Background
Crohn’s disease (CD) is a relapsing–remitting inflammatory bowel disease which mainly affects the gastrointestinal tract. Common symptoms include abdominal pain, fever, bowel obstruction and diarrhoea with passage of blood or mucus, or both [
1]. Disease complications include malnutrition, fistulas, fissures and abscesses [
2]. Approximately 20,000 people in Sweden are affected by CD [
3], with around 10 new cases per 100,000 person-years [
4]. There are indications that CD incidence and prevalence are increasing [
5‐
10] with most patients being diagnosed between the ages of 15 and 30 [
3]. The disease is associated with significant treatment costs, productivity losses and a substantial quality of life impact [
3,
11]. The annual per patient cost of CD in Europe has previously been estimated between €6024 and €22,581. The costs were related to primary and secondary care, medical and surgical hospitalization, and labour force non-participation [
12‐
14]. A Polish study from 2016 estimated the annual indirect costs at around €5550 per patient [
15]. The cost of CD is expected to increase due to the shift from hospitalisation and surgery towards biologic therapy use, shown in the COIN study [
16]. Reported health related quality of life (HRQoL) is significantly lower in CD patients as compared to normal populations [
12]. It has been estimated that more than 20% of Crohn’s disease patients will require surgery within 5 years of diagnosis, and approximately 40% within 10 years [
17]. Common CD surgeries include strictureplasty, resection, and colectomy [
18].
There is no cure for CD, but a variety of disease managing, often long-term, treatments are available [
19]. Biological treatments including tumour-necrosis-factor (TNF)-alpha inhibitors such as adalimumab and infliximab have significantly improved CD management, especially in patients who have failed conventional treatments (e.g. steroids, 5-aminosalicylic acid, and immunomodulators) [
20]. However, approximately one-third of patients fail anti-TNF treatment [
21]. Monoclonal antibodies, for instance biologic agent vedolizumab, have been shown to be effective in TNF-alpha inhibitor non-responders, patients with contraindications to TNF-alpha inhibitors, and in patients who lose response or develop intolerance to such agents [
22]. The human monoclonal antibody ustekinumab is a novel treatment option in CD which blocks the pro-inflammatory cytokines interleukin (IL)-12 and IL-23. The safety and efficacy of ustekinumab as an induction and maintenance therapy in moderate to severe CD has been evaluated and established in three phase 3 trials (UNITI-1 [
23], n = 741; UNITI-2 [
24], n = 628; and IM-UNITI [
24,
25], n = 388). In the induction trials (UNITI-1 and UNITI-2), ustekinumab was found to produce significantly higher clinical response and remission rates compared to placebo both in patients who had failed conventional therapy and patients who had failed treatment with TNF-alpha inhibitors [
23,
24]. Ustekinumab has been approved in the European Union for the treatment of moderately to severely active CD in patients who have an inadequate response with, or have lost response to, or are intolerant to either conventional therapy or a TNF-alpha inhibitor, or have medical contraindications to such therapies [
26].
The objective of this study was to assess the cost-effectiveness of ustekinumab as a treatment of moderate to severe Crohn’s disease in Sweden. The cost-effectiveness of ustekinumab was investigated in two populations versus the relevant comparator in each population. TNF-alpha inhibitor adalimumab was the comparator in patients who had failed conventional care, some of which had been exposed to TNF inhibitor treatment without experiencing treatment failure or unacceptable side effects (“conventional care failure” population). The monoclonal antibody vedolizumab was the comparator in a population of patients who previously failed TNF inhibitor treatment (“TNF-alpha inhibitor failure” population).
Previous cost-effectiveness studies of ustekinumab in Crohn’s have been undertaken in British [
27] and Polish [
28] settings. Considering that cost-effectiveness analyses are highly contingent on local specifics such as cost and utility data, incidence and prevalence, as well as the health care system and clinical practice, the previous ustekinumab analyses are insufficient to draw conclusions for a Swedish context. Scandinavia sees the highest incidence and prevalence rates of Crohn’s in Europe [
11], which highlight the value of evaluations of novel treatment options. The adaptation of the model to a Swedish context included adding productivity costs. A societal perspective captures the indirect costs of CD, which in addition to constituting a substantial proportion of the total costs, reflect the severe impact on Crohn’s patients’ everyday life.
This analysis aims to add to the understanding of the treatment possibilities and to enable future comparisons in a population with a considerable unmet need.
Discussion
Assuming a lifetime time horizon and 2-years treatment duration, the results indicate that ustekinumab dominates adalimumab in the conventional care failure population with a QALY gain of 0.232, and is cost-effective compared to vedolizumab in the TNF-alpha inhibitor failure population. The QALY gain versus vedolizumab was 0.133 and the cost per QALY was €30,282, which is below the Swedish reference WTP. The results were robust to several univariate sensitivity analyses, and the PSA. Ustekinumab was dominating adalimumab in all sensitivity analyses. Considering the Swedish reference WTP, ustekinumab was cost-effective versus vedolizumab in all one-way sensitivity analyses, with the exception of when the treatment duration was increased to 5 years and when the time horizon was decreased to 5 years. Besides treatment duration and time horizon, using different utilities and the exclusion of indirect costs had the largest impact on the results.
To our knowledge, there are no other cost-effectiveness studies of ustekinumab in moderate to severe CD in Sweden. Therefore, it is not straightforward to make a comparison to previous literature. There are no head-to-head trials between biologic agents, and there is a great variation in assumptions and data sources used by cost-effectiveness studies of biologics in CD [
66]. This results in considerable variations in the outcomes of the studies. For example, a review of the cost-effectiveness of biologics in CD patients found 12 such studies, with ICERs versus conventional care ranging from dominating to €549,335 [
67]. Another study assessed the cost-effectiveness of ustekinumab versus vedolizumab from a payer perspective in the US, in a population of TNF-alpha inhibitor naïve, non-TNF-alpha inhibitor refractory patients. Considering a lifetime perspective, vedolizumab accrued 10.461 QALYs whereas ustekinumab accrued 10.326 with incremental costs of $55,523 in favour of vedolizumab [
68]. This comparison was done in a different population than ours, and little information on the assumptions was presented. A straightforward comparison on the difference to our outcome is therefore impossible. Another analysis of ustekinumab’s cost-effectiveness versus vedolizumab as a third line agent after failure with two TNF-alpha inhibitors found that vedolizumab had higher costs of $168,648 as well as a larger QALY gain of 0.029, resulting in an ICER of $5,815,448 [
69]. A polish study found treatment with ustekinumab after failing one TNF-antagonist associated to a QALY gain of 0.349 and an ICER of €18,878 compared to adalimumab after infliximab failure [
28]. The variations in outcomes of cost-effectiveness analyses of biologics in CD emphasise the importance of further research in the field, and a need for a greater consensus over model design and assumptions.
Similar model structures as the one utilized for this study have previously been used in HTAs of biological CD treatments [
29,
63,
70,
71]. The analysis is based on the best available clinical evidence and publicly available data. The ustekinumab induction trials (UNITI-1 and UNITI-2) had high internal validity. As a means of improving flexibility, the model allowed continued induction response assessment. As there are no head-to-head trials between biologics in CD, efficacy between the comparators had to be indirectly estimated. NMAs were performed for the induction phase and treatment sequence analyses were utilised for the maintenance phase for the respective populations. The clinical endpoints in the induction trials were estimated at week 6 for ustekinumab and vedolizumab, and week 4 for adalimumab. The efficacy assessment time points in the NMA were in line with the included studies [
29].
Previous failure to TNF-alpha inhibitor treatment was found to impact the results of the NMA. Therefore, separate analyses were undertaken between trials including conventional care failure patients and TNF-alpha inhibitor failure patients. Moreover, a sensitivity analysis was undertaken in which conventional care failure inputs were replaced with inputs for patients truly naïve to biologics, providing evidence of robustness to the results. To further validate the results of the NMA, and to investigate potential sources of bias, several other sensitivity analyses were undertaken which supported the base case outcome [
29].
A common issue in performing indirect comparisons is heterogeneity between the source trials. The placebo arms were not comparable across the maintenance trials, and could therefore not be used as a common comparator to anchor the indirect comparison, rendering a traditional NMA impossible. Moreover, statistical analyses revealed significant levels of heterogeneity. As often in maintenance trials, patient selection was based on induction response to the evaluated treatment. Additionally, the patient profiles in terms of previous TNF-alpha inhibitor treatment failure differed over the trials. The adalimumab trials only included secondary non-responders, whereas the vedolizumab and ustekinumab trials included primary non-responders as well. As the latter inclusion criteria allowed patients with more severe disease, the effect of ustekinumab and vedolizumab risked being underestimated compared to adalimumab in an indirect comparison. As the maintenance effect is a product of initial treatment effect, the full treatment pathway had to be considered. A treatment sequence analysis was therefore the most plausible option to obtain long-term relative efficacy. Not only did this increase the comparability between maintenance trial placebo arms, it also allowed an evaluation of effect over the full treatment sequence [
36].
The first cost-effectiveness model for CD was developed by Silverstein et al. [
32] using data from an observational US cohort (1970–1993) from Olmstead County. It was published in 1999. More recently, in 2009, Bodger et al. [
20] adapted the Silverstein model to a UK context. CD models have since then been developed for three NICE appraisals. Namely appraisals of infliximab and adalimumab [
71], of vedolizumab [
63], and of ustekinumab [
29]; the latter being the model which this study is based on. The previous models’ limitations include not capturing the relapsing–remitting nature of CD, overly simplifying surgery by disregarding that subsequent surgery often is contingent to first surgery, and using too short time horizons to capture long-term outcomes. As CD is a chronic condition, such outcomes are important to understand the full implications of the disease.
The Bodger et al. model structure on which this model builds have some limitations. It does not allow for treatment sequencing or re-treatments. This is less critical in our analysis, as the populations evaluated are not treatment naïve, but have experienced either conventional care or TNF-alpha inhibitor treatment. Thus, we indirectly model treatment sequences, but with initial treatment taking place outside of the simulation, reflected in the patient data. Additionally, properly modelling treatment sequences would require currently unavailable efficacy data, or ill-founded assumptions. Although the model allows for maintenance transitions from the moderate to severe health state, the structure does not capture increased relapse risk after discontinuing biologics [
28,
72,
73], and—as previously discussed—does not consider the impact of surgery on future surgeries.
In the development of the ustekinumab model, such limitations were considered. For example, in the model patients can transition from the moderate to severe health state in the maintenance phase, thus reflecting that CD is a relapsing–remitting disease. Developments of the surgery modelling was explored, but resulted in non-intuitive outcomes. Other model developments to overcome previous limitations included differentiating the induction length of the comparators, adding the possibility to test different treatment lengths, and—specifically based on input from three health economic experts and a leading clinician—including a gradual decline in post-treatment efficacy in the maintenance period.
The Solver approach to estimate transition probabilities have a number of limitations, as pointed out by the ERG [
30]. The approach is contingent on constraints and starting values, which although based on a previous NICE submission [
63], were criticized for increasing uncertainty while having a substantial impact on the transition probabilities and subsequently the cost-effectiveness. Previous criticism was however accounted for by the inclusion of additional calibration constraints.
To validate the model, a comparison was done between model outcomes and the predicted outcomes of the treatment sequence analysis. The comparison concerned 1-year outcomes of the proportion of patients ending up in each health state. The model outcomes were very close to the predicted outcomes—with a slight bias against ustekinumab in the conventional care failure population [
29].
Patients who did not respond to the initial and second induction doses were assumed to remain in the moderate to severe health state and spend the remainder of the simulation on standard of care. This is a simplification of clinical practice, but was deemed the most reasonable assumption in absence of adequate data. Some inputs were unavailable for a Swedish setting, for instance no adverse event risk data in Swedish CD patients was identified. Therefore, it was assumed that the same risk as estimated in moderate to severe CD patients in the UK apply to Swedish patients. Modelling assumptions inevitably produce uncertainty, but in order to reduce it to the greatest extent possible, assumptions related to data gaps, model inputs, and treatment regimens were validated by Swedish clinical experts in gastroenterology.
As this study was designed to capture Swedish health economic circumstances, the generalisability of the conclusions to an international context is limited. However, the observed cost-effectiveness outcomes are likely similar within other regions with similar CD incidence and prevalence, and comparable health care systems such as for instance, the other Nordic countries.
Authors’ contributions
All authors have made substantial contributions to the study design, and to the acquisition and interpretation of the data. All authors have contributed to the drafting and critical revision of the manuscript, and have given final approval to the version to be published. Each author has participated sufficiently in the work to take public responsibility for the content, and have agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All authors read and approved the final manuscript.