utility and HRQOL [27 ]. Analyses using the linear marginal mean model were adjusted for baseline scores, age, treatment, time, and treatment-by-time interaction (see Online Resource 1 for more details) [28 ]. First, we conducted separate analysis, where only a single analyzable AE (i.e. ten or more incidences) was included in the model. We then conducted simultaneous analysis, where all analyzable AEs were included in the models. Grades of AEs were modeled using linear and quadratic terms. We fitted the models using the generalized estimating equation method [27 ].
All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). All p value evaluations were two-sided, and a p value < 0.05 was considered nominally statistically significant without multiplicity adjustment.
Hagiwara Y., Shiroiwa T., Shimozuma K., Kawahara T., Uemura Y., Watanabe T., Taira N., Fukuda T., Ohashi Y, & Mukai H. (2017). Impact of Adverse Events on Health Utility and Health-Related Quality of Life in Patients Receiving First-Line Chemotherapy for Metastatic Breast Cancer: Results from the SELECT BC Study. Pharmacoeconomics, 36(2), 215-223.