Summary
The objective was to assess the medicoeconomic impact of initial and subsequent treatments based on oral fludarabine, intravenous chemotherapy (mini-CHOP) and chlorambucil in chronic lymphocytic leukaemia.
A Markov model has been defined to encompass the 18 strategies over a 3-year period after starting the first treatment. Costs of treatments, side effects and follow-up have been calculated by crossing data from published prospective trials, specific hospital databases and French resource-based relative value scales. When treatments were based on mini-CHOP, different possibilities offered for hospital care were taken into account.
Probalistic sensitivity analysis was performed.
Whatever the modality of hospital care for mini-CHOP, the strategies based on oral fludarabine as first-line treatment are not only more effective but are also cost effective and dominate other types of scenarios.
Fludarabine given orally should be preferred to mini-CHOP or chlorambucil as a first-line treatment for patients with chronic lymphocytic leukaemia.
Introduction
Chronic lymphocytic leukaemia (CLL), a common haematological malignancy in Western countries, mainly affects adults and elderly patients. The disease is an uncontrolled clonal proliferation/accumulation of B-lymphocytes in the blood and bone marrow, with frequent lymph node and spleen enlargement, and its progression leads to chronic disease. Although the treatments are generally effective, resulting in a reduction of the symptoms of disease and sometimes achieving long-term remissions including clonal clearance from the blood and bone marrow, cure of the disease remains elusive despite a tremendous escalation of treatments using new combinations of cytostatics, monoclonal antibodies and intensification with stem cell rescue. Moreover, these treatments have not improved overall survival and the question still remains whether they could modify the natural history of the disease. Recently, new prognostic factors have been recognised, which could help to develop modern strategies on a risk-adapted basis. However, these promising new approaches must be evaluated in prospective controlled studies before being recommended in clinical practice.
In CLL, the assessment of treatments is generally based on a set of short-term criteria (i.e. percentage and quality of responses, time to progression and time to next treatment). Overall survival is generally considered as a secondary endpoint, given the longer follow-up required. However, in all chronic diseases, including CLL, most patients are treated more than once, obscuring the interpretation of each treatment on long-term results. This last restriction applies to the results of large prospective trials, including patients with previously untreated CLL, comparing first-line treatments with fludarabine and chlorambucilCitation1 as well as fludarabine and cyclophosphamide, vincristine, prednisone plus doxorubicin (CHOP) or cyclophosphamide, doxorubicin plus prednisone (CAP) anthracycline-containing regimensCitation2. In addition, results of another large prospective controlled trial, not yet published, comparing chlorambucil and fludarabine either alone or associated with cyclophosphamide, have recently been made availableCitation3. Considering the results of these trials, it could be assumed that fludarabine provides some advantages over chlorambucil and anthracycline-containing regimens, specifically a better range of complete responses and longer time to progression (versus chlorambucil in the US and UK trials and versus CHOP and CAP in the French study). Moreover, as the side effects of fludarabine and CHOP differ, quality of life assessment based on the Q-TWIST method has been performed, which disclosed only small differences between the two regimensCitation4.
Beyond their efficacy on the abovementioned criteria, each of these treatments brings about a specific cost that has been seldom if ever assessed in the past. For instance, the practical question that the physician, patient and/or third-party payer has to resolve is to evaluate and compare the specific costs required to obtain a longer time to progression or a delay to the next treatment, together with the savings or cost of a possible second-line treatment. These facts should be taken into account while long-term impacts do not differ significantly.
This medicoeconomic approach has been retrospectively assessed on cohorts of patients with previously untreated CLL, taking advantage of an improved accountancy system in French hospitals. In addition, the specific costs of fludarabine-based treatments, significantly modified in most European countries since the availability of the oral formulation, have been assessed with this latter route of administration, which is of importance when comparing the medicoeconomic aspects of treatment decisions in CLL.
Methods
Strategies
Depending on the patient's age and referred physician practice, the most representative regimens prescribed in France for CLL as first- and second-line therapies in the early 2000s were chlorambucil, mini-CHOP and fludarabine.
A 3-year time horizon was applied from the beginning of treatment in order to minimise the risk of having to take into account a third-line treatment and as the best compromise between the necessity of completely capturing the impact of treatment according to the response duration and the robustness of data according to the median time to follow-up. This period of time usually encompasses the first-line treatment period, the median time to progression observed with any of these regimens and, therefore, a significant proportion of patients that required a second-line treatment. Alternatively, a longer period of time would have introduced more hazards in the patient's course and hence increased the complexity of the model.
Combining these events over the 3-year reference period leads to the definition of 18 scenarios (), depending on the choice of first- and second-line treatments. In some cases, no further treatment would be required owing to a response longer than the 3-year period, the patient's decision to accept a stable disease without treatment or death.
Table 1. First- and second-line treatment strategies evaluated
The model
A decision model combined with a Markov model was constructed to evaluate the cost effectiveness of 18 clinical strategies in the management of CLL. The structure of the Markov model is shown in .
Figure 1. Representation of the health states in the model.
At the beginning of the model, patients receive a first-line treatment for 6 months. After the first-line treatment, there are three possible paths:
1. Patients are refractory to the initial treatment and immediately receive a second-line treatment for 6 months;
2. Patients die (natural absorbing state);
3. Patients are in remission for a while then are in progression again.
If this third situation occurs, they receive a second-line treatment for 6 months. Some delay may occur between the first- and second-line treatments. This delay corresponds to the latent character of the disease. In fact, in clinical practice the decision regarding second-line treatment depends not only on the fact of being in progression or not, but also on the progression level. The health states of progression were then considered: progression to first-line treatment without second-line treatment and progression to first-line treatment with second-line treatment. After second-line treatment, three possibilities exist:
1. Patients can be refractory to the treatment;
2. They can die;
3. They can be in remission for a while then be in progression to second-line treatment.
Refractoriness and progression to the second-line treatment are considered as terminal states. A third-line treatment has not been taken into account. A 3-month cycle (corresponding to the frequency of follow-up of patients in remission) has been considered and a 3-year time horizon applied in order to minimise the risk of having to take into account a third-line treatment and as the best compromise between the necessity of completely capturing the impact of treatment according to the response duration and the robustness of data according to the median time to follow-up.
The probabilities of transition applied in the model result from analysis of the literature evaluating the efficacy of chemotherapies in terms of response to treatment according to the criteria of the National Cancer Institute, criteria which is now retained in CLL. Two clinical trials were also retained, both of which included patients never treated before, and compared first-line fludarabine versus chlorambucilCitation1 as well as fludarabine versus CHOP and CAPCitation2. For chlorambucil, the response rate was equal to 46%, the median time to progression was 14 month and median survival was 56 months. For mini-CHOP, the response rate was 71%, the median time to progression was 29 months and median survival was 67 months. For fludarabine, the parameters were 71%, 32 months and 69 months, respectively. The Deale method was used to estimate a probability from a rate. The transition probabilities used in the model are presented in .
Table 2. Transition probabilities
Effectiveness criterion
The effectiveness criterion was progression-free survival, as this is the one most common in oncology.
Costs
Three types of costs have been analysed:
• costs of treatment
• costs of side effects
• costs of medical follow-up for a patient in remission
Only direct costs (medical and non-medical) have been considered from the social security perspective.
Outpatient costs were assessed using the French Resource-Based Relative Value ScaleCitation5 and oncology drugs were assessed using the official tariff. To assess inpatient costs, the French hospitals information system (PMSI) was used. A 3-year follow-up hospital care cost (linked either to treatment courses or to side effects of treatments) was estimated using PMSI data of 206 patients from three French university hospitals (Centre Hospitalier Universitaire de Caen, Centre Hospitalier Universitaire de la Pitié Salpêtrière in Paris and Centre Hospitalier Universitaire de Lille). All CLL incident cases in 2000 were identified through diagnosis coding within PMSI. Finally, all hospital stays and hospital consultations requiring patient transportation have been taken into accountCitation6.
Cost of drugs
The costs of drugs are presented in Chlorambucil is administrated in an ambulatory setting over a 2-year continued treatment. Patient medical management is provided by a hospital physician at 3 months, 6 months and every 6 months thereafter (or five times for a patient in response).
Table 3. Cost of chemotherapy strategies
Administration of mini-CHOP requires hospitalisation. It has been assumed that the costs of anticancer treatment are limited to hospital costs (by 25%) as a consequence of price negotiations. Regarding the hospital stays necessary for treatment administration, analysis of the data provided by the three hospitals highlighted a significant presence of two diagnostic-related groups (DRGs) linked to chemotherapy administration: the first DRG (DRG681) corresponds to chemotherapy sessions (ambulatory care) and the second (DRG817) is a day hospitalisation. Hospital stays were estimated from the National cost data based on the ‘modified DRG” methods. In common practice, administration of mini-CHOP requires the application of a drug delivery system before the first course and its removal at the end of the six courses, which requires 2 days of hospitalisation. It was assumed that a patient is followed by a hospital physician for 15 days after the first cure.
Oral fludarabine is administrated in an ambulatory setting. Follow-up of patients by a hospital physician during courses of fludarabine was assumed to be performed at 1, 3 and 6 months.
Finally, it was considered that all hospital stays and hospital consultations imply costs of patient transportation.
Cost of side effects
Grade 1–2 side effects are supposed to be treated at home and are mainly related to infections requiring antibiotics (i.e. oral amoxicillin for 8 days), whereas grade 3–4 are treated in hospital. Using data from the three university hospitals for each CLL incident case, using the patient identification number, all subsequent hospital admissions after the initial treatment course are identified over a period of 3 years, allowing a direct assessment of costs ().
Table 4. Frequency and costs of adverse effects of treatment
Costs of follow-up
For follow-up of patients in remission, it was assumed, whatever the strategy considered, that a specialist sees a patient once a trimester. Biological follow-up is ensured by a blood count once a trimester.
Cost-effectiveness analysis
The conventional approach has been applied to examine the cost effectiveness of the alternative strategiesCitation8,Citation9. First, it was determined whether any strategies were simply dominated by other strategies (i.e. having both lower costs and greater therapeutic effects). Second, it was determined whether any strategy was dominated through the principles of extended dominance (i.e. whether linear combinations of other strategies can produce greater benefit at lower costs). Finally, among non-dominated treatment options, incremental cost-effectiveness ratios were calculated by comparing each option with the closest more costly and more effective intervention.
This process produced an efficiency frontier of increasingly more costly and more effective strategies ( and ). The slope of the frontier at any point reflects incremental cost effectiveness; the additional cost at which additional effects can occur. In this approach, any modification of the efficiency frontier reflects the impact of the introduction of a new strategy considered as positive when the efficiency frontier expands or shifts. In the case of an expansion of the efficiency frontier, the new strategy is cost effective to the alternative one, whereas in the case of a shift to the right of the efficiency frontier the new strategy is not only cost effective but is a dominant strategy defined as an improvement of the cost-effectiveness ratio.
Figure 2. Baseline cost-effectiveness results: chemotherapy sessions.
Figure 3. Baseline cost-effectiveness results: day hospitalisations.
Probabilistic sensitivity analysis
A probabilistic sensitivity analysis was conducted using a Bayesian second-order Monte Carlo simulation to evaluate the uncertainty in the assumptions and to model the distribution of effectiveness and costsCitation10,Citation11. A probability distribution was assigned to each estimate and assumption that were not fixed values in order to capture the possible range of outcomes. After the probability distributions were defined, the simulation simultaneously drew one value at a time from the feasible range for each parameter. The cost and benefit were then calculated from the specific values of the input parameters. This process was repeated 100,000 times, resulting in 100,000 unique estimates of effectiveness and costs, generating a range of possible values.
Probability distributions were selected based on the expected distributions of the underlying parameters for transition probabilities, hospital costs and resource utilisation. Probabilities were assumed to follow a beta distribution to reflect the normal distribution and a restriction to values between zero and one. Mean hospital costs were expected to follow a log-normal distribution, reflecting the long right tail and restriction to positive values according to the cost data in the French DRG (mean and variance costs). Mean costs of side effects were expected to follow a simulated distribution estimated by a non-parametric bootstrapping method using Monte Carlo simulation. A total of 10,000 sets of input parameters were randomly sampled.
Results
Cost-effectiveness analysis
The results of deterministic cost-effectiveness analyses for the situation in which mini-CHOP administration requires chemotherapy sessions are presented on the cost-effectiveness plane in and in for the situation in which mini-CHOP administration requires daycare hospitalisation. The cost effectiveness categorises the 18 chemotherapy strategies in terms of their health effect assessed by progression-free survival (on the x-axis) and their costs (y-axis), and reveals the dominated strategies and the cost-effective strategies making up the efficiency frontier.
When mini-CHOP is administered in the setting of chemotherapy sessions (), strategies S1, S3, S6, S7, S9, S11, S12, S15, S16 and S17 are dominated and have to be rejected from a medicoeconomic point of view. Strategies S2 and S4 (mini-CHOP as first-line treatment followed by mini-CHOP in case of progression and by oral fludarabine in case of refractory disease; mini-CHOP as first-line treatment followed by oral fludarabine in case of progression or refractory disease, respectively) seem to be equivalent.
The same situation occurs for S8 and S10 (oral fludarabine as first-line treatment followed by mini-CHOP if progression or refractory disease; oral fludarabine as first-line treatment followed by oral fludarabine in case of progression and by mini-CHOP in case of refractory disease) and for S14 and S15 (chlorambucil followed by chlorambucil in case of progression and by mini-CHOP in case of refractory disease; chlorambucil followed by chlorambucil in case of progression and by oral fludarabine in case of refractory disease).
Formally, the efficiency frontier is given by strategies S18 (chlorambucil with no further treatment in case of progression or refractory disease), S13 (chlorambucil following by chlorambucil in case of progression and no further treatment if refractory disease), S14 and S15 (chlorambucil followed by chlorambucil in case of progression and by mini-CHOP or oral fludarabine in case of refractory disease), S11 (oral fludarabine followed by chlorambucil in case of progression or refractory disease) and finally S8 and S10 (oral fludarabine following by mini-CHOP or oral fludarabine in case of progression and by mini-CHOP in case of refractory disease). It was noticed that S5 (mini-CHOP followed by chlorambucil if progression or refractory disease) could be considered to be located on the efficiency frontier.
The introduction of oral fludarabine as first-line treatment (strategies S8 and S10) results in a slight shift of the efficiency frontier to the right: oral fludarabine as first-line treatment followed by mini-CHOP or oral fludarabine in case of progression and by mini-CHOP in case of refractory disease are cost-effective alternative strategies.
In terms of the incremental cost-effectiveness ratio, each month free of progression implies a supplementary cost. It amounts to €347/month when strategies S13 and S18 are considered, €685/month considering S14 and S15 versus S13, €862/month considering S11 versus S14 and S15 and €1,353/month for S8–S10 versus S11.
To sum up, strategies S8 (oral fludarabine as first-line treatment followed by oral fludarabine in case of progression and by mini-CHOP in case of refractory disease) and S10 (oral fludarabine as first-line treatment followed by mini-CHOP in case of progression or refractory disease) are not only the most effective but are also cost effective.
When mini-CHOP is given during daycare hospitalisation, the classification between the strategies with or without oral fludarabine as first-line treatment is more significant (). Strategies S1, S3, S6, S7, S12, S16 and S17 are, as previously, still dominated and so have to be rejected from a medicoeconomic point of view. In addition, strategies S2, S4, S5, S9 and S14 also become dominated and once again have to be rejected from a medicoeconomic point of view. In this setting, no strategies starting with mini-CHOP are cost effective. Moreover, a strong shift to the right of the efficiency frontier was observed when oral fludarabine is taken into account. Indeed, strategy S11 is in the efficiency frontier instead of strategy S5. Strategies S2 and S4 are dominated by strategies S8 and S10.
The efficiency frontier is given by strategies with chlorambucil or oral fludarabine as first-line treatment. These strategies are S18, S13, S15, S11, S8 and S10 and they have been studied in pairs. The incremental cost effectiveness per month free of progression gained ratios was €347 when considering S13 versus S18, €703 when considering S15 versus S13, €832 when considering S11 versus S15 and €1,626 when considering S8–10 versus S11.
To sum up, no strategies starting with mini-CHOP are cost effective. The strategies S8 (oral fludarabine as first-line treatment followed by oral fludarabine in case of progression and by mini-CHOP in case of refractory disease) and S10 (oral fludarabine as first-line treatment followed by mini-CHOP in case of progression or refractory disease) are not only the most effective but also cost effective.
Sensitivity analysis
The results of the probabilistic sensitivity analysis are represented on the cost-effectiveness planes in two complementary ways: by the baseline estimate of the efficiency frontier and by circles for the probability that each strategy has to be in the cost-effectiveness frontier.
In the case of evaluation of costs of chemotherapy sessions needed for mini-CHOP administration, in comparison with the deterministic results, strategy S5 seems to be significant on the efficiency frontier (). In this case, the efficiency frontier is given by strategies S18 (probability = 1.000), S13 (probability = 0.729), S14 (probability = 0.394), S5 (probability = 0.423) and finally S8 and S10 (probability = 0.532 and 0.548, respectively). A probability >0.5 is considered as robust. In the case of chemotherapy sessions, the probabilistic sensitivity analysis only slightly modifies the structure of the efficiency frontier (S5 replaces S11). The robustness of the key result were confirmed: strategies S8 and S10 seem to be not only the most effective but also cost effective.
Figure 4. Probabilistic sensitivity analysis results: chemotherapy sessions.
In the case of evaluation of costs of hospital daycare needed for administering mini-CHOP, the probabilistic sensitivity analysis shows that the probabilities that cost-effective strategies in the deterministic analysis in the efficiency frontier (using strategies S18, S13, S15 and S8–S10) are ≥0.5, which confirms the robustness of the result. For other strategies, the probabilities are ≤0.20, demonstrating the weak robustness of the results (). The probabilistic sensitivity analysis only slightly modifies the structure of the efficiency frontier (S14 disappears). The robustness of the key result is confirmed: strategies S8 and S10 seem to be not only the most effective but also cost effective.
Figure 5. Probabilistic sensitivity analysis results: day hospitalisations.
Discussion
The aim of this study was to evaluate the medicoeconomic contribution of oral fludarabine, particularly as first-line treatment, in the management of CLL in the French context. It seems unrealistic to hope that randomised clinical trials directly comparing the strategies that were chosen could be implemented. In fact, the present clinical trials focus rather on the improvement of results by combinations of different agents. Taking into account the lack of relevant clinical studies, analysis of the therapeutic benefit and costs of 18 therapies has been conducted via an indirect manner of modelling. However, the model, although the simplest possible, is representative of the natural evolution of the disease on the one hand and of clinical practice on the other.
In this model, four types of efficacy parameters (level of response, survival, time to progression and time to second-line therapy) have been introduced in order to describe the clinical evolution of patients. In the framework of cost-effectiveness analyses, different cases have been tested. Two ways of cost calculation have been considered, depending on the type of hospitalisation necessary to administer mini-CHOP treatment (chemotherapy sessions lasting less than half a day or daycare hospitalisation <24 hours). The modelling has its limits. Assuming that the considered strategies are representative of clinical practice linked to the current recommendations of the French marketing authorisation for different treatments, the horizon of the model was fixed at a 3-year period. The choice of this time horizon, which was rather short, leads to an underestimation of the added value of the second-line treatments following mini-CHOP or fludarabine given the clinical effectiveness of these two regimens. Then, effectiveness of strategies in terms of survival was underestimated. This underestimation was, however, compensated by the assumption that the effectiveness of second-line treatment and first-line treatment was equal.
Regarding costs calculation, the method presented here has certain limits. Very simple conventions were applied, for example, the evaluation of transport costs linked to mini-CHOP courses. Another criticism can be related to the management of side effects of grade 3–4 patients, as follow-up of these patients by the private sector is not considered at any moment. With regard to the evaluation of costs of hospital stays necessary for the administration of mini-CHOP, two hypotheses were considered. The first one, rather conservative, assumes that the treatment is administered during a chemotherapy session; the second one considers a hospitalisation in the daycare system. Those two alternatives have been found in the databases provided by the three hospitals.
What may be more controversial is the fact of not taking into account the intangible costs linked to the better safety profile of oral fludarabine and its consequence in terms of patients' quality of life. Indeed, the positive impact of the administration of oral fludarabine is underestimated in this respect. In fact, the advantage of the oral form compared with intravenous administration consists of reducing the number of hospitalisations and decreasing the treatment burden. Side effects, regardless of their nature (haematological, clinical or others (i.e. nausea, fatigue, alopecia etc)) are less significant than those provoked by the intravenous administration of chemotherapy.
Conclusion
Considering the lack of randomised clinical trials, this study has enabled an indirect comparison of effectiveness and costs of 18 strategies of CLL management on the basis of data derived from existing literature.
If the most conservative hypothesis is applied, assuming that mini-CHOP requires only a half-day hospitalisation (chemotherapy session), it can be concluded that prescribing the oral form of fludarabine as first-line treatment associated in the second-line intention with mini-CHOP or oral fludarabine in case of progression and with mini-CHOP in case of refractory disease are not only the most effective strategies but are also cost effective. The willingness to pay for a month progression-free survival amounts to around €1,350.
If it is assumed that the mini-CHOP courses require a daycare hospitalisation (<24 hours), oral fludarabine as the first-line treatment (combined with mini-CHOP or oral fludarabine as second-line intention) remains a cost-effective strategy and dominates the strategies made up of mini-CHOP as the first-line treatment.
References
- Rai K R, Peterson B L, Appelbaum F R, et al. Fludarabine compared with chlorambucil as primary therapy for chronic lymphocytic leukemia. New England Journal of Medicine 2000; 343: 1750–1757
- Leporrier M, Chevret S, Cazin B, et al. French Cooperative Group on Chronic Lymphocytic Leukemia. Randomized comparison of fludarabine, CAP, and CHOP in 938 previously untreated stage B and C chronic lymphocytic leukemia patients. Blood 2001; 98: 2319–2325
- Catovsky D, Richards S, Hillmen P. Early results from LRF CLL4: a UK multicenter randomized trial. Blood 2001; 106: 212a, (abstract 716).
- Levy V, Porcher R, Delabarre F, et al. the French Cooperative CLL Group. Evaluating treatment strategies in chronic lymphocytic leukemia: use of quality-adjusted survival analysis. Journal of Clinical Epidemiology 2001; 54: 747–754
- Nomenclature Générale des Actes Professionnels – Nomenclature des Actes de Biologie Médicale. Union de Caisses Nationales de Sécurité Sociale.
- Selke B, Bercez C, Lenne X, et al. Adjuvant treatment of cancer of the breast: cost assessment of the protocol used. Bulletin du Cancer 2001; 85: 961–966, [in French]
- Rossi J F, van Hoof A, de Boeck K, et al. Efficacy and safety of oral fludarabine phosphate in previously untreated patients with chronic lymphocytic leukemia. Journal of Clinical Oncology 2001; 11: 1260–1267
- Weintein M C. From cost-effectiveness ratios to resource allocation: where to draw the line? In:. Valuing Health Care. FA Sloan. Cambridge University Press, New York, NY 1995; 77–97
- Drummond M, O'Brien B, Stoddart G, et al. Methods for the Economic Evaluation of Health Care Programmes, 2nd edn. Oxford University Press, New York, NY 1997
- Doubilet P, Begg C B, Weinstein M C, et al. Probabilistic sensitivity analysis using Monte Carlo simulation. Medical Decision Making 2001; 5: 155–177
- Pasta D J, Taylor J L, Henning J M. Probabilistic sensitivity analysis incorporating the bootstrap: an example comparing treatments for the eradication of Helicobacter pylori. Medical Decision Making 2001; 19: 353–363