The Feasibility of Combining Cost-Effectiveness, Bayesian Meta-Analysis and Meta-Regression Techniques When Evaluating Melanoma Treatment Methods

Abstract

Background

Melanoma treatment encompasses surgery, chemotherapy, radiotherapy, immunotherapy, and targeted therapy. Despite declining mortality due to therapeutic advances, increasing incidence underscores the growing importance of cost-effectiveness analyses. The integration of cost-effectiveness evaluation with Bayesian meta-analysis and meta-regression may enhance the synthesis of heterogeneous evidence.

Methods

A systematic literature review was conducted to identify studies reporting quality-adjusted life years (QALYs) and direct treatment costs for melanoma in PubMed, MEDLINE, and MEDLINE Ultimate. Bayesian meta-analysis with non-informative priors was applied to derive pooled estimates and quantify uncertainty.

Results

A total of 272 studies were identified, of which only a limited subset met the inclusion criteria. The studies exhibited substantial heterogeneity in treatments, populations, time horizons, perspectives, and costs. Meta-analysis based on the effects provided in one study showed that chemotherapy demonstrated lower effectiveness, as measured by QALYs, compared with treatment sequences such as Anti-PD1 → Bi-TT and Bi-TT → Anti-PD1 in both BRAF-mutated and wild-type populations. Lower costs were associated with chemotherapy, ipilimumab, and Anti-PD1 relative to combination and sequential regimens. Meta-analysis based on the effects provided in another study showed that treatment costs increased with advancing disease stage (≥IIIA) and exhibited greater variability in advanced melanoma.

Conclusion

The combined methodological approach enhances the robustness and comparability of estimates. The observed variability in both clinical effectiveness and costs, across treatment strategies and disease stages, has important implications for healthcare decision-making and resource allocation.

Introduction

According to the World Health Organisation in 2022, melanoma had an incidence of 331,722 cases worldwide. It is the 17th most prevalent malignancy,¹ with the age-standardised incidence rate equal to 3.2/100,000. The incidence rate values for Europe and North America were higher, equal to 10.4/100,000 and 16.3/100,000, respectively, for Oceania, 29.8/100,000.^1,2 The highest incidence rate was recorded in Australia with the value of 37.0/00,000.

The incidence rates were higher in males than in females.^1,2 In the total world population, they were equal to 3.7/100,000 and 2.9/100,000 for males and females, respectively. In North America, 19.0/100,000 and 14.3/100,000; in Oceania, 37.4/100,000 and 23.1/100,000; in Europe, 11.0/100,000 and 10.1/100,000; and in Asia, 0.44/100,000 and 0.40/100,000 for males and females, respectively. Therefore, in Europe the incidence rates for males and females were closer to each other, and in Asia the incidence rates were low, so the difference between females and males that can be observed in the total world population was a consequence of the differences observed in North America and Oceania.

The reported risk of developing melanoma based on data from the European Commission by age 74 is 1:74 for women and 1:66 for men.² The likelihood of five-year survival is high, typically ranging from 80% to 90%, for most European countries. In some countries, ie, Slovakia, Croatia, Estonia, and Lithuania, the percentage is lower (70%–80%). In Latvia and Poland, it falls between 60% and 70%. In Bulgaria, it is equal to 50%.²

Regarding mortality, melanoma is 22nd on the list of malignancy-related deaths. The global age-standardised mortality was equal to 0.53/100,000.¹ Mortality rates were higher for North America at 1.1/100,000, for Europe at 1.5/100,000, and for Oceania, 2.3/100,000. The mortality rates were also higher in males compared to females. In North America, they were equal to 1.5/100,000 and 0.74/100,000; in Oceania to 3.0/100,000 and 1.7/100,000; in Europe to 1.9/100,000 and 1.2/100,000; and in Asia to 0.24/100,000 and 0.19/100,000 for males and females, respectively. Therefore, the differences between females and males were larger in North America and Oceania. Over time, mortality rates decreased, and incidence rates increased, which can be attributed to both better detection methods and higher intensity of risk factors related to lifestyle.³

The main risk factors for melanoma are exposure to ultraviolet radiation from sunlight, tanning bed lamps, sunburn during childhood, having multiple (>100) pigmented nevi on the skin, presence of atypical nevi, melanoma or skin cancer in a family history, a melanoma diagnosed previously, other malignancy in a medical history, especially thyroid cancer, soft tissue cancer, cervical cancer, and acute myeloid leukemia, high lifetime intake of alcohol, a weakened immune system, as a result of HIV infection, transplant, Non-Hodgkin’s lymphoma or chronic lymphocytic leukemia for example, older age, fair complexion,⁴ light eye color, freckles, skin phototype II according to the Fitzpatrick scale (which applies to skin that burns easily and is minimally susceptible to tanning), also skin type I and III when compared to IV. Regarding genetics, CDKN2A, CDK4, BAP1, MITF, TERT, ACD, TERF2IP or POT1 mutations, one or more variants of MC1R, but also having blond or red hair.⁵ The risk of melanoma is also higher among patients with Xeroderma pigmentosum (XP).⁶

Various types of treatment are applied to melanoma, including surgery, chemotherapy, radiation therapy, immunotherapy and targeted therapy.⁷ Also, vaccine therapy is tested in clinical trials as a new type of treatment. The types of treatment are evaluated in terms of health economic evaluation with measures like quality-adjusted life years (QALY), direct costs expressed with the use of a currency as a unit of measure and incremental cost-effectiveness ratio (ICER). QALY combines the length of life and its quality into a single indicator.⁸ It is calculated by multiplying the utility value, which is set at 0 for death and 1 for perfect health by the duration of life. The duration of life is estimated on the basis of the results from clinical trials or from models extrapolating the results of clinical trials. Models, like Markov models,⁹ allow for extending the estimation beyond the timeframes of clinical trials, which are as short as necessary. The utility values come from polls with respondents asked about equivalence between living shorter in perfect health or longer in a specified medical condition, about choices between pairs of health states and about quality of life with the use of questionnaires.¹⁰

ICER combines the information about the direct costs and health outcome,¹¹ which frequently is QALY. It is calculated then as a ratio dividing the difference between the costs of treatments by QALY. Therefore, in this case it evaluates the cost of a single QALY gained.

QALY is essentially an effect size measure that can be used to compare different treatments and different medical conditions. Values of effect size measures from various studies or from various groups examined in a study can be combined and analysed further with the use of meta-analysis.^12,13 It allows for estimation of total effect size taking the results of individual studies and/or the results acquired in examined groups into account and taking the differences between the results into consideration. Also, additional analysis, referred to as meta-regression, is possible to gain additional insight into factors possibly explaining the differences between the results acquired. The necessary condition for meta-analysis to be used is to provide the effect size measure and its standard error, which is a measure of uncertainty regarding the effect size calculated.¹⁴ Meta-analysis can therefore be performed on the results regarding QALY’s or direct costs, among others, provided that the necessary information is obtained from individual studies.

The purpose of the current paper is to provide an answer to the methodological question regarding feasibility of this approach. It aims at demonstrating the use of meta-analyses regarding the results of studies evaluating the treatment of melanoma in the economic terms and encourage scientists to provide variance estimators or credible intervals acquired within cost-effectiveness analyses or cost-utility analyses. Cost-utility analysis (CUA) and cost-effectiveness analysis (CEA) are both forms of full economic evaluation used in healthcare decision-making, but they differ in how outcomes are measured. CEA measures outcomes in natural clinical units (eg, life years gained, seizure reduction, cases detected, ICER), while CUA measures outcomes in quality-adjusted life years (QALYs) or similar utility-based measures.¹⁵ The most frequent approach currently is to perform probabilistic sensitivity analysis (PSA) regarding chosen indicator, usually QALY or ICER, based on random sampling, but taking the results of just a single study as a starting point. The level of uncertainty regarding the results acquired can be expected visually on a scatter plot. The final results of PSA can be provided as the average value of cost or benefits of treatment, and this is the most frequent approach. However, 95% credible intervals can also be provided. These intervals allow inspection of uncertainty regarding the results acquired numerically and allow for extracting a variance measure. The levels of a value of interest acquired in many studies or acquired in a single study, but concerning different treatments, if accompanied by a variance measure, can be combined into a single value with confidence intervals with the use of Bayesian meta-analysis.¹⁶ The differences between the results of different studies can be analysed further with the means of meta-regression. What is more, Bayesian meta-analysis does not require the analysed variables to be normally distributed.

What makes synthesizing economic evaluations difficult is heterogeneity. They differ regarding currency, price-year perspective, time horizon, discounting, study designs, healthcare perspectives, and outcome measures. On the condition that uncertainty estimators are provided or possible to calculate, these differences can be coded and analysed further with meta-regression in the course of meta-analysis. What is more, Bayesian meta-analysis can be performed starting with noninformative priors, which means assigning equal probability to every value and not incorporating any preexisting information if not available or starting with formally encoding pre-existing information from the previous research before analysing the next data, which improves the outcomes.

Materials and Methods

A literature search was conducted to acquire the estimates for QALY and direct costs of treatment for melanoma within MEDLINE, MEDLINE Ultimate and PubMed databases. The following query was applied: (melanoma) AND (treatment) AND (cost-effectiveness), with limiting the scope to papers published in the last ten years. An additional filter limiting the papers acquired to the ones focused on humans was also applied. The search was performed on the 6th of March 2026. Specifically, the papers were searched for the values of QALY’s and direct costs with confidence intervals. Figure 1 below depicts the PRISMA flow diagram and illustrates the study selection process.

Figure 1.

PRISMA flow diagram.

Out of several possible statistical models, like conventional meta-analysis or multilevel Monte Carlo Methods, Bayesian meta-analysis with non-informative priors was performed next. This approach¹⁷ is particularly relevant for synthesising results varying in quantity. It provides a pooled estimate of the effect being studied and a credible interval for it with a technique that modifies the effects acquired in individual studies or groups by shrinking them toward a common estimate or overall mean. With this technique, the reliability of the estimates is improved by reducing the influence of extreme cases or noisy data.¹⁸ Also, Bayesian meta-analysis with non-informative priors does not require normality. With this technique, a prediction interval estimating the results of future studies is calculated as well.

The calculations were performed with the use of bayesmeta package¹⁹ working in R Statistics 4.5.2 statistical software environment.²⁰ The bayesmeta package also allows for calculating meta-regression. Standard errors were calculated with the use of 95% confidence intervals provided by the authors of the papers under the assumption of normality by dividing half of the interval width by the appropriate critical value which for 95% confidence intervals is equal to 1.96. The values included in the meta-analyses are provided in Table S1.

Results

The number of records acquired as a result of the query was 187 papers from PubMed and 96 papers from MEDLINE and MEDLINE Ultimate combined. After removing two duplicates, the total number of 281 papers was examined, including Supplementary Materials. Only seven studies provided the confidence intervals, which makes calculation of standard errors possible.

The first study²¹ evaluated the cost-effectiveness of nine sequential treatment strategies applied in France for BRAF-mutated and wild-type patients diagnosed with melanoma. The drugs are analysed were immune-oncology therapies including ipilimumab (IPI, a CTLA-4 inhibitor), nivolumab, and including pembrolizumab (NIVO and PEM, respectively, two PD-1 inhibitors) and the combination of NIVO+IPI, targeted therapies approved for patients with BRAF mutation including monotherapy BRAF inhibitors (vemurafenib and dabrafenib, as MONO-TT), and combinations of bi-targeted therapies such as BRAF+MEK inhibitors (vemurafenib+cobimetinib, dabrafenib+trametinib, and encorafenib+binimetinib, BI-TT). A multistate model with three states, ie, first line of treatment, subsequent lines and death was used with the time perspective for 10 years. It was based on records from the cohort of 1,719 patients from 26 hospitals. The 95% confidence intervals for costs of treatments and QALYs were calculated with non-parametric bootstrapping with 1000 replicates. Depending on the sequence, point estimates for QALY ranged from 1.1 to 3.2 and for the cost of treatment point estimates ranged from €151,475 to €558,168. The costs included drugs, hospitalizations, radiotherapy sessions, biological and radiological exams. Figure 2 below depicts the forest plot summarizing the results of Bayesian meta-analysis regarding QALYs performed for the purpose of the current study.

Figure 2.

Forest plot for quality adjusted life years depending on the to treatment sequence.

Notes: All effects were acquired from the same study.²¹

The pooled estimate for QALY regarding treatment of melanoma was estimated to be equal to 1.71, with 95% credible interval between 1.31 and 2.19. The credible interval for the type of treatment, as a moderator, acquired in the course of meta-regression, was equal to.26[0.05;0.47], it did not include zero, which means that the differences between types of treatment were significant. Specifically, chemotherapy was less effective in terms of QALYs than the sequence Anti-PD1 → Bi-T 2 and the sequence Bi-TT 2 → Anti-PD1, since the intervals for these treatments differed. Also, it means that the sequences differ and the pooled estimate is not relevant for the clinical practice. This conclusion, however, was formulated for illustrative purposes only. Only when more studies with uncertainty measures are reported in the future will such conclusions be available for the purpose of informing clinical or policy decisions.

Figure 3 presents a forest plot summarizing the Bayesian meta-analysis of direct treatment costs conducted in this study, stratified by treatment sequence.

Figure 3.

Forest plot for the direct costs of treatment depending on the treatment sequence.

Notes: All effects were acquired from the same study.²¹

The pooled estimate for direct costs of treatment for melanoma was estimated to be equal to 214,251 EUR, with 95% credible interval between 185,453 and 242,960 EUR. The credible interval for the type of treatment as a moderator, acquired in the course of meta-regression, was equal to 44,344.13[3355.71; 86,069.53], it did not include zero, which means that the differences between types of treatment were also significant. Specifically, chemotherapy, Ipilimumab or Anti-PD1 involved lower costs than the sequence Anti-PD1 → Bi-T 2, the sequence Bi-TT 2 → Anti-PD1, and the sequence IPI+NIVO → Bi-TT 2 since the intervals for these treatments differed. Again, this conclusion was formulated for illustrative purposes only and not to inform clinical or policy decisions, unless more studies with uncertainty measures are available in the future.

The next study²² used a Markov decision model to evaluate the cost-effectiveness of treating advanced melanoma with tumour-infiltrating lymphocytes when compared to Ipilimumab in the current Dutch and Danish setting. The term advanced melanoma in this case applied to patients with unresectable stage IIIC–IV melanoma after failure of first-line or second-line treatment. Three states were included in the analysis, ie, progression-free survival, progressive disease, and death. The health‑state transition probabilities were based on the data from the TIL-NKI/CCIT clinical trial.²³ The study was conducted to evaluate the cost-effectiveness of the treatment in the Danish setting using Ipilimumab as a comparator over a lifetime horizon. QALY for TIL-NKI/CCIT was equal to 3.52 with credibility interval from 3.30 to 4.59 and to 2.46 with a credibility interval from 1.47 to 3.41 for Ipilimumab. Total lifetime undiscounted societal costs were 347,168 EUR with a credibility interval from 269,889 to 447,100 EUR and to 433,634 EUR with credibility interval from 329,255 to 571,253 EUR for TIL-NKI/CCIT and for ipilimumab, respectively. The final conclusion was that the treatment with TIL-NKI/CCIT was cost-effective.

In another study,²⁴ clinical trial data from patients with advanced melanoma treated in the USA with either ipilimumab or glycoprotein 100 were analysed. QALY in the case of the treatment with glycoprotein 100 was equal to 1.67 95% CI[1.31, 3.28], with the mean cost equal to 89,000 USD 95% CI[66,000, 123,000]. QALY for the treatment with glycoprotein 100 was equal to 4.72 95% CI[3.10, 6.37], with the mean cost equal to 238,000 USD 95% CI[215,000, 272,000]. Therefore, ipilimumab was more beneficial, but at a higher cost.

The next study²⁵ compared cost-effectiveness regarding treatment for patients with advanced melanoma in Hong Kong based on dacarbazine with the treatment based on pembrolizumab. The treatment based on dacarbazine, which was the standard of care, was associated with QALY equal to 1.65 95% CI[0.86, 2.59] and the cost 47,124 USD 95% CI[35,410, 62,214]. The treatment based on pembrolizumab was associated with QALY equal to 4.27 95% CI[4.06, 4.48] and the cost 140,612 USD 95% CI[127,407, 156,690], so it was also more beneficial, but at a higher cost than the standard of care.

The early cost-effectiveness of electrochemotherapy in patients with Stage IIIc/IV skin melanoma in clinical practice in Slovenia was analysed in the next study.²⁶ The treatment was associated with QALY equal to 0.66 95% CI[0.54, 0.74] for the patients with predominantly cutaneous symptoms, to 0.72 95% CI[0.66, 0.80] for the patients experiencing a partial or complete response to electrochemotherapy and to 0.66 95% CI[0.54, 0.74] for the patients in a relapse state. The cost was equal to 4,333 EUR 95% CI[4,139, 4,533].

The results of these studies cannot be combined with the previous one²¹ due to different time horizons and different types of costs analysed. However, if more studies are published with credible intervals for QALY and/or costs analysed, the estimation of standard errors will be possible, and further analyses with meta-analysis and meta-regression will be feasible.

Ipilimumab itself was evaluated in a population-based cohort study in Canada,²⁷ and it was not found to be cost-effective as a second-line monotherapy for patients with metastatic melanoma under conventional willingness-to-pay thresholds when compared to non-ipilimumab therapies, even though it seemed to provide substantial clinical benefit in terms of the prognosis for patients. The results of 329 patients were included in the analysis, 189 patients treated with Ipilimumab and 140 controls. Incremental difference regarding QALY was equal to 0.40 with 95% confidence interval from 0.17 to 0.64. This study also provides the 95% confidence interval for ICER. The incremental cost-effectiveness ratio was equal to 225,885 CAD (Canadian dollars) with the 95% confidence interval from 149,876 to 477,082 CAD. If more studies published confidence intervals or explicit uncertainty measures for the indicators of cost-effectiveness, further analysis involving the use of meta-analysis and meta-regression would be possible.

Another study evaluated the cost of treatment for melanoma over a 5-year period when taking into account stage of the disease, changes in practice, the percentage of doses consumed and the percentage of locoregional and distant recurrences.²⁸ The study concerned the perspective of the healthcare system in Ireland. The model of cumulative cost involved nine initial stages at diagnosis of the disease, ie from IA (localized skin cancer, thin, ≤ 1 mm thick, without metastasis spread to lymph nodes) to IV (the cancer, which has spread from the skin to distant parts of the body like the lungs, liver, brain, bones, or distant lymph nodes). The time horizon was set at five years. The stage-weighted best-case estimate of costs of managing melanoma was equal to 20,268 EUR, with confidence interval from 32,619 to 43,019 EUR. The direct costs taken into account were the costs of the diagnosis, medical procedures, both in-patient and out-patient, systemic therapy, relapse and supportive care. The impact of uncertainty of parameters on the total cost estimates was evaluated with the use of one-way and probabilistic sensitivity analyses (PSA). These analyses were performed by running a model with a hypothetical cohort of 10000 patients. The cost estimates were calculated for each simulation. These analyses allowed for calculating 95% confidence intervals for the average cost per person.

Figure 4 below depicts the forest plot summarizing the results of Bayesian meta-analysis regarding the direct costs of treatment performed for the purpose of the current study on the basis of the estimates provided for subsequent stages by the authors of the study mentioned above.

Figure 4.

Forest plot for the direct costs of treatment depending on the stage of melanoma.

Notes: All effects were acquired from the same study.²⁸

The point estimates showed that the average cost of managing a case of melanoma diagnosed at stage IV was more than twenty-five times more expensive than managing a case diagnosed at stage IA (122,985 EUR vs 4,269 EUR). The pooled estimate for direct costs of managing melanoma in the period of five years was estimated to be equal to 6,232 EUR with 95% credible interval between 5,711 and 6,753 EUR. The credible interval for the stage of the disease as a moderator, acquired in meta-regression, was equal to 11,210.80 [7476.98; 15,130.67]; it did not include zero, which means that the differences between the costs of managing different stages of melanoma were significant. Specifically, the costs of the treatment were much higher if the disease was diagnosed at stage IIIA or later since the intervals for these stages differed. Also, the intervals regarding the treatment of patients diagnosed at more advanced stages were much wider, which suggests much higher variability in the costs of the treatment involved. Again, this conclusion was formulated for illustrative purposes only.

Discussion

The current paper demonstrates the advantages of combining cost-effectiveness, Bayesian meta-analysis and meta-regression techniques. These include more reliable and stable estimates acquired and more precise conclusions regarding the differences between the estimates based on the results from individual studies or samples representing various groups of patients. The necessary condition for applying meta-analytic techniques is that the estimates are published with uncertainty measures or confidence/credibility intervals. Currently, very few papers include these additional estimates. Hopefully, it will change in the future.

The current analyses showed that chemotherapy was less effective in terms of QALYs than the sequence Anti-PD1 → Bi-T 2 and the sequence Bi-TT 2 → Anti-PD1 when treating BRAF-mutated and wild-type patients diagnosed with melanoma. Chemotherapy, Ipilimumab or Anti-PD1 were found to involve lower costs than the sequence Anti-PD1 → Bi-T 2, the sequence Bi-TT 2 → Anti-PD1, and the sequence IPI+NIVO → Bi-TT 2. The costs of the treatment were higher if melanoma was diagnosed at stage IIIA or at a more advanced stage and the treatment of patients diagnosed at more advanced stages was associated with much higher variability regarding the costs of the treatment involved. These results were based on the results provided by two separate single studies. However, combining the results of various studies taking the dependencies between the results from a single study into account is possible with the means of hierarchical Bayesian meta-analysis.²⁹

Some limitations of the approach need to be mentioned as well. Very few studies report variance estimators. The standard errors in the current paper were calculated from the confidence intervals under the assumption of normality, which is rarely the case for measures like QALY. Bayesian meta-analysis with non-informative priors does not require normality. However, extracting standard error from 95% confidence intervals does. ICER, another cost-effectiveness measure, is a ratio. Calculating standard error for a ratio is usually complex, since ratios are unstable as estimates, especially when values are close to zero. Estimating variance of ICER needs special methods like the calculations based on Fieller’s theorem³⁰ or, preferably, bootstrap methods,³¹ including Bias-Corrected and Accelerated bootstrap sampling.³² Gradually, with more and more advanced software, these methods hopefully will become more available. Another approach that can be used is Comparative Efficiency Research³³ based on the incremental net benefit (INB) parameter. The positive and negative INBs directly distinguish cost-effectiveness from non-cost-effectiveness of interventions. Specific recommendations for this approach include methods of data extraction, harmonization and pooling. INB as a measure of cost-effectiveness was used, for example, in a meta-analysis regarding cost-effectiveness studies for pneumococcal vaccination in children.³⁴ It allowed for determining the cost-effectiveness of pneumococcal conjugate vaccines, 13-valent and 10-valent, in both high-income countries and low- and middle-income countries. However, the INB approach introduces another source of variability, ie, the comparator, the treatment being used as baseline comparison, which is different in different studies. We certainly hope our demonstration will encourage scientists to provide variance estimators for the meta-analyses to be robust and representative in the future. The combined methodological approach enhances the robustness and comparability of estimates. The observed variability in both clinical effectiveness and costs across treatment strategies and disease stages has important implications for healthcare decision-making and resource allocation.

Funding Statement

This research received no external funding.

Data Sharing Statement

The data are available from the corresponding author upon request.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare no conflicts of interest.