Cost-effectiveness of 3-year vs 1-year adjuvant therapy with imatinib in patients with high risk of gastrointestinal stromal tumour recurrence in the Netherlands; a modelling study alongside the SSGXVIII/AIO trial

Abstract

Background:

Surgical resection of gastrointestinal stromal tumour (GIST) is rarely curative in patients at high risk of tumour recurrence and therefore 1 year of post-surgery adjuvant imatinib therapy has been recommended in this sub-group. Recently, adjuvant imatinib therapy administered for 3 years has been demonstrated to further increase recurrence-free survival and overall survival. The goal of this study was to assess the economic value of extending the duration of adjuvant imatinib therapy in high-risk patients in the Netherlands.

Methods:

A multistate Markov model was developed to simulate how patients’ clinical status after GIST excision evolves over time until death. The model structure encompassed four primary health states: free of recurrence, first GIST recurrence, second GIST recurrence, and death. Transition probabilities between the health states, data on medical care costs, and quality-of-life were obtained from published sources and from expert opinion.

Results:

The expected number of life years (or quality-adjusted life years, QALYs) was higher in the 3-year group than in the 1-year group, 8.91 (6.55) and 7.04 (5.18) years, respectively. In the 3-year and 1-year group, the expected total costs amounted to €120,195 and €79,361, of which, €74,631 (62%) and €27,619 (35%) were adjuvant therapy drug costs, respectively. The difference in health benefits, that is 1.87 life years or 1.37 QALYs, and costs, €40,835, resulted in incremental cost-effectiveness ratios (ICER) of €21,865 per life year gained, and €29,872 per QALY gained.

Limitations:

A limitation of the study was inherently related to the uncertainty around the predictions of RFS. Scenario analyses were conducted to test the sensitivity of different RFS predictions on the results.

Conclusions:

Delayed recurrence due to treatment with longer-term adjuvant imatinib therapy represents a cost-effective treatment option with an ICER below the generally accepted threshold in the Netherlands.

Keywords::

• Imatinib
• Gastrointestinal stromal tumour
• Cancer
• Reimbursement
• Cost-effectiveness analysis
• Decision-making

Introduction

GIST and risk of recurrence

Gastrointestinal stromal tumours (GISTs) are relatively rare tumours of the gastrointestinal tract that account for less than 1% of all the gastrointestinal tumours1–3. The annual incidence is between 10–20 per million people, with a median age of 60–65 at the time of diagnosis4–6. GISTs occur in both genders at a similar rate, but some studies show male predominance7. Most GISTs, 75–90%, are associated with gain-of-function mutations in the KIT gene8–12, whereas a smaller proportion, 5–10%, are associated with analogous mutations in the gene that encodes platelet derived growth factor receptor-α (PDGFRα)13. Approximately 10% of the GISTs contain no identified receptor tyrosine kinase mutations. These molecular features are important for pathogenesis and are the driving force behind the development of GIST14,15.

In current clinical practice the mainstay of treatment for primary localized GIST is complete surgical excision16,17. Unfortunately, tumour resection is rarely curative and, as a result, many patients face a significant risk of recurrence or metastatic disease. The risk of GIST recurrence may differ for patients and depends on a number of tumour features that can be assessed after surgery. To categorize patients according to the risk of recurrence, several risk classification systems have been developed18. The currently most accepted and adopted classification system was put forward by Miettinen and Lasota14, who categorized patients according to three predictors of recurrence-free survival (RFS): tumour size, mitotic count, and tumour location. Recent guidelines17,19, including the Dutch version20, recommend post-surgery assessment of whether a patient has a very low, low, moderate, or high risk of recurrence based on the Miettinen risk classification system.

RFS 5 years after surgery is fairly high (94%) among those who have low risk of GIST recurrence. However, RFS is relatively low in high risk patients, i.e. patients with a combination of large tumour size, high mitotic count, and a particular tumour location (see Supplementary Appendix A for the complete risk classification table). In particular, it has been documented that ∼80% of the patients in this group experience tumour recurrence within 5 years after surgery, and eventually will die from the disease18. Consequently, because surgery alone is not sufficient in high risk patients, and GISTs are resistant to traditional non-selective chemotherapy21,22, there has been an unmet need for effective adjuvant treatment.

Adjuvant therapy in high risk patients

Understanding the molecular pathophysiology of GIST facilitated the development of medical treatments that specifically target the signalling abnormalities in the cancer cell. Imatinib mesylate (Glivec, Novartis) was the first selective inhibitor of the kinase activities of KIT that has been approved by regulatory bodies such as the European Medicines Agency (EMA) and the Food and Drug Administration (FDA). It is currently licensed for the treatment of adult patients who are at significant (i.e. moderate or high) risk of relapse following resection of KIT-positive GIST23.

The clinical benefit of adjuvant imatinib therapy has been demonstrated in two phase III, multi-centre, double-blind, randomized trials. In the placebo-controlled Z9001 study24, conducted by the American College of Surgeons Oncology Group (ACOSOG), the efficacy of once daily 400 mg imatinib given for 1 year following localized primary GIST removal was assessed. At a median follow-up of 19.7 months, the 12-month RFS in high risk patients was 98.7% in the imatinib-treated group and 56.1% in the placebo-treated group (p-value <0.0001, n = 165)25. Based on the results of this trial, a number of countries, including Scotland26, adapted clinical recommendations for the use of adjuvant imatinib therapy in high-risk patients. The results of the Z9001 study, however, did not demonstrate whether the benefits in RFS can be translated into benefits in overall survival (OS) in the long run by longer-term adjuvant imatinib therapy.

To investigate the efficacy and safety of longer-term adjuvant imatinib therapy, a second, large, phase III trial was conducted by the Scandinavian Sarcoma Group and the Sarcoma Group of the Arbeitsgemeinschaft Internistische Onkologie (SSGXVIII/AIO) trial27. In this trial all patients, including those at high risk (n = 281, 70% of the total patient population), were given once daily 400 mg adjuvant imatinib therapy for either 1 year or for 3 years. Results of the trial conclusively demonstrated that patients receiving 3 years of adjuvant imatinib therapy not only had higher RFS (hazard ratio [HR] = 0.43, p-value <0.0001) but also had higher overall survival (HR = 0.39, p-value = 0.007). Besides the SSG XVIII/AIO trial, two further non-randomized, controlled trials reported OS benefits for long-term adjuvant imatinib therapy in high-risk GIST patients28,29.

Goal of the study

The SSGXVIII/AIO trial was the first study that demonstrated a statistically significant OS benefit for adjuvant imatinib therapy in patients with GIST. The difference of 10% in OS over a 5-year time horizon is noteworthy compared with the benefit of other adjuvant therapies for solid tumours30,31. However, evidence on efficacy and safety per se are no longer sufficient for reimbursement; substantiation of favourable economic profile is also required32. That is, providing information on cost-effectiveness or cost-utility is crucial before healthcare resources can be allocated by policy-makers.

The goal of our study was to assess the value of 3 years vs 1 year of adjuvant imatinib therapy in patients with high risk of GIST recurrence according to the SSGXVIII/AIO trial in the long run. For this reason, a decision analytic model has been built to compare the long-term benefits received and the costs incurred from the Dutch healthcare provider perspective. The model has been developed following the best practice recommendations of the International Society for Pharmacoeconomics and Outcomes Research (ISPOR)33,34 and was repeatedly validated by Dutch oncologists.

Patients and methods

Model structure and health states

A multi-state Markov model has been built to simulate the life course of GIST patients, that is, to evaluate how their clinical status evolves over time in each month until death. Lifetime horizon was chosen to capture all relevant outcomes and costs. Two hypothetical cohorts were compared: patients with high risk of tumour recurrence according to the Miettinen criteria assigned to receive either 1 year of adjuvant imatinib therapy or 3 years of adjuvant imatinib therapy. The model was developed to reflect patient characteristics of the SSGXVIII/AIO trial, thus patients entered the model at an age of 61. Costs and quality adjusted life years (QALYs) were discounted by 4% and 1.5%, respectively, as recommended by the Dutch Health Care Insurance Board (CvZ)35. The model was developed in Excel 2010.

The structure of the model was based on the clinical practice guideline of the European Society for Medical Oncology17 and on Dutch daily clinical practice. It encompassed four primary groups of health states: free of recurrence, first GIST recurrence, second GIST recurrence, and death (from GIST or other causes). These primary groups were combined with a description of the treatment: once daily 400 mg adjuvant imatinib therapy, once daily 400 mg imatinib therapy for recurrence, daily 800 mg imatinib therapy for recurrence, sunitinib therapy for recurrence, and best supportive care. Figure 1 illustrates the model structure showing the health states and the possible transitions between them.

Figure 1. Markov decision analytic model simulating the history of a cohort of patients with high risk of gastrointestinal stromal tumour recurrence. The figure depicts heath states during adjuvant imatinib therapy and for two possible treatment lines after tumour recurrence. The model structure reflects the clinical recommendations of the European Society for Medical Oncology, the clinical practice in the Netherlands, and has been validated by oncologists. At baseline, patients differ in terms of the length of adjuvant imatinib therapy (1 vs 3 years) that they are assigned to receive (A1 health state). No further differences in patients were assumed.

All patients entered the model free of recurrence at initiation of once daily 400 mg adjuvant imatinib therapy. After the fixed treatment period, that is a maximum of 1 or 3 years (corresponding to the two treatment groups) had elapsed, patients completed the adjuvant therapy. Once these patients experienced tumour recurrence they either re-initiated once daily 400 mg imatinib therapy or received sunitinib therapy, depending on whether recurrence occurred sooner or later than 3 months after completion of the adjuvant therapy. Patients could stop adjuvant therapy due to adverse events (AEs) and remain recurrence-free. If, at some point, these patients experienced tumour recurrence, imatinib was re-initiated. Patients who experienced GIST recurrence while they were treated with adjuvant imatinib therapy received either sunitinib treatment or increased-dose (800 mg) imatinib treatment. It was also possible in the model that those who were treated for the first recurrence experienced another tumour recurrence or discontinued treatment due to adverse events. These patients received increased-dose imatinib therapy, sunitinib therapy or best supportive care (BSC), depending on the particular situation. Patients were at risk of dying at all times.

Patients in the 1-year and 3-year adjuvant therapy arms followed the same treatment pathways after recurrence, facing the same risk of further progression and death. As a consequence, the key difference between the two groups was the length of adjuvant therapy and the corresponding recurrence-free survival. It was implicitly assumed that benefits in terms of recurrence-free survival translate into benefits in terms of overall survival. That is, following recurrence, on average all patients experience the same duration with metastatic disease until death.

Transition probabilities

Time-dependent monthly probabilities of first GIST recurrence were derived from the Kaplan-Meier recurrence-free survival curves of the high-risk population in the SSG XVIII/AIO trial for the first 6 years of the model, the period for which RFS was observed. For the period beyond follow-up time, recurrence probabilities were derived from the extrapolated RFS. To extrapolate RFS parametric distributions (Gompertz, Weibull and loglogistic) were fitted to the observed individual patient level data for each treatment group separately. The most plausible distribution was selected using both the Akaike information criterion (AIC) and visual inspection. The log-logistic model was considered to offer the most credible long-term prediction. To avoid a jump-off bias between the last observed (72nd month) and the first forecasted (73rd month) value of RFS, the last observed RFS value was used as a basis for the projections. That is, RFS in each forecasted month was the product of the last observed RFS value (72nd month) and a reduction factor, which was defined as the ratio of RFS estimated by the loglogistic model for a particular forecasted month and the RFS estimated by the log-logistic model for the 72nd month. Reduction factors are often used for mortality forecasting in demography when observed and forecasted mortality are combined36. Figure 2 presents the observed and extrapolated RFS curves in the base case analysis.

Figure 2. Observed and extrapolated recurrence-free survival in patients with high risk of tumour recurrence.

Time-independent monthly recurrence probabilities for the second GIST recurrence, and monthly mortality probabilities following GIST recurrence (first and second separately) during treatment with imatinib, sunitinib, and BSC were derived from published sources10,37,38 and are presented in Table 1. Age-specific general population mortality probabilities, assumed to be associated with non-GIST causes, were derived from the 2011 Dutch government life table39.

Table 1. Main clinical and economic input values (mean, standard error, the distribution used in the probabilistic sensitivity analysis and the source of input data).

	Mean	SE	Distribution	Source
1-month probability of second recurrence (in health state)
Imatinib 400 mg/day	0.0336	0.0015*	Beta	Verweij et al.38
Imatinib 800 mg/day	0.0285	0.0013*	Beta	Verweij et al.38
Sunitinib	0.1172	0.0078*	Beta	Demetri et al.10
1-month death probability following first GIST recurrence
Imatinib 400 mg/day	0.0153	0.0007*	Beta	Verweij et al.38
Imatinib 800 mg/day	0.0125	0.0006*	Beta	Verweij et al.38
Sunitnib	0.0403	0.0025*	Beta	Demetri et al.10
1-month death probability following second GIST recurrence
Imatinib 800 mg/day	0.0403†	0.0025*	Beta	Assumption
Sunitnib	0.0403	0.0025*	Beta	Demetri et al.10
BSC	0.0725	0.0090*	Beta	Reichardt et al.37
1-month discontinuation probabilities
During 1-year adjuvant imatinib (1–6 months)	0.0136	0.0014	Beta	Heintz et al.40
During 1-year adjuvant imatinib (7–12 months)	0.0009	0.0001	Beta	Heintz et al.40
During 3-year adjuvant imatinib (1–6 months)	0.0097	0.0001	Beta	Heintz et al.40
During 3-year adjuvant imatinib (7+ months)	0.0028	0.0003	Beta	Heintz et al.40
Imatinib 400 mg/800 mg for recurrent GIST	0.0029	0.0001*	Beta	Verweij et al.38
Sunitinib for recurrent GIST	0.0499	0.0114*	Beta	Demetri et al.10
Quality-of-life
Recurrence-free health state	0.785	20% of the mean	Beta	Ara and Brazier42
Reduction during imatinib treatment	0.081	20% of the mean	Beta	Chabot et al.41
First GIST recurrence	0.748§			Assumption
Second GIST recurrence	0.712	0.014	Beta	Chabot et al.41¶
Best supportive care	0.577	0.029	Beta	Chabot et al.41¥
Monthly cost of drug therapy
Imatinib 400 mg once daily	€2578	–	Fixed	**
Imatinib 800 mg/day	€5156	–	Fixed	2× imatinib 400 mg
Sunitnib 50 mg	€3700	–	Fixed	***
Medical costs
During/after completing adjuvant therapy, no recurrence (monthly cost)	€12.5– €160.4	20% of the mean	Gamma	See Supplementary Appendix B for details
No adjuvant therapy due to AE, no recurrence (monthly cost)	€47.1–€60.2	20% of the mean	Gamma	See Supplementary Appendix B for details
Recurrent GIST, (400 mg imatinib, 800 mg imatinib, sunitinib) (monthly cost)	€107.4	20% of the mean	Gamma	See Supplementary Appendix B for details
One time recurrence cost	€1146.4	20% of the mean	Gamma	See Supplementary Appendix B for details
BSC (monthly cost)	€1222.5	20% of the mean	Gamma	See Supplementary Appendix B for details
Adverse events (one-time cost)	€244.5	20% of the mean	Gamma	See Supplementary Appendix B for details

*25% of the range of the large-sample approximate confidence interval, i.e. (μ*(1 + 1.96/sqrt(N) − μ*(1 + 1.96/sqrt(N))/4.

**Monthly cost was calculated by 365.25 days in each year times the daily cost and divided by 12 months.

***Average monthly cost was based on regimen of 4 weeks on, 2 weeks off treatment.

†Assumed to be the same as the death rate of patients receiving sunitinib therapy.

¶Based on health state: ‘On sunitinib treatment, no progression’.

¥Based on health state: ‘Progression’.

§Average of RFS and 2nd GIST recurrence.

Time-dependent monthly probabilities of discontinuing adjuvant therapy were derived in a retrospective data analysis based on the high risk population of the SSG XVIII/AIO trial40. The most common reasons for treatment discontinuation were AEs or abnormal laboratory values. Data suggested that discontinuation probabilities were higher in the first 6 months than in later months. Those who discontinued adjuvant therapy were assumed to experience recurrence at the same rate as those who did not discontinue. Discontinuation rates during the treatment for recurrent GIST were taken from published literature10,38. Discontinuation probabilities are summarized in Table 1.

The model structure accommodates choices between different treatment options after recurrence or treatment discontinuation in certain situations. The proportion of patients receiving a particular treatment in these cases was guided by the Dutch oncologist panel. Detailed information on these proportions is shown in Supplementary Appendix B.

Quality-of-life

Quality-of-life values, presented in Table 1, were assigned to patients according to the phase of the disease, i.e. recurrence-free, recurrent disease (first and second recurrence separately), and progressive disease, i.e. receiving BSC. Utility values for each phase of disease were derived or taken directly from published literature41,42. To account for potential adverse events during treatment, a utility decrement of 0.081 was applied while patients were on treatment41.

Costs

Dutch drug acquisition costs of imatinib and sunitinib were taken from the publicly available online data base of Dutch Health Insurance Board (CvZ) providing information on the consumer prices of drugs43. The consumer price is based on the pharmacy purchase price and includes 6% value added tax as well as a 6.82% clawback (with a maximum of €6.80 per prescription). The monthly cost of 400 mg imatnib, 800 mg imatinib, and sunitinib therapy was €2578, €5156, and €3700, respectively. Both drugs are given on an outpatient basis and are administered orally.

Costs of monitoring therapy, test procedures, and surgery for recurrent GIST were used to estimate the monthly cost of care in each health state, i.e. recurrence-free, recurrent GIST, and BSC. The frequency of oncologist visits, general practitioner (GP) visits, and required tests varied by the elapsed time after surgery and were based on recommendations of the European guidelines17 or expert opinion. All cost elements used in the model were inflated to 2011. Table 1 presents the monthly costs associated with each health state in the model. Detailed information on the derivation of these costs can be found in Supplementary Appendix B.

Recurrence-free health states were associated with drug costs of the adjuvant therapy, costs of monitoring therapy, and the costs of managing adverse events. Recurrent GIST health states included three main cost elements. At the onset of GIST recurrence a one-time cost element was incurred (including monitoring and surgery costs), whilst on-going costs included drug therapy costs and associated medical care costs, and a cost element for treating adverse events. Cost of BSC was estimated as the weighted average of the resource use in the period prior to the last year of life (continuing phase) and of the costs in the last year of life. Costs of the continuing phase included costs of specialist visits and tests. Estimation of these costs was guided by Dutch oncologists. The expenditures during the last year of life of a cancer patient in the Netherlands was taken from published literature44.

All patients experienced some adverse events in the SSGXVIII/AIO trial. Grade 3 or 4 adverse events were reported by 20.1% of the patients in the 1-year arm and 32.8% in the 3-year arm. In the model the costs of managing adverse events were determined based on the 10 most frequently occurring grade 3 or 4 adverse events. Other adverse events were assumed to require minor tests or inexpensive medications, which were not calculated.

Sensitivity analysis

A number of deterministic univariate sensitivity analyses were conducted to test the robustness of model results to changes in key parameters. This was done by varying the parameters one at a time through plausible ranges and re-calculating the cost effectiveness. Parameters investigated in these analyses included (1) the time horizon of the model, i.e. 10 years and 5 years, (2) the method of estimation of recurrence rates, i.e. derived from RFS curves that were modelled with log-logistic, Weibull and Gompertz parametric models for the entire time horizon, equal RFS in the two arms after the observation period, and gradual convergence of RFS by the end of the model horizon, and (3) the cost of imatinib, i.e. a 50% price reduction in both arms, and 100% reduction in the 1-year arm during adjuvant phase. Besides these scenario analyses, the effect of a number of other parameters were investigated which included recurrence and discontinuation probabilities, the dose of adjuvant imatinib therapy, the cost of AEs, the quality-of-life of patients in recurrence-free and recurrent GIST health states, the costs and quality-of-life of patients receiving best supportive care, age at model entry, and mortality probabilities.

A probabilistic sensitivity analysis (PSA) using 1000 simulations was performed to estimate the simultaneous effect of uncertainty surrounding the model parameters. The results of the PSA were used to simulate the joint distribution of the model outcomes, i.e. expected effects, expected costs for each treatment group, the difference of these, and the incremental cost-effectiveness ratios (ICER). Similarly, the results of the PSA were used to estimate the 95% confidence intervals (95% CI) around the model outcomes. For each simulation the model outcomes were calculated and the 25th (or the 975th) of the ordered values indicated the 2.5% (or the 97.5%) boundaries of the CIs.

The probability distributions and the standard errors of the model parameters are shown in Table 1. Simulating uncertainty surrounding the monthly probability of first GIST recurrence was done by non-parametric bootstrapping method from the individual level time-to-recurrence data. The bootstrap draws were used to calculate the Kaplan-Meier recurrence-free survival estimates that were forecasted repeatedly for each bootstrapped dataset using reduction factors, as described above. From each simulated RFS profile, the monthly recurrence probabilities were derived.

Results

Eighty-four per cent and 71% of the patients completed 1 year and 3 years of adjuvant imatinib therapy, respectively. The extended duration of adjuvant therapy resulted in a decrease in the probability of recurrence that, in turn, was translated into benefits in terms of long-term overall survival. Estimates of overall survival revealed that 76% and 37% of the patients in the 3-year treatment group were expected to be alive after 5 and 10 years, respectively. In the 1-year therapy group the corresponding figures were 56% and 26%. An illustration of the predicted overall survival curves is shown in Figure 3.

Figure 3. Predicted overall survival for patients receiving 1-year and 3-year adjuvant imatinib therapy.

In terms of the distribution of patients dying from GIST and other causes in the model, 81% of the patients died from GIST in the 3-year group and 88% in the 1-year therapy group. The model predicted that 15% and 10% of the patients in the 3-year and 1-year arm, respectively, experienced no recurrence over the whole lifetime and thus died from another reason than GIST. Fifty-three per cent of the patients in the 3-year arm and 49% of the patients in the 1-year arm received best supportive care and died from this health state.

Overall survival, recurrence-free survival, and quality-adjusted life years for an average patient are presented in Table 2. Over the whole lifetime, the average number of life years was higher in the 3-year group than in the 1-year group, i.e. 9.83 undiscounted life years (95% CI = 8.18–11.35) and 8.91 discounted life years (95% CI = 7.24–10.10) in the 3-year therapy group, while 7.65 undiscounted life years (95% CI = 6.79–8.52) and 7.04 discounted life years (95% CI = 6.10–7.73) in the 1-year therapy group. Patients assigned to the 3-year therapy arm lived, on average, 8.00 years (81%) out of the 9.83 years free of recurrence, and spent 0.56 years (6%) in the stage of best supportive care. The corresponding figures in the 1-year therapy arm were 5.65 years (74%) and 0.59 years (8%) out of the 7.65 life years. Similarly to life years, quality adjusted life years were higher in the 3-year group than in the 1-year group; 6.55 (95% CI = 5.08–7.52) and 5.18 (95% CI = 3.84–5.86) QALYs, respectively.

Table 2. Overall survival, quality-adjusted life years (undiscounted and discounted with 1.5%) and costs per patient (undiscounted and discounted with 4%), base case analysis.

	Undiscounted			Discounted
	3-year adjuvant therapy	1-year adjuvant therapy	Difference	3-year adjuvant therapy	1-year adjuvant therapy	Difference	p Value
Overall survival	9.83	7.65	2.18	8.91	7.04	1.87	<0.001
Recurrence-free	8.00	5.65	2.35	7.29	5.22	2.07	<0.001
With recurrent GIST	1.27	1.40	−0.13	1.13	1.28	−0.15	0.813
Receiving BSC	0.56	0.59	−0.03	0.49	0.54	−0.05	0.623
QALYs	7.24	5.63	1.61	6.55	5.18	1.37	0.020
Recurrence-free	6.08	4.36	1.72	5.52	4.02	1.50	<0.001
With recurrent GIST	0.85	0.93	−0.09	0.75	0.85	−0.1	0.813
Receiving BSC	0.32	0.34	−0.02	0.29	0.31	−0.02	0.623
Drug costs	€124,131	€78,205	€45,926	€109,489	€68,084	€41,405	<0.001
Imatinib during adjuvant therapy	€77,368	€27,619	€49,749	€74,631	€27,619	€47,012	<0.001
Imatinib during treatment of recurrence	€36,549	€39,802	−€3.253	€26,903	€31,573	−€4670	0.795
Sunitinib	€10,215	€10,785	−€570	€7955	€8892	−€937	0.651
Medical costs	€12,807	€12,792	€15	€10,707	€10,345	€362	0.880
During adjuvant therapy	€1558	€861	€697	€1520	€861	€659	<0.001
During treatment of recurrence (imatinib 400 mg/800 mg or sunitinib)	€3089	€3219	−€130	€2312	€2633	−€321	0.888
Best supportive care	€8610	€8711	−€101	€6053	€6851	−€798	0.866
One-time recurrence costs	€970	€1040	−€70	€772	€887	−€115	0.890
Adverse event costs	€59	€52	€7	€50	€45	€5	0.048
Total costs	€137,967	€92,088	€45,879	€120,196	€79,361	€40,835	<0.001

QALYs, quality adjusted life years; BSC, best supportive care.

p Values indicate the probability that a particular outcome is larger for the 1 year arm than for the 3 year arm. P-values were derived from the results of the probabilistic sensitivity analysis and were the same for undiscounted and discounted outcomes.

A breakdown of the average undiscounted and discounted costs per patient is presented in Table 2. Over the lifetime, the average total discounted costs amounted to €120,196 and €79,361 in the 3-year and 1-year arm, respectively. Of these, €74,631 (62%) and €27,619 (35%) were adjuvant imatinib therapy drug costs. The higher adjuvant therapy cost in the 3-year arm was partly offset by lower drug costs associated with treatment of recurrence (€34,858 vs €40,465). Other costs associated with physician visits and tests were approximately equal in the two arms.

Adjuvant imatinib therapy given for an extended period of time resulted in 1.37 (95% CI: = 0.17–2.45) additional QALYs, 1.87 (95% CI = 0.37–3.21) life-years, and €40,835 (95% CI = €34,891–€46,983) incremental costs. It yielded an incremental cost-effectiveness ratio of €29,872 per QALY (95% CI = €12,690–€98,571), and €21,865 per LY (95% CI = €10,820–€80,845).

Results of deterministic univariate analyses are presented in Table 3. Decreasing the time horizon of the model, estimating recurrence-free survival (and corresponding recurrence rates) with different methods, and the price of imatinib had the largest impact on the model results. Incremental costs did not change significantly if the time horizon was set to 10 or 5 years, but the incremental QALYs decreased considerably. These scenarios, however, mimicked extreme situations in that no more differences between the strategies accrue beyond 10 or 5 years. Using parametric distributions for modelling recurrence-free survival also had a substantial impact on the results. Of the parametric distributions, the log-logistic model had the best fit and was considered to offer the most credible forecasts, even though extrapolated health benefits were likely to be under-estimated in the 1-year arm. In this scenario the incremental QALYs increased by 0.98 and the ICER decreased by 44% compared to the base case. The Weibull distribution indicated shorter recurrence times than the log-logistic model, which ultimately resulted in patients progressing more quickly through the health states with poorer quality-of-life. In the end, the Weibull model resulted in larger incremental total costs and lower QALY gains than those of the log-logistic model. The Gompertz model resulted in future RFS estimates that were considerably lower in both arms. The impact of equal RFS beyond the observation period immediately or in 10 years decreased the incremental QALYs and increased the ICER compared to the base case analysis by 118% and 22%, respectively. Supplementary Appendix C presents figures that depict the different RFS scenarios.

Table 3. Results of the sensitivity analyses, incremental costs, outcomes, and cost-effectiveness ratios.

	Incremental			ICERs
Analysis	Costs	QALYs	LYs	Cost per QALY	Cost per LY
Base case	€40,835	1.37	1.87	€29,872	€21,865
Time horizon
10 years	€38,791	0.75	1.07	€51,399	€36,264
5 years	€33,167	0.25	0.39	€131,880	€84,332
Recurrence-free survival
Loglogistic parametric model	€35,882	2.16	2.85	€16,597	€12,575
Weibull parametric model	€39,165	1.61	2.17	€24,347	€18,023
Gompertz parametric model	€44,784	0.80	1.18	€55,720	€38,091
Equal RFS beyond end of observation period	€44,186	0.68	1.01	€65,143	€43,631
Convergence of RFS curves beyond follow-up time	€42,190	1.16	1.62	€36,391	€26,119
Imatinib 400 mg cost: −50%	€19,664	1.37	1.87	€14,385	€10,529
Imatinib 400 mg cost: −100% in the 1-year arm, adjuvant phase	€68,454	1.37	1.87	€50,075	€36,654
Equal recurrence and discontinuation probabilities in the two arms during the first years	€40,500	1.36	1.86	€29,796	€21,826
Real imatinib dosage used: 3-year imatinib: 394.3 mg, 1-year imatinib: 393.1 mg	€40,835	1.37	1.87	€29,322	€21,463
QoL in recurrence-free: perfect health	€40,835	1.80	1.87	€22,706	€21,865
QoL in recurrence-free −10%	€40,835	1.21	1.87	€33,744	€21,865
QoL in recurrent GIST +10%	€40,835	1.36	1.87	€30,016	€21,865
QoL in recurrent GIST −10%	€40,835	1.37	1.87	€29,729	€21,865
QoL in BSC +10%	€40,835	1.36	1.87	€29,929	€21,865
QoL in BSC −0%	€40,835	1.37	1.87	€29,814	€21,865
Cost of AE +10%	€40,838	1.37	1.87	€29,873	€21,867
Cost of AE −10%	€40,833	1.37	1.87	€29,870	€21,864
Cost of BSC +10%	€40,516	1.37	1.87	€29,638	€21,695
Cost of BSC −0%	€41,154	1.37	1.87	€30,105	€22,036
Age at model entry: 55	€41,459	1.51	2.06	€27,398	€20,140
Age at model entry: 65	€40,138	1.24	1.71	€32,284	€23,534
Death probability from GIST: −50%	€38,949	1.30	1.76	€29,956	€22,177
Death probability from GIST: −25%	€40,835	1.37	1.87	€29,872	€21,865

A price reduction of 50% for imatinib had a large positive impact on the model results, i.e. 52% decrease in the ICER. The second price scenario, i.e. zero adjuvant therapy drug cost in the 1-year arm, was constructed to mimic the hypothetical situation in that 3-year adjuvant therapy is compared with placebo. Although implicitly this scenario made the very conservative assumption that 1-year adjuvant imatinib therapy has no benefit over placebo, 3-years of adjuvant imatinib therapy compared to placebo would still be cost-effective in the Netherlands (€50,000 per QALY gained).

The impact of assuming equal progression and discontinuation rates in the first year, actual dose of adjuvant imatinib therapy in the clinical trial, various AE costs, and various quality-of-life values for the modelled health states had limited impact on the results. The ICER varied between a minimum and maximum value of €22,706 and €33,744 per QALY gained, respectively, in these analyses.

The results of the PSA are presented as a scatterplot and the derived cost-effectiveness acceptability curve in Figure 4. The probability of being cost-effective was 12.2%, 51.0%, 68.1%, and 80.4% at a willingness-to-pay threshold of €20,000, €30,000, €40,000, and €50,000 per QALY, respectively.

Figure 4. Scatterplot of incremental cost and incremental QALYs (base case scenario), and cost-effectiveness acceptability curve.

Discussion

A multistate Markov model was built to assess the cost-effectiveness of 3 years vs 1 year of adjuvant imatinib therapy. The model simulated how the clinical status of primary GIST patients evolved over time. Multistate Markov models are widely used tools that help describe complex processes, such as possible clinical pathways after resection of GIST, whilst acknowledging relevant aspects of the disease from the decision-maker’s perspective. In the model, two hypothetical cohorts were compared in terms of health benefits and costs; patients at high risk of tumour recurrence were assigned to receive either 3 years or 1 year of adjuvant imatinib therapy. The principal data source for the model was the SSGXVIII/AIO study. Cost effectiveness was expressed in terms of the incremental cost per QALY gained and incremental cost per life-year saved.

Results of the model suggested that adjuvant imatinib therapy given for an extended period offers considerable health benefits; on average, 1.37 (95% CI = 0.17–2.45) additional QALYs and 1.87 (95% CI = 0.37–3.21) life-years. The health benefits were estimated to be reached at an incremental cost of €29,872 per QALY (95% CI = €12,690–€98,571). Probabilistic sensitivity analysis suggested that the probability of cost-effectiveness at a €30,000 threshold was 51.0% and at a €50,000 threshold was 80.4%. Time horizon of the model and the method of extrapolation of the RFS curves were the major drivers of the cost effectiveness ratio, showing that the validity of the survival gain estimated in the SSGXVIII/AIO trial is also the main determinant of the validity of the economic evaluation.

Other studies

The benefits of giving adjuvant imatinib therapy for 3 years for patients with GIST and a high risk of recurrence has been demonstrated in the large-scale SSGXVIII/AIO study27. Further data indicating the benefits of long-term imatinib therapy were reported in Li et al.29 and in Jiang et al.28. These studies were single-centre, non-randomized, prospective studies that assessed the efficacy of adjuvant imatinib treatment, given for 36 months vs no treatment in patients following surgical resection of GIST, who were considered to be at intermediate or high risk of recurrence. Unfortunately the low number of patients (n = 105 in Li et al., n = 90 in Jiang et al.) and the treatment strategies in these studies (3 years of adjuvant imatinib therapy vs no therapy) did not allow for a formal comparison of RFS and OS between our study and these studies. However, each of these studies independently and consistently demonstrated the ability of adjuvant imatinib treatment to suppress tumour recurrence following surgery. In Li et al., it was additionally reported that, similar to what our study predicted, adjuvant treatment with imatinib significantly reduced the risk of death in patients with GIST at high or intermediate risk of recurrence (hazard ratio = 0.25; 95% CI = 0.07–0.93, median follow-up time 45 months).

Two other cost-effectiveness studies have been published in the treatment of GIST based on efficacy data from phase III trials45,46. Similarly to our study, Sanon et al.46 assessed the cost-effectiveness of 3 years vs 1 year of adjuvant imatinib therapy from the US payer’s perspective using clinical data from the SSGVIII/AIO trial. Due to differences between the Netherlands and the US in prices of interventions (e.g. monthly cost of daily 400 mg imatinib/sunitinib was ∼€3700/€4600 after conversion, in contrast to only €2578/€3700 in the Netherlands) and approaches to modelling recurrence rates, the results of these studies differed noticeably. Similarly to our study, monthly recurrence rates were used in Sanon et al., however the recurrence probabilities were kept constant within a year. In addition, within each treatment group, the same recurrence rates were applied beyond 5 years as was used for the fifth year. Using such an approach to model recurrence probabilities potentially over-estimates the true health benefits of the therapies, although not necessarily the incremental health benefits, and a more refined method, for example in our study, may be more appropriate. In particular, the authors estimated that patients receiving 3 years vs 1 year of adjuvant imatinib therapy accumulated QALYs of 8.53 (+1.98 compared to our study) and 7.18 (+2.00), and lifetime costs of $302,100 (∼+€125,000) and $217,800 (∼+€90,000), respectively. Accordingly, the ICER was estimated at $62,600 per QALY gained (∼€47,500), 50% higher than what our study estimated for the Netherlands.

Paz-Ares et al.45 estimated the cost-effectiveness of sunitinib vs BSC as second-line treatments in patients with metastatic and/or unresectable GIST after progression or intolerance to imatinib therapy from the perspective of the Spanish National Health System45. A Markov model was developed for which the transition probabilities between the three health states (progression-free survival, progression, and death) were obtained from the sunitinib pivotal trial10. Projected LYs and QALYs were higher for sunitinib compared with BSC: 1.59 vs 0.88 and 1.00 vs 0.55, respectively. The ICERs obtained were: € 30,242 per LY and €49,090 per QALY gained. The study results could not be directly compared with the results of our study because of the difference in the complexity of the two models and because of the line of treatment in which imatinib and sunitinib were considered.

Limitations

While in most cases using a parametric distribution function to model and extrapolate RFS is considered an appropriate and convenient method, there are several limitations associated with its application. In general, when a curve fitting approach is employed, the goal is to find the distribution that fits the observed data best because then the predicted survival times are close to the observed data and are described with a smooth curve. However, there are situations when none of the typically employed parametric functions (e.g. exponential, log-logistic, Weibull) fit the data particularly well. In these cases, selecting the proper distribution may be a concern, and using the observed data may be more appropriate. For our study, therefore, the observed RFS values, and the corresponding recurrence rates, were used for the first 6 years of the base case analysis.

Although none of the parametric functions captured the drop in the RFS curves after 1.5 and 3.5 years of follow-up in the 1-year and 3-year groups, respectively, the log-logistic model had a reasonable model fit and offered the most credible forecasts among the three parametric models. Consequently, recurrence rates beyond the observation period were derived from RFS estimates that were obtained by the log-logistic model and were rescaled to avoid a jump-off bias between the first extrapolated and the last observed RFS values. When RFS was extrapolated by the Weibull and the Gompertz functions, both models predicted a steeper decrease beyond the trial period that reached zero relatively quickly. These projections were considered less plausible, thus the log-logistic assumption was favoured for the base case analysis. To demonstrate that the Weibull and Gompertz functions did not reflect clinical reality, we refer to the results of a study of Miettinen et al.47 in which long-term follow-up data on the natural history of disease were presented from 1765 patients. The study included 341 patients who qualified as high-risk patients according to the Miettinen criteria. In patients with gastric tumours >5–10 cm, with mitosis of >5/50 per high-power field, 26% were still alive without evidence of disease at a median follow up of 13 years. In patients whose tumours were >10 cm and who had mitosis of >5/50 high-power field, 10% remained alive and were disease-free at a median follow-up of 19 years. These figures suggest that there is a group of patients with high survival that could not been captured by the Weibull and Gompertz models; that is, the RFS would have been under-estimated.

To test the sensitivity of model results in response to various functional forms of RFS, a number of deterministic scenario analyses were conducted. It was found that the results of the base case analysis were less favourable, i.e. used conservative assumptions, than the model results based on fully parametric log-logistic or Weibull RFS models. When RFS curves were assumed to converge within 10 years after follow-up time, similar model results were found as in the base case analysis. Undoubtedly, lacking long-term data makes it challenging to tell which RFS curves are projected the most accurately by these models, a limitation that most extrapolations suffer. However, modelling and forecasting RFS in our study was supported by the clinical experience, and, except for extreme situations, the model results of the scenario analyses indicated that 3-years vs 1-year adjuvant imatinib therapy was cost-effective in the Dutch setting.

Another limitation of our study was the difference in the level of overall survival between the SSGXVIII/AIO trial and what was projected by the model. In the SSGVIII/AIO trial the OS was 89% in the 3-year imatinib group and 74% in the 1-year group after 5 years of follow-up. Somewhat lower values were predicted by the model, 76% and 56% of survival for the two groups, respectively. There are a number of reasons why such a difference could occur. First, in the model, mortality and progression probabilities after the first tumour recurrence were treatment-specific; that is, they were different for patients who received imatinib 400 mg, imatinib 800 mg, sunitinib, and BSC. These probabilities were derived from studies that included patients with certain characteristics or disease history, that were potentially different from those patients who experienced recurrence in the SSGXVIII/AIO trial. For example, patient characteristics and disease severity after recurrence in the SSGVIII/AIO trial may be different from the patient characteristics in the sunitinib trial, which was used to derive secondary recurrence and mortality probabilities for the model. Another possible reason for the OS differences comes from the fact that recurrence was measured differently in the trials that were used to derive recurrence rates after the first GIST recurrence. Derived recurrence rates based on a stricter criterion artificially transitions patients to further treatment lines in a relatively short period of time, for which the mortality is higher. Finally, in the model, it was simplistically assumed that, once recurrence occurs, the patient immediately receives a next-line therapy. However, in reality it may not be true for every patient. All taken together, lower predicted (than measured) OS was an issue for both arms, but OS did not only affect the estimated health effects but also the estimated costs. Sensitivity analyses of mortality rates showed that the impact of this downward bias on the ICERs was small.

Another limitation of our study was the number of options available in the model for the third line treatment of GIST. In current clinical practice drugs, such as nilotinib, sorafenib, or regorafenib48–50, are also given for some patients with metastatic GIST refractory to first-line imatinib and second-line sunitinib. While this is an admitted limitation of the study, it likely has a marginal effect on the results because the expected time spent in BSC health state was small and differed between the two groups only to a small extent.

Lastly, health effects may depend on the specific mutation. The population under study comprised patients with GIST of KIT or PDGFRA mutation; however, patients with a certain KIT mutant isoform (exon 9) or with PDGFRA D842V-mutated GIST reflect lower sensitivity to imatinib therapy27,51. Regrettably, RFS data over time were not reported for sub-groups in the SSGXVIII/AIO trial; therefore, further analyses could not be carried out for specific patient groups.

Conclusions

A multi-state model was developed to assess the cost-effectiveness of 3 years vs 1 year of adjuvant treatment with imatinib following surgical resection of primary, localized GIST in patients at high risk of disease recurrence. The model assumed a Markov process to evaluate GIST recurrence, quality-adjusted survival and costs from the initiation of the treatment until death. Although based on a single trial, and associated with uncertainty, there is clear evidence that that delayed recurrence due to treatment with longer-term adjuvant imatinib therapy represents a cost-effective treatment option according to currently accepted standards of cost-effectiveness in the Netherlands.

Transparency

Declaration of funding

Funding for this study was provided by Novartis Oncology (the Netherlands).

Declaration of financial/other relationships

IMM and BGV are employees of Pharmerit Internations, who were paid consultants to Novartis with regard to the development of this manuscript. AJG and WBH have nothing to disclose. EG is an employee of Novartis Oncology. IMM, BGV, AJB, and WBH have no conflict of interest to report. We, the authors, attest that we have herein disclosed any and all financial or other relationships that could be construed as a conflict of interest and that all sources of financial support for this study have been disclosed and are indicated in the acknowledgement. JME Peer Reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgements

The authors would like to thank Myrna Hennequin for her analytic support in this study.