A cost-effectiveness analysis of a preventive exercise program for patients with advanced head and neck cancer treated with concomitant chemo-radiotherapy

To investigate the cost-effectiveness of a preventive (swallowing) exercise program (PREP) compared to usual care (UC) in advanced head and neck cancer, data of two recent clinical trials in the Netherlands Cancer Institute were used [3, 4]. In both studies the protocol was approved by the Protocol Review Board of the Netherlands Cancer Institute - Antoni van Leeuwenhoek Hospital (NKI-AVL) and written informed consent was obtained from all patients before entering the study. All patients had advanced (stage III and IV) functional or anatomical inoperable head and neck cancer [5]. All received identical concomitant chemo-radiotherapy (CCRT), which consisted of 100-mg/m² Cisplatin as a 40 minutes intravenous (IV) infusion on days 1, 22, 43 and combined with radiotherapy, and identical intensive supportive care. Details about patients, methods, and clinical results in both studies have been published previously [3, 4, 6]. The patient characteristics are summarized in Table 1.

The UC data are derived from a multi-center randomized controlled trial (RCT), comparing intra-arterial (IA) and intravenous (IV) chemo radiation in advanced head and neck cancer [3]. Only the data of the 53 patients treated at the NKI-AVL, randomized in the IV arm, and still alive and disease free at 12 months were analysed for this cost-effectiveness study.

The PREP data are derived from a clinical trial conducted immediately following the former RCT. In this second RCT the effects of preventive strength and stretch exercises on (long-term) swallowing and/or mouth opening problems caused by CCRT, as an adjunct to UC, were assessed in 55 advanced head and neck cancer patients [4]. Before treatment all patients were randomized into two groups: an experimental group that was provided with the TheraBite^® Jaw Motion Rehabilitation System™ and a group receiving standard intervention (Standard group). The rationale and a detailed description of the exercises have been published previously [4]. In short, both regimes consist of comparable stretch and strength exercises to keep the swallowing musculature active before, during, and after CCRT, even when patients are not swallowing because of (naso) gastric tube feeding. Patients were encouraged to practice 3 times a day and to integrate the exercises into other daily activities at home. Participants were provided with verbal and written instructions prior to treatment, at which time they also started practicing, thus, when oral intake was not yet influenced by the treatment. Thirty-seven of the 55 included patients were still alive and disease free at 12-months. Since no significant differences in QoL, costs and functional outcomes were found between the two arms, for the present cost-effectiveness study both PREP arms were taken together [4].

The main outcome variables of interest for this cost effectiveness assessment were tube dependency at 12 months, and number of days patients were admitted to the hospital after completion of the CCRT in the first year. In the UC cohort, tube dependency was 13/53 (25%), and in the PREP cohort 1/37 (3%). The mean number of extra admission days in the hospital post-CCRT was 4.5 (SE 2.8) in the UC, and 3.2 (SE 1.2) in the PREP cohort.

A Markov decision model was developed to compare the PREP versus UC for advanced head and neck cancer. The model was constructed with three mutually exclusive health states: "complete remission", "recurrent disease" and "death" (death of cancer or other causes). The input regarding treatment success rates, and probability of recurrence were based on the published outcome data from our institute [6]. We assumed that the PREP has no direct influence on survival [7‐9]. Input on aspiration for UC was based on the empirical data, for PREP the value was assumed, based on the literature. The input of feeding substitutes and hospitalization were based on the above-described databases: the series of Ackerstaff et al. [3], as UC and that of Van der Molen et al. [4], as the PREP strategy. The model simulated the course of events in a hypothetical cohort of 1000 patients aged 55 years with a stage III or IV squamous cell carcinoma of the head and neck treated at the NKI-AVL. Possible complications from the treatment were modelled up to 1 year from the start of treatment. The cycle length of the model was one month, with a total simulated time horizon of 1 year. The analysis was performed from the health care perspective of the NKI-AVL. All costs were reported in year 2008 Euros (Table 2).

In the NKI-AVL the costs for treatment where measured by means of clinical pathways that patients followed when receiving CCRT. Besides treatment costs, feeding substitutes, pneumonia as adverse event and hospital days were derived from the NKI-AVL hospital charts and administration. The professional costs of PREP were derived from the Dutch Diagnosis Treatment Combination (DBC) "DBC-system" list, this tariff includes all possible involved disciplines in the PREP (19-49 hours for €3,252). Use of feeding substitutes (tube feeding) was calculated per disease severity stage from the two databases. It was assumed that from the patients needing tube feeding, 50% received a nasal tube and 50% received a gastronomy-tube (Table 2).

The quality of life of patients treated with CCRT was examined by Ackerstaff et al. [3]. For UC during treatment the QoL result of 7 weeks was incorporated (0.517), for UC after treatment, the QoL result of 12 months was taken (0.754) [3]. Assumptions as to how these results would be influenced by the PREP were based on published literature and informal expert elicitation (Table 2).

We programmed the model in Microsoft Excel (Microsoft, Redmond, WA) and validated it using various sensitivity analyses. Future costs and effects were discounted to their present value by a rate of 4% and 1.5% per year respectively, according to Dutch guidelines [10]. Incremental cost-effectiveness ratios (ICERS) were calculated by dividing the incremental costs by incremental quality adjusted life years (QALYs). Stochastic uncertainty in the input parameters was handled probabilistically, by assigning distributions to parameters (Table 2) [11]. Parameter values were drawn randomly from the assigned distributions, using Monte Carlo simulation with 1000 iterations. The results of the simulation of the hypothetical cohort of 1000 patients are illustrated in a Cost-Effectiveness (CE) plane, each quadrant indicates whether a strategy is more or less expensive and more or less effective [12]. Cost-effectiveness acceptability curves (CEACs) to show decision uncertainty are presented. CEACs show the probability that a pathway has the highest net monetary benefit, and thus is deemed cost-effective, for a range of Willingness to Pay (λ) values for one additional QALY, also referred as the ceiling ratio. This definition involves a Bayesian definition of probability i.e. the probability that the hypothesis ('PREP is cost-effective compared to UC') is true given the data. The two curves therefore always sum to 100% for one given value of λ [13]. In the Netherlands an informal ceiling ratio of €80,000 per QALY exists (Dutch Council for Public Health and Health Care 2006), and for preventive care programs of €20,000 per QALY. The National Institute for Health and Clinical Excellence in the United Kingdom uses a general ceiling ratio between £20,000-£30,000 per QALY. In this analysis, we use the Dutch threshold for preventive care programs, €20,000 per QALY.

We performed four one-way sensitivity analyses. The first two sensitivity analyses tested the robustness of the model outcomes against changes in the utility estimates (i.e. higher and lower estimates), as the current estimates are preliminary. For the two other sensitivity analyses, lower and higher cost estimates (€1,213 and €7,058) were imported for the resource use associated with paramedical care delivery in the rehabilitation program (for respectively 7-18 hours or 50-129 hours). In addition, we performed two two-way sensitivity analyses to test the most uncertain parameters, such as a variation in utilities in combination with the various DBC tariffs, and the variation of utilities in combination with the probability of aspiration.

For various scenarios regarding costs and QALYs, we also present the findings as cost-effectiveness acceptability frontiers that illustrate the probability of any intervention being optimal compared to its alternative. The optimal intervention is defined as the one with the highest expected net health benefit. Each cost-effectiveness frontier also illustrates the potential crossover when one intervention is substituted by another as the one with the highest probability of being optimal, and therefore provides useful information for policy makers.

Uncertainty in the cost-effectiveness results was also presented and used to inform future research priorities using Value of information analysis (VOI) analyses based on the expected value of perfect information (EVPI). VOI can be used to support decisions on focus and design of further research, assuming that additional evidence on the relevant aspects can be desired, but that the amount and specific requirements for further research will depend on the parameters which are causing the most uncertainty [14]. Generally, information is valuable when there is great uncertainty surrounding a decision and when that decision likely affects a large number of people in a meaningful way. If one had perfect information about the risks and benefits of a particular technology, decision makers in theory would always be able to make correct choices regarding the use of the technology. The difference between the expected net benefit obtained using perfect information and the expected net benefit obtained in the presence of uncertainty (that is, the maximum expected net benefit obtained with less than perfect information) is known as the expected value of perfect information (EVPI). It can be interpreted as the maximum amount the decision maker would be willing to spend to obtain perfect information [15].

The total health care costs (treatment costs + preventive exercises) per patient were: €42,271 for the preventive exercise program (PREP), and €41,986 for usual care (UC). The quality adjusted life years amounted to: 0.77 (PREP), and 0.68 (UC). The difference in costs per QALY of the PREP strategy compared to the UC strategy amounted to €285 (= 0.7% of the total treatment costs). In comparison to UC, the PREP for advanced head and neck tumours costs €3,197 per QALY gained and was found to be more effective and slightly more costly (Table 3).

When focusing on quality adjusted survival, the PREP has a higher probability of being cost-effective compared to UC, as long as the willingness to pay threshold for 1 additional QALY is at least €3,200/QALY (Figure 1 and 2). At the prevailing threshold of €20,000/QALY the probability for PREP being cost-effective compared to UC was 83%.

The sensitivity analysis using lower utility estimates resulted in an incremental cost-effectiveness ratio (ICER) of €6,393/QALY; higher utilities resulted in an ICER of €2,131/QALY. The sensitivity analysis considering a lower resource use (fewer hours of professionals) resulted in UC being dominated by PREP, i.e. PREP is more effective and less costly than UC. Modelling a higher resource use (more hours) for the PREP resulted in a higher ICER of €45,906/QALY (Figure 3). The results of the two-way sensitivity analyses are listed in Table 4, the majority of the analyses resulted with an ICER below the ceiling ratio of €20,000/QALY.

At a ceiling ratio of €20,000/QALY the probability for PREP being cost-effective compared to UC was 83%, which shows a considerable decision uncertainty using the current available data. The EVPI for the base case resulted in €398,063, providing the upper boundary for investing research funds in further clinical trials to obtain perfect information on the cost-effectiveness of PREP versus UC. The EVPI demonstrated potential value in undertaking additional research to reduce the existing decision uncertainty (Figure 4).