Cost–effectiveness of obinutuzumab plus bendamustine in Chinese patients with relapse and refractory follicular lymphoma

The model cohort in this cost–effectiveness analysis was defined as Chinese patients with rrFL. The decision analytic model was constructed using the Markov framework with three health states, including PFS, post-progression survival (PPS), and death. The model simulated the cohort under treatments with O-B-O and the selected conventional therapies. The simulation time horizon was set to a lifetime, with each simulation cycle set to one month. The model outputs, including life years, quality-adjusted life years (QALY), and medical costs, were discounted at 5% per year. The structure of the constructed cost–effectiveness model is illustrated in Figure 1.

The constructed cost–effectiveness model was used to guide the search for appropriate evidence to estimate the model inputs. However, obinutuzumab is a relatively new treatment, and there are limited clinical studies directly comparing O-B-O with the conventional therapies examined in our study for clinical effectiveness and toxicity. As a result, this study was unable to estimate the differences in effectiveness and toxicity between O-B-O and the conventional therapies used for rrFL [4]. To address this limitation, we conducted a systematic literature search of clinical studies published in English or Chinese from 2011 to 2021 that assessed O-B-O and/or any of the six conventional therapies examined in this study. Subsequently, a single-arm meta-analysis was conducted to synthesize the reported PFS, overall survival (OS), and serious adverse events of the treatment methods from the identified studies [5–10]. To better simulate the survival outcomes associated with O-B-O and the six reference treatments in the model, this study used the plotted Kaplan–Meier (KM) survival curves for PFS from the phase III clinical trial (GADOLIN) comparing O-B-O with B-mono in patients with rrFL to identify the most appropriate survival function from six survival functions (Exponential distribution, Weibull distribution, log-logistic distribution, log-normal distribution, generalized gamma distribution, and Gompertz distribution). Based on the Akaike Information Criterion (AIC), the predicted average PFS over lifetime simulation, and the 10-year PFS rate, Weibull distribution was likely to be the most appropriate survival function to fit the observed PFS associated with O-B-O and B-mono. Thus, Weibull distribution was selected to simulate PFS associated with O-B-O and B-mono in the model. Exponential distribution was used to simulate PFS associated with other reference treatments and OS associated with all treatments in the model due to the lack of plotted KM survival curves in the identified clinical trials.

The data source for the utility values of the three health states in the cost–effectiveness model was another study assessing the cost–effectiveness of O-B-O relative to B-mono in Norwegian patients with rrFL [11]. However, because the GADOLIN trial contained data mainly from European countries and didn't assess the impact of age and gender on quality of life, the quality of life data of the Chinese population by age and gender was applied in the model to adjust utility associated with health states over lifetime simulation according to the demographics of the model cohort [12].

The data sources for cost inputs in the model included the latest drug reimbursement bidding prices, patient survey (for administration costs of induction therapy for rrFL in the hospital setting, medical costs per outpatient visit, outpatient visit frequency, and average hospital costs per admission for progressive disease), and physician survey (for hospital admissions in the last year prior to death and health resource utilization associated with managing the reported serious adverse events by treatments). These data sources were used to estimate the average medical costs associated with health states for each model cycle over lifetime simulation. The treatment costs included the drug acquisition costs based on treatment dosage, treatment administration frequency, and treatment cycles or duration, treatment administration costs, and adverse events management costs based on the risk of adverse events by types and medical costs managing each type adverse event. Disease management costs were estimated according to the health resources utilization and unit costs of utilized health resources for PFS and post-progression survival (PPS), respectively. The estimations of the model inputs are summarized in Table 1.

To adhere to clinical guidelines and practices, we constructed a cost–effectiveness model based on the following assumptions. Firstly, the model assumed that maintenance therapy associated with the treatment scenarios of interest would continue until the disease progressed. Secondly, we assumed that maintenance therapy requiring intravenous administration would be administered in a day ward, which is widely implemented in tier III hospitals across China to facilitate cancer treatment and reduce medical costs. Additionally, the cost–effectiveness model only considered the medical costs resulting from the management of serious adverse events, as mild or moderate adverse events usually do not require medical attention in real-world settings.

Since the patients in the clinical studies referenced by our study received subsequent treatments after disease progression, the cost–effectiveness model assumed that the survival outcomes of subsequent treatments for progressive disease were accounted for in the post-progression survival estimated in these clinical studies. Therefore, our cost–effectiveness model did not separately simulate survival outcomes but included the medical costs associated with subsequent treatments in patients with progressive disease under the treatment scenarios.

The constructed decision analytic model was used to assess the cost–effectiveness of O-B-O relative to conventional therapies for rrFL in Chinese patients. The cost–effectiveness analyses included a base case analysis, one-way sensitivity analysis (OWSA), and probabilistic sensitivity analysis (PSA). The base case analysis used the model inputs' baseline values to estimate the incremental cost–effectiveness ratio per gained QALY (ICER). Two treatment regimens, BR and R2, were selected to conduct OWSA and PSA to assess the uncertainty of O-B-O's cost–effectiveness. OWSA was conducted by varying the model inputs within the 95% confidence interval (CI) or the range of +/-25% of the baseline value if a 95% CI was lacking. PSA was performed using the decision analytic model with the distributions of the selected model inputs with a 95% CI. Ten thousand Monte Carlo simulations were performed to plot the distribution of the generated ICER and calculate the median and 95% credible intervals of the ICER for O-B-O relative to BR and R2. cost–effectiveness acceptability curves were plotted for O-B-O relative to the selected treatment regimens under varying willingness-to-pay to gain one additional QALY. To aid in interpreting the results of the cost–effectiveness analysis under the recommended cost–effectiveness threshold in China (triple the 2022 China gross domestic products per capita [GDPPC]), all ICER results were converted to a ratio to the 2022 China GDPPC (¥85,698 or US$12,743 based on the 2022 average exchange rate [$1: ¥6.725]). The statistical software R (v 4.1.0) was used for common evidence synthesis methods to estimate model variables. Microsoft Excel (version 16.63) was used to construct the decision analytic model and perform the cost–effectiveness analyses, including base case analyses, OWSA, and PSA.

According to the simulation, O-B-O was associated with the longest PFS (2.965 years) and the fourth-longest PPS (3.450 years) among the included treatment regimens, just after R-mono (4.358 years), Lenalidomide (4.228 years), and R2 (3.758 years). The simulation of OS indicated that O-B-O could add the most years to a patient's OS (6.415 years), and B-mono could add the least duration to a patient's OS (4.454 years) among the treatment regimens. The simulated QALY associated with the treatment regimens in the model displayed a similar trend as the simulated survival outcomes. Of the included treatment regimens, O-B-O was associated with the most QALY derived from PFS (2.471 QALY) and the most QALY gained (5.100 QALY). The QALY gained associated with B-mono (3.535 QALY) was ranked the lowest among the included treatment regimens. The lifetime direct medical costs simulated in the model were classified by PFS and PPS. O-B-O was ranked third for the cumulative direct medical costs associated with PFS (¥252,684) but in the bottom for the cumulative direct medical costs associated with PPS (¥288,495). O-B-O was ranked fourth for the overall lifetime direct medical costs among the included treatment regimens in the model. The survival outcomes, QALY, and lifetime direct medical costs associated with the included treatment regimens simulated by the model are summarized in Table 2.

The base case analyses indicated that O-B-O was superior in cost–effectiveness over R2 and L-mono by producing more QALY gained but costing less for Chinese patients with rrFL. In addition, O-B-O indicated highly attractive cost–effectiveness over the other four treatment regimens (B-mono: ¥76,859 or 0.897 GDPPC; BR: ¥38,331 or 0.447 GDDPC; R-mono: ¥87,620 or 1.022 GDPPC; R-CHOP: ¥54,463 or 0.636 GDPPC) with an ICER under 2 GDPPC. The results of the base case analyses are illustrated in Figure 2.

The ICER of O-B-O, compared with the six conventional therapies in OWSA, varied substantially. The cost–effectiveness of O-B-O relative to BR was driven by key model inputs, including subsequent treatment acquisition costs associated with progressive disease (ICER varied from -¥6676 to ¥83,338), the median OS associated with O-B-O (ICER varied from ¥24,210 to ¥71,566), and the subsequent treatment rate for progressive disease (ICER varied from ¥22,611 to ¥65,193) (Figure 3A). For the cost–effectiveness of O-B-O relative to R2, only median PFS associated with R2 (ICER varied from -¥5,969,345 to ¥1,187,554), median OS associated with O-B-O (ICER varied from -¥381,641 to ¥507,718), and median OS associated with R2 (ICER varied from -¥374,533 to ¥526,159) could drive the ICER above $0 (Figure 3B). Median OS associated with O-B-O (ICER varied from ¥47,166 to ¥314,826) and R-mono (ICER varied from ¥46,710 to ¥281,472) were the key model inputs driving the cost–effectiveness of O-B-O relative to R-mono (Figure 3C). Only median PFS associated with L-mono could drive the ICER of O-B-O relative to L-mono above ¥0 (ICER varied from -¥193,361 to ¥2189) (Figure 3D). The cost–effectiveness of O-B-O relative to B-mono was driven by median OS associated with O-B-O (ICER varied from ¥54,956 to ¥121,739) and subsequent treatment acquisition costs associated with B-mono (ICER varied from ¥39,613 to ¥114,105) and O-B-O (ICER varied from ¥44,312 to ¥109,406) (Figure 3E). Similarly, the cost–effectiveness of O-B-O relative to R-CHOP was driven by median OS associated with O-B-O (ICER varied from ¥32,267 to ¥123,722) and subsequent treatment acquisition costs associated with R-CHOP (ICER varied from -¥3431 to ¥112,357) and O-B-O (ICER varied from ¥4962 to ¥103,964) (Figure 3F).

The 10,000 Monte Carlo simulations for the cost–effectiveness of O-B-O relative to the six conventional therapies for rrFL were used to identify median and 95% credible intervals of ICER and the cost–effectiveness proportions under the cost–effectiveness thresholds as of 1, 2, and three-times of GDPPC (Table 3). Based on these PSA, the cost–effectiveness proportions of O-B-O relative to all six included conventional therapies were above 50% under the guideline recommended cost–effectiveness threshold (3 GDPPC). When the cost–effectiveness threshold reduced to 1 GDPPC, the cost–effectiveness proportions of O-B-O still remained above 50% for the comparisons with BR (93.7%), R2 (59.9%), L-mono (94.7%), B-mono (61.7%) and R-CHOP (79.0%).