Background
Combination chemotherapy with cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) is the standard care for diffuse large B cell lymphoma (DLBCL), an aggressive, common form of non-Hodgkin lymphoma. In the last decade, four randomized controlled trials (RCTs) and two small observational studies demonstrated that the addition of the humanized monoclonal antibody rituximab to this combination (RCHOP) significantly improved the overall survival of patients undergoing primary treatment, although very elderly patients (≥80 years) were underrepresented [
1‐
7]. Our recent population-based study (n = 4,021) showed that RCHOP was associated with a significant increase in overall survival compared to CHOP in all ages, including ≥80 years, without evidence of any significant increase in serious toxicity detected [
8].
However, the high cost of rituximab brings its cost-effectiveness into question. This is problematic because cost-effectiveness information is a critical complement of comparative effectiveness research for producing efficient care and promoting fairness; it supports clinicians’ professional commitment to fair distribution of finite resources and helps health care payers and plans ensure value for money [
9,
10]. Economic models comparing RCHOP to CHOP have found RCHOP to be either a dominant strategy [
11], or a cost-effective alternative to CHOP [
12‐
15], but these models have relied on efficacy findings from RCTs and required assumptions regarding resource use since economic data were not prospectively collected. This is particularly relevant given the repeated demonstrations that patients who are eligible for RCTs are not representative of the wider population expected to use the treatment [
16]. While these economic models may be useful in informing coverage decisions, they may not represent the true cost-effectiveness of rituximab in practice. There remains a lack of evidence needed by payers to assess the extent to which the innovation is medically beneficial and financially sustainable for typical patients in routine clinical settings.
We evaluated the real-world cost-effectiveness of rituximab in patients with newly diagnosed DLBCL, using routinely collected widely available data. Our objective was to provide an assessment of value for money and accountability for spending on rituximab for DLBCL in practice from a population-based health care system’s perspective using administrative data on real world patients.
Discussion
Our overall results show that RCHOP for DLBCL was associated with a mean improvement in survival of approximately 3.2 months over a 5-year period but approximately $16,000 higher costs than standard CHOP chemotherapy, with an ICER of $62 K/LYG and a high probability of being cost-effective if the willingness-to-pay were at least $100 K for an extra year of life. However, cost-effectiveness decreased significantly with age, suggesting that the use of rituximab is not as economically attractive in the very elderly.
Our study had several strengths. First, our large population-based analysis included very elderly patients previously excluded from RCTs and young patients who were not captured in other databases such as Medicare. We also included a more comprehensive list of cost elements than previous cost-effectiveness studies [
11,
14], which allowed us to analyse the cost components with respect to age and time up to five years. Costs from this study are not only relevant to countries with a universal single-payer healthcare system similar to Ontario’s, but also to systems with multiple payers in which these healthcare costs would be distributed among private insurers, government-sponsored insurance, and patient out-of-pocket costs. Second, this study used administrative datasets exclusively, rather than prediction models [
11‐
14], to address the knowledge gap on the cost-effectiveness of rituximab for DLBCL in routine clinical practice. The results from this evaluation provide additional evidence needed to make or re-evaluate coverage decisions to ensure medical benefits, safety, and affordability of innovation. Third, we used a rigorous matching protocol to reduce bias [
20]. Finally, we applied IPW to account for censoring in the cost and survival data. Although there are guidelines for the statistical analysis of censored cost data, few studies apply them [
23].
There are several limitations to our study. First, the OCR for the study period did not contain stage data or full prognostic information (e.g. IPI score) for DLBCL patients, which are clinically useful predictors of survival outcomes that help guide treatment planning. We used ACG scores, a population-patient case-mix system, to estimate the burden of co-morbid illness, and used treatment intensity as a proxy for disease severity, but differences in treatment practices may lead to misclassification, although how this would bias the apparent incremental benefits and costs of rituximab is unclear. Since RCHOP is associated with improved survival compared to CHOP, it is possible that a CHOP patient who achieved the same number of cycles of therapy as his/her RCHOP match was actually healthier, and hence matching on treatment intensity could lead to estimates that might be biasing against rituximab. However, we expect this selection bias, if any, to be small. Second, outpatient prescription drug data were not available for most patients aged <65 years. However we expect minimal bias because we hard-matched the treatment groups by age. Third, we relied predominantly on Activity Level Reporting data to select our CHOP cohort, and therefore did not include patients from hospitals or clinics that did not submit data, potentially explaining the smaller size of the CHOP cohort before matching. However, our matched cohort was large, potentially improving its representativeness. Fourth, cost and survival benefits accumulated at a different rate in our study. While most costs were incurred in the first years after diagnosis, survival gains extend into later years. The ICER estimate is very sensitive to survival benefits, and it is possible that rituximab would be more favourable if the follow-up time was to extend beyond 5 years. Our approach measures, at best, a 5-year estimate. Finally, we did not use a contemporaneous cohort design due to rapid rituximab uptake post-approval. In fact, only a small subset (5-6%) of patients did not receive rituximab after 2005. These patients could be sicker, weaker, or have other health conditions, and using them as contemporary comparators could introduce unnecessary bias. With a historical cohort design, however, temporal improvement in patient management and differential censoring are challenges. To account for differential censoring, we limited our study time period and applied IPW to each treatment group separately. It is possible that recent widespread efforts in Ontario to shift end-of-life care from acute care settings to home care and community care centres [
24], and to shift complex continuing care from a lighter care residential model to active rehabilitation of more medically complex patients partially explain the higher home care and continuing care costs we detected among our very elderly RCHOP patients [
25], but these trends were not evident in the other age groups. Nonetheless, the costs we reported were the actual costs observed and we feel that our results represent valid estimates of the cost of care of RCHOP patients in the context of contemporary management for the period observed.
Compared to other studies that used same time horizons, our survival benefit from rituximab among young patients is lower than an Italian model (1.6 vs. 2.2 months at 3 years) [
11], but it only included patients with good prognosis [
3]. Our 5-year overall survival gain for elderly patients was similar to a study in British Columbia (BC) (0.2 vs 0.4 year) [
15], but much lower than the 1.04 years reported in a US study that extrapolated survival data from the European phase III GELA trial [
14]. Different modelling assumptions and extrapolation of trial data can generate a substantial variability in outcomes, highlighting the importance of validating findings with follow-up comparative effectiveness research such as this study.
While rituximab extended survival in all age groups, we found that its major impact on healthcare resource use was the reduction in hospitalization among patients <80 years old, especially for the youngest patients (<60 years). For the very elderly (≥80 years), however, RCHOP did not reduce hospitalization, while costs of other non-cancer resources significantly increased with age among RCHOP patients more than among CHOP patients, resulting in a high incremental cost for this age group. This is consistent with a recent Medicare study that reported more expensive non-chemotherapy-related and non-cancer related care among elderly rituximab patients as a result of longer survival [
26]. In our elderly rituximab patients, some of the additional costs were offset by the reduction in hospitalization, partially explaining our lower incremental cost than the Medicare study (4-year: $20 K vs. Medicare $28 K) (all values converted to 2009 Canadian dollars and rounded). Also that study only included patients >65 years old. In contrast to the Medicare study, our very elderly patients experienced an even more significant increase in non-chemotherapy and non-cancer costs, resulting in our higher incremental costs (4-year $37 K vs. $25 K), and suggesting rituximab is not cost-effective by standard thresholds (Medicare ICER: $60 K/LYG vs. our ICER: $114 K/LYG). This may be related to the fact that very elderly patients who received RCHOP had greater survival benefit than other age groups, and continued to incur more cost-intensive medical costs due to age and other conditions.
We found that real-world costs, incremental costs and cost-effectiveness ratios are higher than in published economic models and differ by age [
11,
14,
15]. For example, we did not observe lower costs in rescue therapy that could offset the high costs of rituximab to make it a cost-saving intervention for young patients, as projected by an Italian model [
11], or lower costs in palliative care for the elderly patients that could significantly reduce incremental cost, as described by the US model [
14]. These models, however, excluded key drivers of total and incremental costs such as the costs of hospitalization and prescription drugs. Compared to a British Columbia microsimulation, our 5-year incremental cost was comparable ($9 K vs. $10 K) for young patients [
15], but significantly higher for elderly patients ($19 K vs. $8 K). That study projected a reverse trend of incremental costs and ICER with age ($51 K/LYG for young patients and $21 K/LYG for the elderly) than what we observed, but its cost estimates were based on aggregated and literature-based data, and it did not observe a relationship between non-chemotherapy cost components and age as did our observational study. Variations in the ICERs found in these economic analyses are driven by different model assumptions.
Cost-effectiveness results are sensitive to study timeframe. Since most costs were incurred within the first years following DLBCL diagnosis, a longer study horizon resulted in a more economically attractive assessment because benefits extend into subsequent years. In fact, a follow-up study on the GELA trial showed that the survival benefits of the addition of rituximab to CHOP persisted over a 10-year follow-up [
27]. Our study’s goal was to highlight the usefulness of providing cost-effectiveness information alongside comparative effectiveness data that reflect routine clinical practice on representative patients, so we did not extrapolate beyond our data. Follow-up studies could examine the cost-effectiveness of rituximab over a longer time horizon and compare against findings in published models that used standardized methods for life-time projections of survival benefit and costs.
Competing interests
All authors have no conflict of interest, or financial or other relationships to declare that may influence or bias this work.
Authors’ contributions
SK participated in the design and coordination of the study, performed data and statistical analysis and drafted the manuscript. JB assisted in data analysis. MK participated in the design of the study and the interpretation of data. DH participated in the design of the study and the interpretation of data. LL participated in chart review and in the design of the analysis plan. MC participated in the design of the study and the interpretation of data. KEB participated in the design of the study and the analysis plan. JL performed data analysis. MM participated in the design of the study. CMB participated in the design of the study. CS participated in the design of the study. SG participated in the design of the study. TS participated in the design of the study. MT participated in the design of the study. SP participated in the design of the study. JSH conceived of the study, participated in its design and interpretation of data. All authors revised the article critically for important intellectual content, and provided approval of the final version.