FormalPara Key Points for Decision Makers

Respondents with stage III melanoma prefer adjuvant immunotherapy that reduces their risk of recurrence, reduces treatment adverse events and costs less.

Almost three-quarters of respondents chose adjuvant immunotherapy over no adjuvant immunotherapy. Preferences were found to statistically differ by the amount of immunotherapy experience and being female; women were significantly less accepting of treatment. Respondents were willing to pay between AU$794 and AU$4266 (US$527–2833) depending on income, for adjuvant immunotherapy per year to reduce their risk of immediate death and recurrence by 1%.

Health policymakers should consider respondents’ trade-offs and remove financial barriers to enable access to effective therapies.

1 Introduction

Melanoma incidence continues to increase globally [2]. In Australia, the lifetime risk of melanoma rose from 1 in 46 people in 1982 to an estimated 1 in 15 in 2020 [3]. Over the past decade, advances in treating stage IV metastatic melanoma using checkpoint inhibitor immunotherapy or BRAF-targeted therapy have revolutionised care, with 5-year overall survival now greater than 50% [4, 5]. Such gains have been applied in the adjuvant stage III setting. In an era before immunotherapy, 10-year overall survival for resected stage IIIA–D melanoma was 24–88%, worsening with sub-stage [6].

However, adjuvant checkpoint inhibitor immunotherapy (adjuvant immunotherapy) is associated with possible short- and long-term side-effects, including fatigue, skin rash, arthralgia, diarrhoea, hormone deficiencies (thyroid, pituitary and pancreatic) and other immune-related adverse effects. In the post-operative setting, respondents without active overt residual (metastatic) disease might be reluctant to commence a 12-month regimen of treatment (i.e. adjuvant therapy) to prevent future melanoma recurrence. Previous research suggests around one-tenth of respondents with stage III resected melanoma are reluctant to commence immunotherapy, with a further fifth requiring early cessation of treatment due to toxicity [7]. Additionally, such treatments may be costly, as in Australia adults receiving immunotherapy may have substantial out-of-pocket costs [8]. Therefore, it is imperative to understand and quantify what treatment risks respondents are willing to accept to gain benefits (trade-offs).

Generating meaningful information to inform policy and allocation of healthcare resources requires quantification of respondents’ preferences and should be considered by regulatory agencies [9, 10]. Consistent with prior research from our group, key drivers for respondents with resected stage III melanoma considering adjuvant immunotherapy included extending life, uncertainty about treatment efficacy, appetite for risk, treatment burden and out-of-pocket (OOP) costs [7]. Discrete choice experiments (DCEs) are commonly used to quantify preferences for healthcare services or treatments, exploring the trade-offs respondents are willing to make [11]. A DCE is a series of questions (a survey) showing two or more alternatives, described by a set of attributes (factors) with varying levels. Multiple choice tasks or scenarios are presented to respondents, in which the levels of each attribute are varied. Respondents consider all the information presented in the choice task and are asked to select the alternative they most prefer [12, 13]. This generates a dataset that enables the analyst to quantify preferences, the relative importance of attributes, and trade-offs between attributes for the group as a whole [14].

This study aimed to quantify adults’ preferences for adjuvant checkpoint inhibitor immunotherapy for resected melanoma and the influence of varying levels of key attributes, and baseline characteristics. Additionally, calculation of trade-offs, marginal rates of substitution (MRS) for adjuvant immunotherapy and marginal willingness to pay (mWTP) per year are reported.

2 Methods

2.1 Respondent Recruitment and Setting

Australian adults who self-reported having surgery for resected stage III melanoma were invited to participate in an online DCE about adjuvant immunotherapy. Respondents were excluded if they were younger than 18 years, had stage I–II or stage IV melanoma or did not consent. Respondents did not receive payment for survey completion. The DCE was designed and conducted according to best practice methods [15,16,17,18]. The survey was hosted on the Qualtrics survey platform (version July 2021) and disseminated by eight Australian melanoma consumer and professional groups, including the Melanoma Institute Australia, Melanoma Patients Australia, Melanoma Research Victoria, the Melanoma and Skin Cancer Advocacy Network, Melanoma WA, Skin Cancer Tasmania, Melanoma and Skin Cancer Trials, and the Hunter Melanoma Foundation. All respondents provided written consent, and the Declaration of Helsinki underpinned this study. Data collection occurred between July and September 2021, with ethical approval granted by the Sydney Local Health District Human Research Ethics Committee (HREC/18/RPAH/736).

2.2 Selection of Attributes and Levels

Attributes and levels were informed by existing literature [19,20,21,22,23,24,25,26], focus groups and interviews with respondents with resected stage III melanoma (n = 24) [7], expert consultation, adjuvant immunotherapy trials of resected stage III melanoma, and pilot testing. As per Table 1, the DCE included six attributes with two to three levels each.

Table 1 DCE attributes and attribute levels
Full size table

2.3 Study Design

This DCE presented respondents with two treatment alternatives and asked them to choose one or the other. The DCE included an explanation of the choice questions, definitions of the attributes and levels, 12 hypothetical choice tasks and socio-demographic questions. Respondents were provided with a brief description of the choice context, i.e. “As a person with resected stage III melanoma, we ask you to consider treatment with adjuvant immunotherapy, a drug that uses the body’s immune system to fight any leftover melanoma cells” (Fig. 1).

Fig. 1
figure 1

Example of Discrete Choice Experiment task

Full size image

Respondents were provided with a brief description of the choice context: i.e. “As a person with resected stage III melanoma, we ask you to consider treatment with adjuvant immunotherapy, a drug that uses the body’s immune system to fight any leftover melanoma cells.”

A literature review and in-depth discussion with investigators drawing on their clinical knowledge and experience informed the alternatives, attributes and levels included in the choice set. A D-efficient design was generated using Ngene version 1.2.1 software, consisting of two labelled alternatives of 72 choice questions in six blocks of 12 questions. Options consisted of drug treatment (adjuvant immunotherapy) versus no adjuvant immunotherapy. In both scenarios, respondents undergo regular monitoring, including clinical assessments, imaging and blood tests. Respondents’ socio-demographic and clinical factors collected included age, gender, have children (yes/no), education, annual pre-tax household income, melanoma sub-stage and treatment received. Accordingly, the author hypothesised that the following traits might influence treatment choices: age, disease sub-stage, children and income. Respondent characteristics were coded numerically with ‘unknown’ melanoma sub-stage coded as missing data.

As per recommended guidelines, selecting the final six attributes was based on stakeholder feedback and ease of survey completion [15, 27]. Adjuvant immunotherapy trials of respondents with resected stage III melanoma informed the levels for four of the six attributes, namely the chance of 3-year melanoma recurrence and the three treatment toxicity attributes. These included the chance of melanoma recurrence at 3 years (levels 36%, 42%, 50%, 56% or 62%); the chance of a mild adverse event (AE) (levels 0%, 1%, 12% or 37%); the chance of a permanent AE (levels 0%, 1%, 10% or 14%); and the chance of fatal AE (levels 0%, 1% or 3%). For the remaining two attributes, drug regimen (levels 0-, 3- or 4-weekly treatment) was as per approved treatment guidelines [28, 29], and respondent OOP cost (levels AU$0, $750, $1500 or $6000) was informed by focus groups, interviews and the literature [8]. Following extensive consultation with medical oncologists, the levels for no adjuvant immunotherapy AEs (mild, permanent and fatal) were set at 0%.

Annual household income consisted of four categories (≤ AU$43,999, AU$44,000–$87,999, AU$88,000–$175,999 and ≥ AU$176,000) for analysis purposes, splitting categories into two, low-income (combining the two lower-income ranges, ≤ AU$87,999) and high-income (combining the two higher-income ranges, ≥ AU$88,000). The authors hypothesised that respondent OOP costs would likely interact with the annual pre-tax household income; therefore, an interaction between income and cost was created, and coded in the design to respecify the model. Further, a priori, the author hypothesised that respondents would be more likely to prefer adjuvant immunotherapy that meant less frequent treatment (i.e. 4-weekly versus 3-weekly); reduced their chance of recurrence; lowered the number of mild, permanent and fatal AEs; and reduced OOP costs.

The survey design in Ngene included six blocks of 12 choice tasks. The design had two constraints; first, the AU$6,000 ‘out-of-pocket cost per year’ was not displayed for 4-weekly immunotherapy treatments, as more frequent drug delivery (i.e. 3-weekly) was likely to be associated with the higher cost. Second, to address dominance across the alternatives, the ‘chance of melanoma recurrence’ could not be equal (i.e. 50%) across both alternatives. The pilot and main DCE study used this design. Figure 1 provides an example DCE choice question.

2.4 Pilot Study

The DCE was piloted on respondents with resected stage III melanoma (n = 21) to assess face validity and acceptability. Additionally, 10 of the 21 respondents undertaking the pilot study completed face-to-face ‘think aloud’ feedback with the researcher. The ‘think aloud’ process allows respondents to verbalise their thoughts when completing any task [30]. Respondents verbalised their reasoning for choosing choice tasks, discussed their understanding of the questions and the survey look and feel, and confirmed face validity. Examples of survey refinement included revision of clinical terms [e.g. diabetes or hypophysitis (loss of pituitary gland function)] and removal of treatment options (e.g. interferon). The pilot study results generated prior parameter estimates for the main DCE. The pilot results were not included in the main study results; however, six respondents completed the pilot and main study DCEs.

2.5 Analyses

DCE survey analysis assesses the relationship between the choices made and the differences in the attributes used to describe profiles (i.e. the distinct combinations of attributes levels displayed) resulting in those choices [18, 31, 32]. DCE analyses used NLOGIT version 6 software [33]. Analyses initially used a multinomial logit (MNL) model; however, a limitation of this model is the use of the same preference weights (homogeneous) applied to all respondents [16]. Further analyses used mixed multinomial logit (MMNL) models; assumptions accommodated the data’s cross-sectional panel structure and preference heterogeneity, meaning one or more parameter estimates are randomly distributed among individuals [15, 16]. Additionally, the model choice was based around the assumption that error terms were independent and identically distributed, preference heterogeneity and how flexible respondents were about substituting between options [34]. A linear functional form was used, and testing of other functional forms completed; however, they did not improve the model fit. Further, sensitivity analyses of the model specification were considered, resulting in testing the plausible variations in the model assumptions, for example, subgroup analyses. MNL analyses are not presented as they did not add any explanatory value over the final MMNL analyses.

To minimise respondent fatigue, choice sets were limited to 12 per person. Additionally, MMNL model assumptions and the use of NLOGIT software accounted for the non-independence between observations completed by the same respondent [33].

The MMNL expresses the utility to person n from choosing alternative j in choice scenario t:

$$U_{njt} = \left[ {\beta + \eta n} \right]X_{njt} + \varepsilon_{njt} ,$$

where β is the weighting of the attributes, ηn is a random term capturing non-observable person-specific variation, such as preferences. The random error ε includes random variation across discrete choices. The following equation expresses the utility function (i.e. the total satisfaction received from consuming a good or service) [35] of the final MMNL model:

$$\begin{gathered} {\text{Vimmunotherapy[Imm] = }}\beta {\text{Imm}} + \beta 1{\text{ThreeweeklyRegimen + }}\beta 2{\text{ChanceofRecurrence}} \hfill \\ { + }\beta 3{\text{One\% ChanceofMildAE + }}\beta 4{\text{Twelve\% ChanceofMildAE + }}\beta 5{\text{ChanceofPermAE}} \hfill \\ { + }\beta 6{\text{ChanceofFatalAE + }}\beta 7{\text{cost*lowincome[costL] + }}\beta 8{\text{cost*highincome[costH]}} \hfill \\ \end{gathered}$$

Of the six attributes, two categorical attributes were dummy coded and did not reach statistical significance (p > 0.05) in the MNL model (i.e. the chance of mild AEs and drug regimen). The remaining four attributes were treated as continuous (linear) (i.e. the chance of recurrence at 3 years, fatal and permanent AEs, and OOP cost). All 15 parameters were specified as random, and if not statistically significant (i.e. p > 0.05), parameters were re-specified as non-random. The final MMNL model had 12 random parameters (eight normally distributed parameters and four cost parameters that used a constrained triangular distribution) [36] and three non-random parameters. All parameters were specified as main effects, except the cost parameters, which were interacted with low- and high-income categories; the model was run using 10,000 Halton draws. The log-likelihood ratio test, McFadden’s pseudo-R [2] value and normalised Akaike’s information criterion (AIC/N) assessed the goodness of model fit.

Socio-demographic and clinical characteristics hypothesised to influence choices included age, gender, children yes/no and melanoma sub-stage (i.e. American Joint Committee on Cancer, AJCC, 8th edition stage IIIA-D) [6] were interacted in the model with the alternative specific constant. These were added to models, and then removed if the improvement in model fit was not statistically significant (i.e. p > 0.05). Melanoma sub-stage was added to the model as the likelihood of recurrence increases with more extensive disease. Authors hypothesised that as the chance of recurrence increased, so would the preference for immunotherapy. Coefficients of the parameters, and post-estimation odds ratios (OR) with their 95% confidence intervals (CI) and p-values are reported. Odds ratios are interpreted regarding the likelihood of preferring adjuvant immunotherapy per specified change in the level of that attribute. Thus, the smaller the OR, the less likely the respondent is to select treatment per each increase in the level of that attribute (OR < 1); conversely, the larger the OR, the more likely the respondent is to prefer treatment per each decrease in the level of that attribute (OR ≥ 1).

Calculation of the marginal rate of substitution (MRS) and marginal willingness to pay (mWTP) used the delta method (a method used to measure the variability of data, and the variability between variables often used to calculate the 95% confidence intervals) [17]. The MRS was calculated by dividing the coefficients of the attributes of interest by each other, providing information on the rate respondents were willing to trade levels of one attribute for another [37]. Similarly, calculation of the mWTP by dividing the coefficient of an attribute of interest by the cost coefficient, providing an individual’s maximum price to trade levels of that attribute [38]. Cost attributes used a constrained triangular distribution [36]. MRS values for trading the chance of a permanent AE for a reduction in the chance of 3-year recurrence, and the chance of 3-year recurrence for a reduction in the chance of a fatal AE, are reported. mWTP for a 1% lower chance of recurrence and a 1% lower chance of a fatal AE were reported separately for respondents with lower and higher household incomes.

The ESTIMATE checklist informed our approach for reporting a DCE (Table 2) [18].

Table 2 ESTIMATE checklist for DCEs
Full size table

3 Results

3.1 Respondents

Analysis was undertaken using complete survey data from 116 respondents. We excluded 28 duplicates (more than one response by the same respondent), 35 incomplete responses and 90 surveys from respondents who were ineligible or did not consent (Fig. 2). The final 116 respondents had a median age of 60 years (range 26–82 years); 53% were female, and 79% had received adjuvant immunotherapy. Most respondents (84%) resided in New South Wales or Victoria and were aged 45–64 years (Table 3).

Fig. 2
figure 2

Flow diagram of patient DCE recruitment

Full size image
Table 3 DCE respondent characteristics
Full size table

3.2 Mixed Multinomial Logit Model Results (MMNL)

Respondents chose adjuvant immunotherapy over no adjuvant immunotherapy in 70% of the choice tasks. Over a third of respondents (n = 41) always chose drug treatment, i.e. drug treatment was selected in all 12 of the 12 choice tasks. MMNL results suggest respondents were less likely to prefer treatment that increased the absolute risk of recurrence (OR 0.76 per 1% higher risk of recurrence over 3 years, 95% CI 0.70–0.83, p < 0.001); or treatment that had a higher chance of fatal AEs (OR 0.60 per 1% increase in the chance of fatal AE, 95% CI 0.44–0.80, p = 0.006); or treatment that had a higher chance of permanent AEs (OR 0.94 per 1% higher chance of permanent AE, 95% CI 0.89–0.99, p < 0.05). Further, for each $1000 increase in annual OOP costs, respondents with incomes that were low compared to high were less likely to prefer adjuvant immunotherapy, (OR 0.63, 95% CI 0.47–0.85, p < 0.01) and (OR 0.84, 95% CI 0.15–4.86, p = 0.06) (Table 4 and Fig. 3).

Table 4 Preferences for adjuvant immunotherapy relative to no adjuvant immunotherapy in resected stage III melanoma, MMNL model (n = 116)*
Full size table
Fig. 3
figure 3

Odds ratio of attributes for adjuvant immunotherapy versus no adjuvant immunotherapy treatment

Full size image

Of the respondent characteristics included in the model (age, have children yes/no, melanoma sub-stage and annual household income), preferences were found to statistically differ by gender, with females having a lower preference for immunotherapy than males (OR 0.10, 95% CI 0.02–0.66, p = 0.02) (Table 4).

Further, the model was rerun using subgroups of the sample. Subgroup analyses investigated the preferences of respondents who had received immunotherapy (n = 92) versus those who had not (n = 22). Respondents treated with immunotherapy were grouped into those ceasing immunotherapy because of toxicity (n = 17) versus those who continued treatment (n = 75) (Table 4).

Compared with the total sample, respondents who ceased immunotherapy because of toxicity (n = 17) chose adjuvant immunotherapy treatment over no adjuvant immunotherapy 63% of the time, 7% less than the whole cohort. Other socio-demographic and clinical characteristics were similar for the entire group, except for zero stage IIIA disease, and 10% more respondents reported the highest income level (≥ AU$176,000). Compared with the total sample, preferences for adjuvant immunotherapy were similar, except for a stronger preference to lower the chance of a permanent AE (OR 0.77, 95% CI 0.61–0.96, p = 0.02). For this cohort, the reduction in the likelihood of a fatal AE (OR 0.46, 95% CI 0.15–1.38, p = 0.16), and each $1000 increase in annual OOP costs, were less of a concern with incomes that were lower (OR 1.18, 95% CI 0.74–1.89, p = 0.10) rather than higher (OR 2.08, 95% CI 0.86–5.01, p = 0.94). Further, preferences did not statistically differ by gender (Table 4).

Compared with the total sample, survey respondents that underwent immunotherapy and did not stop it due to toxicity (n = 75), chose adjuvant immunotherapy 76% of the time, which was 6% higher than the total sample. Compared with the whole sample, a reduction in the chance of a permanent AE was of less concern (OR 0.97, 95% CI 0.91–1.04, p = 0.39). Similarly, the acceptability of an increase in the chance of experiencing a mild AE from 1% to 37% concerned this group less (OR 1.17, 95% CI 0.58–2.38, p = 0.66) (Table 4).

For the 22 respondents who did not receive adjuvant immunotherapy, mild, permanent or fatal AE attributes did not significantly affect preferences. Of those not treated with immunotherapy, almost a quarter would have been eligible for treatment (i.e. stage IIIB-D melanoma). Further, preferences did not statistically differ by gender (Table 4).

3.3 Trade-Offs (Marginal Rates of Substitution) and Marginal Willingness-to-Pay Estimates

On average, respondents were willing to accept an additional 4% chance of a permanent AE to reduce their absolute risk of recurrence at 3 years by 1% (95% CI – 10.44 to 23.41). Respondents were willing to accept an extra 2% chance of recurrence at 3 years to lower their chance of a fatal AE by 1% (95% CI – 12.35 to 34.54) (Table 5).

Table 5 Trade-offs (marginal rates of substitution) and marginal willingness-to-pay estimates for MMNL model
Full size table

Respondents with lower household income were, on average, willing to pay an extra AU$794 (95% CI – AU$741 to AU$1,745) (US$527, 95% CI – US$492 to US$1159) per year and, for higher income, an additional AU$2190 (95% CI −AU$1651 to AU$4,571) (US$1454, 95% CI – US$1096 to US$13,035) for each 1% reduction in their absolute risk of recurrence at 3 years with adjuvant immunotherapy. Further, respondents with lower income were, on average, willing to pay AU$1531 (95% CI – AU$1527 to AU$8417) (US$1017, 95% CI – US$1014 to US$5589), for a 1% lower chance of fatal AE compared with a higher income, AU$4266 (95% CI – AU$3484 to AU$13,997) (US$2832, 95% CI – US$2314 to US$9295) (Table 5). Additionally, respondents were willing to accept an increase in their likelihood of experiencing a mild AE from 1% to 37% (OR 2.06, 95% CI 1.13–3.78, p = 0.02) in return for adjuvant immunotherapy (Table 4).

4 Discussion

Results suggest that respondents with resected stage III melanoma had an overall preference for adjuvant immunotherapy over no adjuvant immunotherapy, with 70% choosing adjuvant treatment across the choice tasks. Respondents in our sample prefer treatments with an absolute reduction in the risk of recurrence, reduced probabilities of fatal and permanent AEs, and lower yearly OOP costs. However, respondents accepted a substantial increase in their chance of experiencing mild AEs in return for adjuvant immunotherapy. Respondents were willing to trade a slight increase in the chance of a permanent AE to reduce their chance of recurrence at 3 years and a slight increase in the likelihood of 3-year recurrence to lower their chance of a fatal AE.

Subgroup analysis of the 17 respondents who ceased adjuvant immunotherapy for toxicity revealed that lowering the chance of a fatal AE was of less concern; however, this finding may be due to the small sub-sample size. Additionally, despite ceasing treatment because of toxicity (n = 17), this group appeared willing to choose adjuvant immunotherapy over no adjuvant immunotherapy almost as frequently as the total sample, 63% of the time compared with 70% of the time. Further, the 75 respondents who received immunotherapy without cessation because of toxicity chose adjuvant immunotherapy 76% of the time, 6% more than the entire group. This finding suggests respondents receiving immunotherapy treatment without AEs, still chose no adjuvant immunotherapy in almost a quarter of the choice tasks. Preferences were found to statistically differ by gender (p = 0.02), with females having a lower preference for immunotherapy treatment than males. For respondents who had not received immunotherapy treatment, AEs did not drive their treatment choices; suggesting those who have experienced AEs possess additional information to inform decisions.

Prior DCE research confirms that respondents with stage III–IV melanoma prefer to avoid fatal AEs, such as colitis, and permanent AEs, including changes in hormone gland function [23, 24]. Similarly, US respondents with self-reported unresectable or metastatic melanoma preferred to maximise their objective response rate and minimise severe treatment toxicities in the stage IV setting [20]. Further, a conference abstract of an Australian DCE of stage III–IV melanoma respondents asked about single and double immunotherapy and targeted therapy preferences [25]. Likewise, results reported that efficacy attributes were most preferred, and adverse events and higher costs were least preferred. Our results align with prior melanoma research, confirming that respondents never treated with adjuvant immunotherapy had fewer concerns about AEs [21]. Lastly, our findings align with two earlier DCEs reporting patients with resected melanoma selected the ‘no treatment’ option, 32% [26], and 29% of the time [24], compared with 30% in our resected stage III melanoma sample.

Regarding demographic characteristics, females had a significantly lower preference for immunotherapy than males; this differed from prior melanoma DCEs showing preferences for treatment did not differ by gender [21]. Conversely, other cancer research supported our findings of gender differences for cancer treatment, with women with resected stage I–III non-small cell lung cancer more likely to decline adjuvant chemotherapy than men [39]. Additionally, the authors acknowledge that these findings may be relevant in other cancers where adjuvant immunotherapy is routine care.

To our knowledge, this DCE is the first to investigate and publish preferences for adjuvant checkpoint inhibitor immunotherapy alone, for respondents with resected stage III melanoma. Interestingly, respondents preferred to avoid the short-term risk of death from melanoma by lowering their likelihood of experiencing a fatal AE over their longer-term risk of death, such as reducing their chance of recurrence in 3 years. One possible explanation for respondent behaviour may be that recurrence was viewed differently to death. In stage IV melanoma, immunotherapy may be curative, unlike other cancer treatments such as chemotherapy. Therefore, recurrence might not result in death, whereas a fatal AE ensures immediate loss of life. Concordantly, respondents willingly accepted a slight increase in recurrence at 3 years to lower their chance of a deadly AE. Findings may explain the potential trade-offs respondents were willing to accept to lower the likelihood of fatal treatment toxicity and reduce the risk of recurrence.

Given the increasing cost of cancer care, health service providers need to rationalise costly cancer treatments and consider respondent OOP costs. In Australia, the government reimburses adjuvant immunotherapy drug costs for respondents with resected stage III melanoma. However, respondent OOP costs may be considerable, meaning therapy has a financial cost [8]. Consequently, unlike many earlier melanoma DCEs, our study included a respondent OOP cost attribute, as prior qualitative research identified this as an important consideration in treatment decision-making [7].

Further, lower- and higher-income respondents were willing to pay an additional AU$794 (US$527) and AU$2190 (US$1454), per year respectively, for adjuvant immunotherapy to reduce their absolute risk of recurrence at 3 years by 1%. Findings support that governments and healthcare payers should focus on removing respondents’ financial barriers, enabling equitable access to effective treatments that align with respondent preferences, such as adjuvant immunotherapy.

Study limitations include the potential for selection bias as most of the respondents recruited to this study received adjuvant immunotherapy treatment. Additionally, research has shown that respondents prefer what they know instead of the unknown, which may account for most respondents choosing immunotherapy [40]. Second, the authors hypothesised that as the sub-stage of disease increases, so might the preference for adjuvant immunotherapy. Unfortunately, over a quarter of respondents reported their melanoma sub-stage as ‘unknown’, resulting in the coding of these responses as missing data. Such misinformation might contribute to the sub-stage not being statistically significant at the 5% level. Additionally, it is acknowledged that subgroup sample sizes were small. Specifically for the 17 respondents who ceased immunotherapy due to toxicity, the authors accept it is difficult to conclude with certainty these findings, which are at best hypothesis generating and require further research with larger samples, and control for differences in scale [41, 42]. Nevertheless, the design generated a small D-error of 0.001708; the smaller the number, the better the design [16], proving that the sample size was adequate to quantify preferences and identify statistically significant attributes. Further, the attribute levels for AEs for the no adjuvant immunotherapy alternative were assumed to be zero. However, adjuvant trials of checkpoint inhibitor immunotherapy demonstrate that a small percentage of respondents receiving placebo report a substantial number of AEs [43]. Lastly, we did not include tests of internal validity [44].

Globally, rapid changes in the treatment of earlier stage melanoma continue. Findings from two phase 3 randomised clinical trials have demonstrated a significant reduction in the risk of recurrence with anti-PD1 checkpoint inhibitors, with a relative reduction in the risk of recurrence of approximately 40% versus no therapy [43, 45]. These findings have led to the approval of both nivolumab and pembrolizumab to treat adjuvant melanoma. Authors deliberately focused on issues about adjuvant anti-PD1 checkpoint inhibitor immunotherapy for respondents with resected stage III melanoma and the potential trade-offs of treatment benefits and risks. Following the recent results indicating the benefits of adjuvant immunotherapy in high-risk stage II disease, this warrants further research [46]. Additionally, preferences for immunotherapy in stage IV are likely to differ and require further investigation.

5 Conclusions

This DCE is the first to investigate the preferences for adjuvant checkpoint inhibitor immunotherapy for respondents with resected stage III melanoma. Most respondents chose adjuvant immunotherapy over no adjuvant immunotherapy, especially if treatment affords high efficacy and improves safety. Preferences were found to statistically differ by the amount of immunotherapy experience and being female. Further, those reporting treatment toxicities preferred to reduce the risk of permanent AEs.

Evidently, trade-offs demonstrate that respondents accept an increase in the odds of disease recurrence to lower their immediate risk of death. Respondents would trade a higher risk of permanent AEs to reduce their absolute risk of recurrence. Consideration of respondent preferences and trade-offs for adjuvant immunotherapy following surgery for melanoma may assist policymakers regarding access and subsidisation of effective yet expensive treatments.