Study Design
This was a post hoc pooled analysis (GSK study 206351) of patient-level data pooled from the six retrospective, observational cohort OBSErve studies from 97 study centers based in Argentina (GSK study 201282), Canada (GSK study 117300), Germany (GSK study 117214), Spain (GSK study 200883), Switzerland (GSK study 201232), and the USA (GSK study 117295) [11,12,13,14,15,16,17]. Follow-up times varied between the original studies (OBSErve Canada, Germany, Spain, and Switzerland: 6 months; OBSErve Argentina and USA: up to 24 months), for the current analysis, patient data collected at belimumab start (index date, between September 2012 and March 2016) and at 6 months after index, or at time of discontinuation, were analyzed, because this was the longest duration of follow-up time for which all countries could contribute data (Supplementary Fig. 1). As the studies were retrospective, study protocols had no influence on the physicians’ treatment decisions or collection of patient data.
Ethics committee/institutional review board approval was not required for this analysis, as it used previously published data. All procedures performed in the original OBSErve studies were in accordance with the ethical standards of the institutional and/or national research committee and with the Helsinki Declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the original OBSErve studies. The reporting of this study conforms to the CONsolidated Standards of Reporting Trials (CONSORT) 2010 guideline [18].
Physicians
Complete physician eligibility criteria for each individual OBSErve study have been published [11,12,13,14,15,16]. In brief, rheumatologists and/or internal medicine specialists who prescribed belimumab to patients with SLE as part of their usual clinical practice were invited to participate. Physicians extracted patient information about demographics, and disease and treatment characteristics. All patient data were anonymized.
Patients
The current analysis included all patients from the individual OBSErve studies. The following patient eligibility criteria were used in the primary studies: adults ≥ 18 years of age with a diagnosis of SLE, who were prescribed belimumab as part of standard SLE therapy, who initiated their belimumab therapy ≥ 6 months prior to study enrollment (Argentina, Germany, Spain, Switzerland), or had received ≥ 8 belimumab infusions during the ≥ 6 months prior to study enrollment (Canada, USA), and for whom reasons for belimumab initiation could be identified. Patients enrolled in an SLE clinical trial or who had initiated belimumab as part of a clinical trial were not included.
In OBSErve Argentina, Canada, Germany, Spain, and Switzerland, physicians examined their patient records to identify all patients with SLE receiving belimumab. To avoid selection bias, all patients who met the eligibility criteria (according to each country’s indication) were selected for inclusion. In OBSErve US, a random sample of eligible patients was selected by retrospective chart review of all available patients for each individual study center.
Endpoints and Assessments
The primary endpoint (physician-assessed overall clinical response to belimumab at month 6) was evaluated as worse, no improvement, < 20% (minimal) improvement, 20–49% (clear but moderate) improvement, 50–79% (large) improvement, or ≥ 80% (near normalized) improvement, compared with index.
Patient demographics (e.g., age, gender, country of enrollment), and clinical (disease severity and duration, number and type of SLE manifestations, number of comorbidities) and treatment characteristics (e.g., concomitant medications) at belimumab initiation were described as part of the secondary endpoint. Secondary endpoints (at month 6 vs. date of belimumab initiation) also included physician-assessed overall clinical response among patients in high disease activity subgroups; patient characteristics associated with a ≥ 50% improvement in physician-assessed overall clinical response to belimumab; change in SLE Disease Activity Index-2000 (SLEDAI-2K)/Safety of Estrogens in Lupus National Assessment-SLE Disease Activity Index (SELENA-SLEDAI) score (choice based on center preference for each individual patient); reasons for belimumab initiation and discontinuation (physician-reported) and changes in concomitant GCS use (i.e., mean change in dose, proportion of patients experiencing dose change, and proportion of patients achieving dose reduction to ≤ 7.5 mg/day (prednisone-equivalent); and reasons for belimumab initiation and discontinuation (physician-reported).
Exploratory endpoints were also analyzed at month 6, including physician-assessed overall clinical response and concomitant GCS use based on immunomodulator (azathioprine, cyclophosphamide, cyclosporine, methotrexate, mycophenolate mofetil, thalidomide, leflunomide, or tacrolimus) use at belimumab initiation; and physician-assessed improvement in specific clinical manifestations reported as present at belimumab initiation.
For the secondary endpoint of physician-assessed overall clinical response, patients were assigned to one of two high disease activity subgroups to reflect the indicated population in Europe [5]: Subgroup 1, patients with a SLEDAI-2K/SELENA-SLEDAI score ≥ 10; and Subgroup 2, patients with high anti-dsDNA or low complement (C3 and/or C4, < lower limit of normal [based on physician’s assessment of the medical record]).
Sample Size and Statistical Analysis
Pre-specified primary and secondary endpoints of the current study were selected based on consistent data reporting/availability across the original OBSErve studies. Thus, except for SLEDAI-2K/SELENA-SLEDAI scores, which are not routinely assessed in clinical practice, the extent of missing data was expected to be minimal. Where data were missing, it has been clearly indicated throughout the results, and no imputation was performed for missing data.
Descriptive statistics were used to summarize continuous data (mean, median, standard deviation [SD], 95% confidence intervals [CIs], minimum and maximum, and range) and categorical data (counts and frequencies). The overall clinical response to belimumab was compared between the patients exposed to one immunosuppressant (IM) versus those exposed to at least two IMs at belimumab initiation, using the Wilcoxon rank-sum test. A logistic regression model was fitted to estimate the predictors of ≥ 50% improvement (combined responses of 50–79% and ≥ 80% improvement) physician-assessed overall clinical response to belimumab at month 6. The response categories were dichotomized (≥ 50% or < 50% improvement) to facilitate interpretation by using a generalized linear model specifying a binomial distribution rather than ordinal regression incorporating multiple response levels; ≥ 50% was selected as this represents a large improvement and used to differentiate patients likely to derive a large benefit from belimumab treatment. The following covariates were included in the model based on clinical understanding of characteristics that influence treatment response: age (reference case: 18–27 years); gender (reference case: female); ethnicity (reference case: Caucasian/white); disease severity at belimumab initiation (reference case: mild); prior IM use (reference case: no prior use); SLE disease duration (reference case: ≤ 5 years); GCS dose at belimumab initiation (reference case: ≤ 7.5 mg/day); presence of SLE manifestations at belimumab initiation (reference case: no manifestations at belimumab initiation); presence of high anti-dsDNA and/or low complement at belimumab initiation based on physician judgment (reference case: no anti-dsDNA and/or high complement at belimumab initiation). SLEDAI-2K/SELENA-SLEDAI score was not included in the final model due to missing data (n = 484 patients at baseline), which would substantially reduce the number of observations that could be included. Prior to fitting the model, multicollinearity of covariates was assessed using the variance inflation factor (VIF), with VIF > 10 indicative of one covariate being a linear combination of other covariates. A general rule of one covariate for every 20 observations in the smallest response group was applied to eliminate overfitting the model. Odds ratios (OR) and p values were reported for each covariate included in the model. The Hosmer–Lemeshow test was used to assess whether the observed rates matched the expected rates in the population subgroups (i.e., how well the model predicted overall clinical response to belimumab treatment). A p value (< 0.05) indicated a poor fit.