Cross-sectional nationwide mixed-methods population-based study of living conditions, and identification of sexual and fertility profiles among young women after breast cancer in France: the Candy study protocol

Quantitative component

Booklet response rates will be provided as well as the proportion of missing items for each questionnaire. Sexual function, HRQoL, alcohol consumption, anxiety and depression, social support and deprivation scores will be generated according to validated algorithms. They will be categorised and described in addition to the other quantitative variables as mean (SD) or median (range). Fertility data, professional situation of participants as well as other qualitative variables (clinical data, treatments, etc.) will be described as number and percentage. Using scores of the six subscales of sexual function, sexual function profiles will be identified by ascending hierarchical classification, and fertility profiles will be identified by latent class models. A generalised linear mixed model will be constructed to characterise fertility and sexual function profiles as well as to identify the clinical and socioeconomic determinants of HRQoL in young women. This modelling will take into account the date of differential diagnosis between participants, the department effect as well as a possible process of missing-not-at-random (MNAR) data by adjusting for the year of diagnosis and the non-random missing data. In univariate analysis, variables to test as predictors for sexuality will be, among others, age at the time of the study, time since diagnosis, tumour stage, Charlson comorbidity index, hormone receptor status, surgery (lumpectomy, mastectomy, breast reconstruction), ovarian suppression, oophorectomy, endocrine therapy, radiotherapy, anxiety, depression, body image and current partner relationship. For fertility profiles, variables will include age at the time of the study, education, employment status, having children before diagnosis, current partner relationship, time since diagnosis, mastectomy, radiotherapy, chemotherapy, targeted therapy, current endocrine therapy, menstrual cycles before and after treatments, fertility preservation and desire for children at diagnosis. Dependent variables to be tested in the univariate model for each dimension of HRQoL will include age at the time of the study, time since diagnosis, Body Mass Index, anxiety, depression, deprivation, sexual function, social support availability, social support satisfaction, tumour stage, tumour grade, hormone receptor status, HER2 status, KI67 index, Charlson comorbidity index, surgery, chemotherapy, radiotherapy, endocrine therapy, targeted therapy, current partner relationship, having children, employment status and education. Correlations and interactions will be tested for eligible variables. The variables eligible for multivariate analyses will be those with a p value<0.10 by univariate analysis. Correlations and interactions will be tested for eligible variables. Results will be reported as multivariate analysis coefficients, SDs and p values. Because dimensions of the FSFI, SF-12, EORTC QLQ-C30, EORTC QLQ-BR23 and EORTC QLQ-INFO25 questionnaires cannot be considered independent of each other, Bonferroni correction will be applied to adjust the α risk according to the number of dimensions analysed (α'= α/n with n corresponding to the number of dimensions analysed) for each self-report questionnaire. Multiple imputation by chained equation (MICE) will be used in the event of missing-at-random (MAR) data. Before performing the MICE, an analysis of the observed data will make it possible to define MAR mechanisms depending on certain variables, and thus to anticipate variations between observed and imputed data. A graphical comparison making it possible to detect faults in the superposition of the observed and imputed distributions will be performed. This step is crucial to ensure both the validity of the MICE model used and the plausibility of the MAR hypothesis. After MICE, the Wald test will be approximated by a Student test to test the regression coefficients. A Fisher test will also make it possible to jointly test a series of regression coefficients on all the imputed bases. In addition, sensitivity analysis will be performed to take into account any missing-not-at-random (MNAR data and to assess the impact of a MNAR mechanism on the results of multiple imputation. Statistical analysis will be performed with R and SAS software V.9.4 (SAS Institute, Cary, NC, USA).

留言 (0)

沒有登入
gif