Treatment cost and life expectancy of diffuse large B-cell lymphoma (DLBCL): a discrete event simulation model on a UK population-based observational cohort

With an annual incidence of around 10.2 per 100,000 in adults, diffuse large B-cell lymphoma (DLBCL) is the commonest lymphoma subtype, accounting for around 40 % of the total [1, 2]. Although rapidly fatal if left untreated, DLBCL is potentially curable [3]. Introduced in the 1970s, chemotherapy with cyclophosphamide combined with doxorubicin, vincristine and prednisone (CHOP) [4] resulted in a response rate of around 60 % and a long-term cure rate of 30 % [5, 6]; the addition of the monoclonal antibody rituximab in the 1990s increased the latter to 45 % [7].

Hitherto, although several economic studies have been carried out, the majority have focused on comparing the cost-effectiveness of CHOP and R-CHOP (CHOP plus rituximab) [8‐17]. In addition, most of the relevant data has come from trials that only include patients treated with curative intent [9‐12, 14, 16]; making findings difficult to extrapolate to the general patient population.

The objective of this study was to develop an economic model that could (1) model across the whole treatment pathway, rather than being limited to first-line treatment or a specific agent alone, (2) reflect real world practice rather than the idealized predefined setting of a randomised controlled trial, and (3) predict medical costs and life expectancy. Such a predictive disease model is particularly important for evaluating the cost-effectiveness of new interventions and for allocating health resources efficiently. To the best of our knowledge, no such model has been previously developed for DLBCL.

This is the first DLBCL model to simulate and predict treatment costs and life-years gained throughout the treatment pathway. Whilst several economic models have previously been constructed and published for DLBCL, all were built for the purpose of assessing the cost-effectiveness of adding rituximab to CHOP, not for overall disease treatments [8‐10, 14‐16]. Hence, although these reports confirmed the fact that adding rituximab to CHOP was cost-effective, none could examine the overall economic impact to health insurers or policy makers. In addition, this is the first DLBCL model to incorporate data on patients who were not treated with curative intent; enabling the production of more accurate estimates of the economic impact of DLBCL among different patient groups, as well as across the patient population as a whole. This flexibility allowed quantities such as the prevalence-based economic impact at a national level to be explored for the first time.

In the current study, a discrete event simulation (DES) model was built for analysis. The DES model generated individual treatment histories within the set time horizon for hypothetical DLBCL patients based on ‘real-world’ observational data. Estimates for the 5-year expected medical cost and life-years gained were £18,096 and 2.8 years, respectively, while the lifetime expected medical cost and survival were £18,396 and 10.0 years, respectively. Curative treatment results in a high number of life-years gained at the relatively moderate cost of £1535 per life-year gained (95 % CI £1534–£1537 per life-year gained) compared to non-curative care only. The variability was driven largely by initial treatment, age at diagnosis, and whether the patient had an ASCT. The expected lifetime medical cost ranged from £79,131 to £2930, while the life expectancy ranged from 30 days to 19 years (Table 3).

With respect to appropriate time horizons, this study demonstrated that the differences in estimated costs derived from 5-year, 15-year, and lifetime horizons were relatively minor, reflecting the high response rate amongst treated patients. However, the differences in estimated life-years gained were considerable (Table 3); confirming, as has been suggested by others [14], that the lifetime horizon is the optimal approach as it allows the overall effects of treatment to be fully captured. Furthermore, as expected, patients younger than 70 years had better survival and therefore incurred higher medical costs (Fig. 3). Importantly, however, in the 5-year time frame the differences in costs and life-years gained between these two age groups were small, demonstrating that patients over 70 years who receive chemotherapy responded as well as their younger counterparts. In this context, it is important to note that a patient’s performance status has been shown to be more discriminatory of survival than chronological age [25]. Finally, although the average cost of treating DLBCL is considered moderate in comparison to some other cancers (£18,396, see Table 1), the annual economic impact of treating existing and new DLBCL patients in the UK is considerable (in the region of £88–£92 million, see Fig. 4); accounting for approximately 1/6th of the annual UK NHS budget for haematological diseases as a whole [26] and providing around 35,000 life-years gained per year (data not shown).

Predicated on ‘real-world’ data, this model produced findings that can be extrapolated to the general patient population; which is not the case for models built using data from clinical trials [9‐12, 14, 16]. The reliability and robustness of the model are also confirmed by the internal and external validations. For internal validation, the simulated cost and life-year results were close to the empirical HMRN data (Fig. 5). For external validation, the model results were in line with the findings from other observational studies: the micro-costing study conducted in Canada [17] and the Medicare claims study conducted in the US [16]. Moreover, the structure of our DES model provided an opportunity to make detailed comparisons with findings from more restricted datasets: and when the parameters (e.g. specific age groups, chemotherapy regimens and time horizons) from those studies were applied to the model, the results were found to be in line with most of the relevant studies [8, 10, 16, 17] (please refer to “Model validation”).

The molecular heterogeneity of DLBCL coupled with recent advances in diagnostic technologies [27, 28] is resulting in the development of more targeted approaches to the treatment of this complex cancer. The recent UK Phase 3 trial (REMoDL-B), for example, used gene expression profiling to assign DLBCL patients to different treatment arms [29]; and, as science continues to advance and new treatments emerge, reliable models such as the one developed here will become increasingly important in this rapidly changing field. Furthermore, the reduced side-effects and toxicities of many of these novel therapeutic agents mean that the proportion of the patient population who receive treatment will continue to increase. Accordingly, our model’s ability to accurately predict the impact of these changes across the entire population of patients with DLBCL will not only support commissioners to allocate resources, but will also aid clinical decision making.

With respect to model limitations, the current study did not deal with an administrative censoring effect. However, this is unlikely to impact on results, as the degree of administrative censoring is non-informative (independent of treatment). Also, the model is validated by the results which are broadly similar to the findings in earlier literature. In addition, whilst cost inputs were mainly confined to inpatient and day-case settings, the majority of costs will have been captured since around 85 % of cancer spending is incurred in acute/secondary care settings [26]. Moreover, as the model is based on an empirical follow-up time of only 5 years, the results for treated patients need to be interpreted with caution. Nonetheless, as the majority of treatments are carried out within the first 5 years, the costs estimates will be largely unaffected (see Table 3). It is, however, worth noting that our long-term estimates are conservative since they do not include health care costs for comorbidities that could potentially have been caused by DLBCL therapy. Quality adjusted life years (QALY) were not used in the model, as this information was not available for this cohort of patients and currently no other high quality QALY information is published in the existing literature, an issue recognised by other researchers [10, 15]. However, the model has been built to allow QALY data to be incorporated in the future and we are now collecting European Quality of Life-5 Dimensions (EQ-5D) information from patient cohort members.

Life expectancy and costs of treating DLBCL patients vary according to patient characteristics and treatment pathways. However, the population-based model developed in the current study demonstrated a good capability of capturing the medical costs to healthcare commissioners, as well as the life-years gained in a real world setting. Importantly, the model produces different outputs for different purposes; estimating total costs along with health benefits at varying time points for specific patient cohorts, as well as generating prevalence-based costs for all patients over specific time periods. Future application of the model could include evaluation of new technologies/treatments to support healthcare decision makers, especially in the era of personalised medicine.