Transition probabilities used in the model were derived from the most up-to-date CLL11 clinical information available, corresponding to data cut-off of May 2015.17 The results of the latest data cut analyses showed GClb to be more effective than RClb, with a median PFS of 28.7 and 15.7 months for GClb and RClb, respectively, and a hazard ratio of 0.46 (95% confidence interval [CI] 0.38–0.55;P<0.0001). The time to next antileukemic treatment was 51.1 and 38.2 months for GClb and RClb, respectively, with a hazard ratio of 0.57 (95% CI: 0.44–0.74; P<0.0001).17
The probability of remaining in the PFS state was parameterized and extrapolated beyond CLL11 follow-up time using a Weibull distribution. The Weibull distribution was selected since it provided the best fit to the CLL11 PFS data, based on statistical tests such as the Akaike information criterion18and visual inspection. Sensitivity analysis was conducted using other statistical distributions: exponential, log-logistic, log-normal, gamma, and Gompertz.15 The probability of transiting from PFS to death was the maximum between the death rate during PFS in CLL11 and the Spanish age- and sex-specific mortality. The probability of moving from PFS to progression was then calculated as the complementary probability of remaining in the PFS state and of death from PFS.
Due to the immaturity of the overall survival (OS) data of the CLL11 trial at the time of data cut-off (median OS was not reached),17 the probability of transiting from progression to death was obtained from the CLL5, a Phase III clinical trial,19 which compared fludarabine versus Clb as the first-line therapy in patients with CLL. The probability of death after progression was based on an exponential function according to the Akaike information criterion and visual inspection. Age at progression was used as a covariate to account for age differences in the populations. Sensitivity analysis on this input was performed using the CLL8 trial, which compared FCR versus fludarabine and cyclophosphamide as the first-line therapy in patients with CLL.20 The projection of OS was then based on the extrapolated PFS and postprogression survival curves. Predicted survival curves are depicted in Figure 2.
(To view a larger version of Figure 2, click here.)
The health state utilities were obtained from a utility elicitation study involving 100 subjects in the UK, based on the time trade-off method (Table 1).21 The impact of being on/off treatment was accounted for. Differences in the administration route (intravenous [IV] or oral) and hospital visits associated with each treatment were also reflected in the utilities applied in the model. The utilities were used to adjust survival time and obtain the quality-adjusted life years (QALYs) for each of the regimens compared. Sensitivity analysis considered the 95% CI for each health state utility value.
The model considered direct health care costs only, which included drug acquisition, patient follow-up, the management of grade ≥3 AEs, and the cost of the IV administration of the medicines. All costs were obtained from Spanish sources and are reported in euros for the year 2016.22,23
The price considered for obinutuzumab in the base case was the reported ex-factory price (EFP). It should be acknowledged that the price of obinutuzumab is lower than the reported EFP, when charged to the Spanish NHS. This reimbursed price by the NHS is confidential and could not be used in this study. The price considered for rituximab corresponded to the EFP, whereas there is no lower confidential price of rituximab for the NHS. On the other hand, the price of Clb was not publicly available in Spain; therefore, this price was provided by the panel of experts (Table 1).
Sensitivity analyses on the acquisition prices charged to the NHS were carried out for both treatments. Regarding rituximab, a 15% reduction was applied to the EFP (base case) due to application of the Spanish Royal Decree-Law 9/2011 (RD).24 Concerning obinutuzumab, different acquisition cost levels (10% and 20% lower than the base case) were considered, in which case the RD deduction was also applied (4%; due to its status as an orphan drug) (Table 1).24
With regard to the doses of GClb and RClb, the model considered the mean doses recorded in the CLL11 trial, accounting for dropout, treatment discontinuation, or dose reduction, since this could be what might be expected in clinical practice. Maximum vial sharing was assumed. Sensitivity analysis was conducted using the doses and duration of each treatment as specified in the corresponding Summary of Product Characteristics (Table 1).