Outcome measures and baseline predictors of response to belimumab treatment.

For the first part of the study, we will investigate specific items as predictors of response to treatment with belimumab. For this part, patients in the belimumab arms will be included in the analysis, and, most probably, the expert opinion will propose that only patients in the belimumab 10 mg/kg arm will be included in the analysis, as this dose received approval and is used in clinical practice. First, ROC-curve analysis could reveal whether the respective item is predictive of treatment response, and also help identify the optimal threshold value of the respective item showing best sensitivity and specificity for predicting the outcome. For example, ROC-curve analysis could result in the identification of an optimal cut-off baseline BLyS value. Then, we could proceed with predictive tests in order to calculate the positive and negative predictive values. The cut-off value can then be used for dichotomisation of the patients based on the baseline BLyS levels. This way, BLyS levels can be analysed in either logistic or Cox regression models (depending on whether the time to response should be part or the model according to the expert opinion and the statistician - this will be most probably dictated by how the dataset is built) not only as a continuous variable, which would only give implications of the value of baseline BLyS, but also as a categorical variable (below or under the optimal threshold). The latter gives a more straightforward message to the readership and can better be utilised by physicians treating patients with SLE. The smoking status is already a categorical variable, but could also be dichotomised as never smoker versus ever smokers, and never or former smokers versus current smokers. The organ damage will be categorised according to the expert opinion, based on relevant literature. For dichotomised variables, even the chi-square test might be used. In regression models, covariates will include age, sex, ethnicity, and other demographic characteristics, as well as clinical characteristics such as disease activity and disease duration. The dependent variables will be the clinical treatment outcome according to the SRI and possibly other measures, e.g. LLDAS and clinical SLEDAI=0.The LLDAS and clinical SLEDAI=0 will be applied to the dataset. In the second part of the study, all treatment arms (placebo, belimumab 1 mg/kg, and belimumab 10 mg/kg) will be used. First, response rates will be calculated using these outcome measures. To investigate whether the response rates differ between the different groups, we will make use of the non-parametric Mann-Whitney U-test for non-related samples. Then, we will compare SDI, LLDAS and clinical SLEDAI=0 in terms of reproducibility of the results shown in the BLISS-52 and BLISS-76 trials. This part is mostly descriptive and the interpretation of the results will need expert opinion. For the evaluation of the effects of drug therapies, simple statistical methods like the Mann-Whitney U test will be used in order to identify differences between groups. Logistic regression will be used when adjustments for possible confounding factors have to be made, as appropriate. The SAS software provided from the data share system or the IBM SPSS software will be used for the statistical analyses.