Introduction
The role of budget impact (BI) in healthcare decision-making varies across different jurisdictions as recent reviews indicate [
1‐
4]. Germany and the USA are examples of jurisdictions that do not have a formal or informal role for budget impact in decision-making. Other countries, for example The Netherlands, France and Australia, do have guidance or even legislation on BI, but the actual role of BI or the impact on decision-making remains rather informal and, moreover, politically driven [
1‐
4]. On the other end of the spectrum, England has one of the best defined systems with a clear role for BI in healthcare decision making [
2,
3]. In general, however, there is an informal role for BI and its contribution to reimbursement decisions often remains unclear. As a result of that, the role of BI in decision-making remains an important topic for debate [
1,
5,
6]. In particular, the growing attention for healthcare and pharmaceutical expenditures in combination with price negotiation mechanisms increasingly raises questions about the role of cost-effectiveness (CE).
Whilst the role of BI is often unclear in reimbursement decision-making, there are ample examples where BI did play a significant role in either the reimbursement decisions or where high BI estimates resulted in restricted reimbursement for a specific patient population [
6‐
11]. Recently, the introduction of new, very effective but high priced Direct-acting antivirals (DAA) in hepatitis C sparked worldwide affordability concerns and access restrictions [
7,
8]. Also in oncology, patients have limited access to many high priced products due to concerns regarding affordability as a result of high BI [
9‐
11].
Especially in the hepatitis C case, the cost-effectiveness of the innovations were generally regarded as positive and medical need was high [
12‐
16]. The future will bring new products with potentially high short-term BI, which could spark further BI-guided restrictions and will call for further deliberation of the role of affordability in the political and societal debate and as such of relevant meaning in healthcare decision-making [
17,
18]. Therefore, clarity on the role and hierarchy of CE vs BI will not only be of interest but seems to become very important in informing reimbursement decisions.
Unfortunately, the (methodological) quality and accuracy of BI analysis does not seem to match the proven scientific rigor of CEAs [
6,
19,
20]. A review by Van de Vooren et al. [
21] reports that (methodological) quality of many published BI analyses is poor. Furthermore, Broder et al. and Cha et al. illustrate that the accuracy of BI predictions is regarded as low [
22,
23]. These observations, in light of the increased debate on drug prices, growing interest in price negotiation, BI of pharmaceuticals as part of healthcare budgets (e.g. Hospital) and, therefore, burden to societies, warrant questions about whether BI is being used properly and what the extent of influence is on patient access.
In this paper, the accuracy and role of BI in reimbursement decisions is investigated by assessing the life cycle of hepatitis C DAAs in the Netherlands. This case was selected as there were concerns for an extremely high BI (up to €1.78 billion). The final reimbursement decision of Sofosbuvir (Sovaldi), the first DAA, resulted in restricting treatment to the most critically ill whilst this seems rather irrational from a cost-effectiveness perspective and was likely triggered by other elements such as price, BI considerations and affordability discussions [
12,
13,
24‐
28].
The aim of this hepatitis C case study is twofold: First, we aim to quantify the accuracy of the BI predictions used for informing the Dutch reimbursement decisions. Second, we attempt to characterize the influence of market-dynamics on actual BI and the way these are implemented in the BI predictions. This includes, for example, timing of regulatory decisions, influence of introductions of new hepatitis C products and the influence of a restricted reimbursement decision that limits the product’s indication.
Discussion
In a Dutch setting, we showed that BI estimates reported in ZIN reimbursement dossiers largely overestimated the actual BI for hepatitis C DAAs. Although the most severely ill patients did get access to the innovative hepatitis C therapies, access was initially not granted to the extent of the recommendations in the then prevailing EASL guidelines.
In the EU, the crude hepatitis C incidence is estimated to be 7.4 per 100,000 persons but with a very large spread (0.1–73.3) between countries, at least partly driven by varying quality of surveillance systems and data completeness [
47‐
49]. ZIN dossiers as well as studies by Iyengar et al., Cornberg et al. and Saraswat et al. report larger potential eligible populations (22,000–28,000, 0.14–0.17%) than those that ZIN actually used for the BI calculations (2000–3000, 0.012–0.018%) [
12,
48‐
50]. The former population, all those with chronic hepatitis C, should be treated according to the most recent EASL guideline. Interestingly, the estimates of 28,000 stated in the reimbursement dossiers are denoted as a scenario where the ‘indication is broadened’ to all hepatitis C patients. No further notion is given as to whether the presumable non-symptomatic patients are actually F0 patients. We can safely conclude that (1) current patient volumes have not been near this level and (2) the estimates of 2000–3000 are likely to be a conservative estimate.
In 2015, when Sovaldi was reimbursed, the BI and estimates of number of patients appear to be rather accurate as we see that, according to the reimbursed indication, treated patients are within the F4–F3/F2 range. Then, with the unrestricted reimbursement of Harvoni and Viekirax + Exviera, treated patients plateau to the predicted F4–F0 population for approximately 4 months. These observations would suggest that the ZIN estimate of 2000–3000 patients per year is quite accurate. From June 2016 onwards, however, patient numbers decline to an F4 + F3 level that is below the expected F4–F0 range. If we, for now, assume that the estimate of number of patients was indeed reasonably accurate, other factors must have been responsible for the large deviations between estimated and actual BI.
A first reason for this deviation could be the inadequate implementation of timing of regulatory decisions. The Sovaldi reimbursement dossier was published on 20 May 2014 based on which ZIN formally advised the minister of health on 23 May 2014. The final reimbursement status was granted per 1 Nov 2014. The delay between advice and reimbursement could have been unforeseen. The manufacturer and/or ZIN could, however, have assumed that reimbursement of Sovaldi during entire 2014 (MA was 16 Jan 2014) was highly unlikely. Still, the BI estimate assumed access during the entire year. This alone contributed to a €46 million overestimation which could have been prevented. Of course, the relevancy of this overestimation can be questioned as it is common-practice to start the period of estimation from the initiation of reimbursement. Applying this logic would shift the Sovaldi ‘Estimated BI’ line in Fig.
1 to the right and would cause a nearly equal overestimation from 01 Jan 2017 onwards due to declined market share.
In line with the overestimation due to a declined market share, inadequate correction for the introduction of new products could be a second reason. The Sovaldi BI estimations did not at all account for the introduction of new products in the same class whilst this was nearly inevitable as various manufacturers were in advanced stages of clinical development or regulatory approval [
30,
51,
52]. The ISPOR BI guideline, to which ZIN refers in their own guidance on BI, states that an attempt should be made to forecast introduction of new interventions for the chosen time horizon [
39]. Harvoni and Viekirax + Exviera were introduced nearly simultaneously and both were estimated to reach a market share of 35%. As our results show, the latter product only reached a small fraction of the estimated 35% whilst the former outperformed the expectations.
Of course, the patient estimates or the distribution over METAVIR scores could have been inaccurate. In addition, as DAA BI was in general overestimated and patient estimates are on the low side of estimations, we should consider the possibility that other forms of access restrictions were present. Stringent reimbursement criteria or volume caps issued to hospitals by payers might have been a factor in this regard as payers in The Netherlands have gained influence [
53].
Transitioning towards the role BI played in governing access for this specific case study, we like to reiterate that actual BI stayed well below the BI estimations ZIN deemed realistic. Consequently, the actual BI was also considerably lower than alternative scenarios postulated by ZIN where a ‘broader DAA indication’ would be adopted. This, according to the reimbursement dossiers, could theoretically lead to a BI of €1.78 billion [
12]. Such a wide call for treating all viraemic hepatitis C patients, culminating in the EASL’s 2018 guideline recommendation, has not been apparent in our data. Our data did, however, show that access is strictly governed by a positive reimbursement decision as a products’ BI is very low before it is reimbursed.
The reimbursement decision was, on average, taken 258 days after MA whilst the reimbursement dossier was published after on average 117 days. Price negotiations were conducted for all products for which the HTA report was used as guidance. There is, however, no report on the role that BI played in this process and whether BI estimations, which are known and proven to be uncertain, are necessary for either the reimbursement decision or the price negotiation. Additionally, one can extend this way of thinking with a debate on whether the 258 days of access restriction is worthwhile. This especially in light of the fact that DAAs are generally considered cost-effective [
12‐
16].
The implications of over- and underestimations differ for various stakeholders. Patients and manufacturers of the specified products, would probably incur no real loss due to an initial underestimation. It could potentially even facilitate the reimbursement process whereas the contrary could be true for an initial overestimation. For payers, however, an underestimation could be more troublesome than an overestimation as the former could cause direct and measurable budgetary deficits. We have no evidence to support or quantify these statements, but it is known that payers and manufacturers have confidential price negotiations where their BI estimates together with other parameters serve some purpose for bargaining [
54].
If there is a more aligned and agreed estimate of patient numbers, as the basis of a BIA, we believe that more accurate BI estimates are achievable. As is stated in the most recent BIA guideline, it is important to include treatment dynamics including new introductions and displacement effects as well as pricing dynamics to provide more accurate and meaningful BI estimates. This would allow for a more prominent role and value of BI in the decision-making process for reimbursement of different types of treatments (e.g. chronic and one-off).
Our study has various strengths. First, we used real world data to assess access using validated and monthly updated data. Our data covers in- and outpatient dispensing data, is irrespective of the healthcare provider or insurance company and, therefore, captures the entire Dutch DAA access data.
Second, we made an accurate representation of average treatment cost per patient using the distribution of several subpopulations. As our sensitivity analysis shows, not including patients with longer than typical treatment durations have a profound influence on outcomes.
Our study also has some limitations. First, several assumptions underlie population size estimates and the distribution of subgroup characteristics. We, therefore, crosschecked our estimates with the GIP-database which reports comparable figures [
33]. We, furthermore, aimed to illustrate the influence of population size and treatment costs on outcomes by means of sensitivity analysis.
Second, our Dutch scenario induces limitations regarding generalizability. Of course, the Dutch healthcare and reimbursement system is not directly comparable to others but the, albeit imperfect, method of BI estimation is rather similar [
21‐
23,
39].
Third, confidential discounts or rebates are not included in this analysis. In the Netherlands, the general outcome of price negotiations is published but the actual discount remains confidential. A study by Morgan et al. indicated that, in ten high-income countries (including The Netherlands), discounts are common and confidential discounts are most frequently in the 20–29% range but can also be substantially higher (> 60%) [
55]. We thus know that the actual BI, as costs to society, are lower than presented here. Yet, we based our analyses on BI estimations that disregard potential discounts and rebates. Lack of inclusion of pricing agreements is, therefore, not a concern.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.