Overall Survival of Patients with Melanoma of Unknown Primary Versus Melanoma of Known Primary Under Immunotherapy and Targeted Therapy: A Systematic Review and Meta-Analysis

Abstract

Background: Melanoma of unknown primary (MUP) is a rare and distinct clinical subtype of metastatic melanoma, in which no identifiable primary tumor is found. The prognosis of MUP compared to melanoma with known primary (MKP) remains unclear, especially in the era of novel therapies like immune checkpoint inhibitors (ICIs) and targeted therapies. This meta-analysis aims to compare the overall survival (OS) of MUP and MKP patients under these therapies. Methods: This systematic review was conducted in line with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA). A systematic search of major databases was conducted, yielding six eligible studies (nine study arms) that assessed the survival outcomes of MUP and MKP patients treated with immunotherapies and targeted therapies. We pooled the hazard ratios (HRs) for OS using both fixed and random effects models. Heterogeneity was assessed with the I² statistic followed by a Baujat plot, and publication bias was evaluated using funnel plots and Egger’s test. Results: Our analysis revealed a borderline significant HR of 0.90 (95% CI: [0.81, 1.00], p = 0.04) under the fixed effect model, suggesting a potential survival benefit for MUP patients. However, the random effects model, accounting for study heterogeneity, showed no significant difference in OS between MUP and MKP (HR = 0.87, 95% CI: [0.73, 1.05], p = 0.15). Significant heterogeneity (I² = 66.9%, p = 0.0022) was observed across studies. No substantial publication bias was detected. Conclusion: While the trend observed in the fixed effect model suggests a potential benefit for MUP patients, the random effects analysis indicates no significant difference between MUP and MKP in terms of OS. These findings underscore the importance of accounting for study heterogeneity and highlight the need for further prospective studies to better understand the impact of novel therapies on MUP.

1. Introduction

Melanoma of unknown primary (MUP) is a distinct clinical presentation of metastatic melanoma in which no identifiable primary tumor is found. MUP is estimated to account for approximately 3–4% of all melanoma diagnoses. The pathogenesis of MUP remains unclear, with theories suggesting spontaneous regression of the primary lesion or primary dermal or visceral melanomas without a recognizable cutaneous source. Compared to melanoma with a known primary (MKP), MUP often presents at a more advanced stage, involving lymph nodes, subcutaneous tissue, or visceral organs. Treatment strategies for MUP have largely mirrored those for MKP. Surgical resection has been the cornerstone of treatment in patients with resectable disease, while radiation therapy has been employed in select cases for local control. Traditional systemic therapies such as cytotoxic chemotherapy have previously shown limited efficacy in both MUP and MKP, yielding modest survival benefits. Recent advances in immunotherapy and targeted treatments have significantly improved outcomes in metastatic melanoma, including MUP. Immune checkpoint inhibitors (ICIs), such as anti-PD-1 agents (e.g., nivolumab, pembrolizumab) and CTLA-4 inhibitors (e.g., ipilimumab), have revolutionized the treatment landscape by enhancing anti-tumor immunity. Additionally, targeted therapies directed at key oncogenic mutations, particularly BRAF and MEK inhibitors, have demonstrated substantial efficacy in patients harboring these genetic alterations [1,2,3,4]. Previously, outcomes for MUP have often been regarded as on par with—or even superior to—those for melanoma with known primary (MKP) when patients are matched by disease stage and metastatic extent. This counterintuitive finding may arise from intrinsic differences in tumor biology, variations in immune surveillance, or biases in how tumors are detected. As a result, the relative prognosis of MUP versus MKP remains uncertain and likely varies with the specific clinical context. Before the introduction of immune checkpoint inhibitors, reported 5-year overall survival (OS) rates for MUP ranged from 8% to 18%, most frequently involving metastases to lymph nodes and the gastrointestinal tract [4,5,6,7].

MUP presents a unique clinical challenge: patients manifest regional or distant metastatic disease (AJCC 8th edition stage III or IV) without an identifiable cutaneous, mucosal, or ocular primary lesion. In AJCC terms, stage III encompasses metastases to regional lymph nodes, satellite, or in-transit sites, whereas stage IV denotes spread to distant skin, subcutaneous tissue, distant lymph nodes, or visceral organs [8]. The absence of a detectable primary tumor complicates both prognostication and treatment selection, since standard staging and risk stratification pivot on primary-tumor characteristics (e.g., Breslow thickness, ulceration). Melanoma of unknown primary (MUP) likely arises through multiple, overlapping mechanisms: one hypothesis holds that a nascent cutaneous melanoma undergoes complete, immune-mediated regression—leaving only dermal fibrosis, melanophages, and lymphocytic infiltrates as histologic footprints [9], while another proposes that ectopic nevus cell nests within lymph nodes may transform de novo into malignant lesions [10]. Additionally, small or atypically located primary tumors in mucosal or ocular sites can evade detection during routine examinations [11], and highly aggressive clones may disseminate early, seeding distant sites before the primary lesion becomes clinically apparent. These models collectively suggest that MUP represents a heterogeneous “syndrome” in which robust host immunity and perhaps unique tumor biology play a central role, providing a strong rationale for the application of ICIs in this subgroup. However, despite the success of immunotherapies and targeted agents in stage III and IV melanoma, data specifically addressing their efficacy in MUP patients remain scarce and underscore the need for dedicated clinical studies [12,13,14,15,16,17].

In this systematic review and meta-analysis, we evaluate whether novel treatments confer improved outcomes for MUP patients compared to those with MKP.

2. Materials and Methods

2.1. Protocol

This review and meta-analysis were carried out in accordance with the PRISMA statement (see Supplementary Table S1) and are registered on the Open Science Framework (DOI: 10.17605/OSF.IO/FTKVD; accessed on 9 March 2025) [18].

2.2. Search

We performed a comprehensive, language-unrestricted search of PubMed, Web of Science, and Scopus for studies published from 2011 onwards (the start of the novel melanoma therapy era). The full search strings are detailed in Supplementary Table S2.

2.3. Selection of Studies

Following identification of potential studies, two independent reviewers (T.G., P.G.) selected the included articles in three phases. In screening phase 1, the reviewers evaluated the titles and abstracts based on the eligibility criteria. In phase 2, the reviewers reviewed the full texts and selected the articles according to the same criteria as in screening phase 1. Subsequently, the reviewers verified all data extracted concerning critical inclusion and exclusion criteria. In cases of disagreement, another reviewer (L.S. or N.A.) was consulted. Inclusion criteria were as follows: Studies including at least 30 patients with stage III/IV MUP as well as MKP cohorts, novel interventions (e.g., immunotherapies and targeted therapy), hazard ratios (HRs) including the 95% confidence intervals (CI) for OS. Abstracts, posters, reviews, and case reports were excluded from further analysis.

2.4. Data Extraction

We systematically extracted the following information from each study: author(s), study design, year and country of publication, type of intervention, key findings, and authors’ conclusions. Our primary endpoint of interest was either overall survival or melanoma-specific survival.

2.5. Assessment of Study Quality and Risk of Bias

Each article’s methodological rigor and potential for bias were evaluated using the Critical Appraisal Skills Programme (CASP) checklist [19]. We selected this tool for its clear, structured framework, which accommodates a range of study designs and facilitates consistent appraisal of internal validity and applicability.

2.6. Data Synthesis and Statistical Analysis

A meta-analysis was conducted to evaluate the differences in OS between patients with MUP and MKP. Data, including HRs with corresponding 95% CI, number of patients per cohort, and treatments were extracted from each included study. All statistical analyses were performed using R and RStudio (version 4.4.3). A forest plot was generated to visually represent the individual study estimates and the pooled fixed and random effect sizes for OS. The presence of statistical heterogeneity among studies was assessed using the I² statistic (Cochran’s Q test). I² of 30% or less was considered to be a low degree of heterogeneity, 30% to 60% to be a moderate degree, and 60% or more to be a high degree. Contributors to heterogeneity were identified using a Baujat plot. The publication bias was assessed using the funnel plot including Egger’s test.

3. Results

3.1. Study Selection

Our initial search of the three databases on 8 March 2025 (updated 14 March 2025) yielded 289 records. Following duplicate removal and other exclusions, 68 articles advanced to the first screening stage (title and abstract review). Twenty of these proceeded to full-text evaluation in the second screening phase. In the final eligibility assessment, we applied our predefined inclusion and exclusion criteria, resulting in six studies (including relevant sub-analyses) being retained for the review (Figure 1).

3.2. Study Characteristics

The final analysis included six studies (with sub-studies) that met the inclusion criteria. These studies were primarily focused on comparing MUP with MKP under novel therapies, assessing OS outcomes. The included studies were a mixture of retrospective cohort studies and one randomized controlled trial (RCT), which investigated stage III and IV patients in an adjuvant setting after complete surgery [16]. A total of nine study arms were included, representing data from both MUP and MKP patient populations which have been detailed in

Table 1. Selected studies for the evaluation of survival of patients with melanoma of unknown primary (MUP) versus patients with melanoma with known primary (MKP). Outcomes after novel therapies beyond 2011.

#	Study (Country) Stage	Treatment	HR OS §/MSS $	95% CI (Lower-Upper)	MUP n	MKP n
1	Ellebæk 2019 [12] § (Denmark) unresectable IIIC/IV (M1a or M1b)	* ICI	1.03	0.78–1.37	80	496
2	Gambichler 2019 [13] $ (Germany) unresectable IIIC/IV (M1a or M1b)	* ICI	0.29	0.11–0.75	9	32
3	Verver 2021 [14] § (Netherlands) unresectable IIIC/IV (M1a or M1b)	* ICI or TT	0.74	0.61–0.90	385	2321
4	Tarhini 2022 [15] § (USA) resected IIIB, IIIC, IV (M1a or M1b)	* CTLA-4i or hd-INF	0.61	0.42–0.86	214	1669
5 A B	Rousset 2023 [16] § (France) unresectable IIIC/IV (M1a or M1b)	* ICI * TT	1.20 0.90	1.00–1.50 0.60–1.10	165 86	1059 488
6 A B C	Persa 2024 [17] § (Germany) unresectable IIIC/IV (M1a or M1b)	* PD-1i * PD-1i + CTLA-4i * BRAFi + MEKi	0.79 1.22 0.93	0.50–1.26 0.74–1.99 0.58–1.49	142 152 101	142 152 101

* first treatment regimen; ICI = immune checkpoint inhibitor (PD-1i + CTLA-4i, PD-1i, or CTLA-4i); TT = targeted therapy; OS = overall survival; MSS = melanoma-specific survival; hd-INF = high-dose interferon. § represents overall survival; $ represents melanoma-specific survival.

. The studies varied in terms of sample sizes, treatment regimens, and outcome measures (e.g., overall survival and melanoma-specific survival). The patient population in these studies was predominantly adult, with a wide range of sample sizes (from as low as 9 to as high as 2321). The therapies under study were almost exclusively novel (post-2011 era), including a mix of immunotherapy and targeted therapy regimens, whereas first-line immunotherapy modalities dominated. A summary of study characteristics of the selected papers is shown in Table 1.

3.3. Individual Assessment of the Risk of Bias and Study Quality

In our quality appraisal and assessment of risk of bias using CASP analysis (Table S3), all five cohort studies and the lone randomized trial demonstrated a very low risk of bias overall, with only minor, study-specific limitations. Among the cohorts, Gambichler et al. [13], Verver et al. [14], and Persa et al. [17] each met every CASP criterion, reflecting exceptionally rigorous design, thorough confounder control, valid exposure and outcome measurement, and near-complete follow-up. Ellebæk et al. [12] also performed strongly across most domains but did not explicitly account for one key potential confounder, suggesting that residual confounding cannot be fully excluded. Likewise, Rousset et al. [16] fulfilled almost all CASP items but experienced modest attrition, raising a slight concern about the completeness of follow-up.

The randomized trial by Tarhini et al. [15] achieved 10 of 11 “yes” ratings, indicating robust randomization, allocation concealment, balanced baseline characteristics, and appropriate intention-to-treat analysis. Its sole shortcoming was the absence of participant and assessor blinding, which introduces a modest risk of performance or detection bias, particularly for subjective outcomes.

Taken together, these studies exhibit high methodological quality (≥90% of CASP criteria met), lending strong credibility to their findings. Nevertheless, when synthesizing results, we give particular attention to residual confounding in Ellebaek et al. [12], attrition in Rousset et al. [16], and the unblinded design of Tarhini et al. [15], especially for patient-reported endpoints. By accounting for these nuanced limitations, we ensure a balanced and transparent interpretation of the evidence.

3.4. Hazard Ratios and Pooled Analysis

We performed a meta-analysis to compare the HRs for OS in patients with MUP versus MKP under novel therapies. The HR estimates for each of the nine study arms are also summarized in Figure 2.

The overall pooled HR for the fixed effect model was 0.90 (95% CI: [0.81, 1.00], p = 0.04), indicating a statistically significant treatment effect for novel therapies in improving OS for patients with MUP versus MKP, assuming homogeneity across the studies. However, the random effects model, which accounts for heterogeneity, yielded a pooled HR of 0.87 (95% CI: [0.73, 1.05], p = 0.15), which was not statistically significant.

3.5. Heterogeneity and Baujat Analysis

Heterogeneity across studies was evaluated using the I² statistic, which indicates the proportion of observed variability attributable to true differences between studies rather than random sampling error. The I² value for the forest plot was 66.9%, indicating moderate-to-substantial heterogeneity in the studies.

When excluding our previous study [13], the I² value dropped only to 60.13%, indicating that the exclusion of this study did not significantly alter the overall heterogeneity. The Q-statistic for residual heterogeneity was 24.16 (p = 0.0022), confirming significant variability across the studies included in the meta-analysis.

To investigate sources of heterogeneity, a Baujat plot was generated (Figure 3). The Baujat plot visually displays the contribution of each investigation to the overall heterogeneity as well as the effect estimate. On the x-axis, the plot shows how much each study influences the pooled effect estimate, while on the y-axis, it shows each study’s contribution to the total heterogeneity. In the Baujat plot, studies that are located far to the right on the x-axis have a strong influence on the pooled effect estimate, while studies far up on the y-axis have a high contribution to heterogeneity. Ideally, a symmetrical distribution of studies suggests that no single study is disproportionately influencing the results. From the Baujat plot, it was evident that our previous study [13] was a significant outlier. This study had a very low HR (0.29), and its contribution to the overall heterogeneity was much higher compared to other studies. Indeed, our previous study [13] had a much smaller sample size compared to other studies, and this could have exaggerated its influence on the overall pooled estimate. However, after reviewing the plot and conducting a sensitivity analysis, we decided not to exclude this study from the analysis. While the study was an outlier, excluding it would have resulted in a reduction of heterogeneity from I² = 66.9% to I² = 60.13—a very low reduction. Moreover, excluding our previous study [13] did not significantly alter the overall pooled HR, and excluding studies with lower sample sizes or outliers could introduce a bias toward more uniform results.

3.6. Publication Bias

We evaluated publication bias by constructing a funnel plot (Figure 4), which appeared fairly symmetric around the pooled effect size, suggesting minimal bias. Egger’s regression test for asymmetry produced a p-value of 0.15, indicating no statistically significant bias at the 0.05 threshold. Nonetheless, given the limited number of studies analyzed, the possibility of undetected publication bias cannot be entirely excluded.

3.7. Meta-Regression and Moderator Analysis

Meta-regression was performed to explore whether any study characteristics, such as sample size, might explain the observed heterogeneity. The moderator analysis for sample size yielded a non-significant p-value of 0.6154, suggesting that sample size did not significantly explain the variability across studies. The I² value after accounting for sample size decreased slightly to 60.13%, confirming that the moderator did not substantially reduce heterogeneity.

3.8. Sensitivity Analysis

To assess the influence of individual studies on the overall effect, a sensitivity analysis was conducted by excluding our previous study [11] and the study of Rousset et al. [16]. Removing our previous study [13], which had a very low HR value, resulted in a small reduction in heterogeneity (from I² = 66.9% to I² = 60.13%), but it did not significantly change the overall treatment effect in either the fixed or random effects model. Excluding Rousset et al. [16] had a similar effect, with no substantial change in the overall HR or heterogeneity.

4. Discussion

In a previous review, Bae et al. [5] selected studies up to the end of 2012 and performed a meta-analysis of 18 studies including 2084 patients with MUP and 5894 with MKP. Patients with MUP had a better OS compared with MKP in stage III (HR 0.83, 95% CI 0.73–0.96, p = 0.010) and stage IV (HR 0.85, 95% CI 0.75–0.96, p = 0.008). In a subgroup analysis, advanced disease stage and older age emerged as significant risk factors among MUP patients. Bae et al. [5] also found that, when stratified by the same tumor stage, individuals with MUP demonstrated superior survival compared to those with MKP. Notably, the studies included in the study of Bae et al. [5] almost exclusively reported patients treated with chemotherapy (e.g., dacarbacine) and/or interferon (INF). In 2020, Verver et al. [6] reported that the introduction of novel therapies (ICIs, targeted therapies) have significantly improved OS of patients with MUP, indicating that these new treatments also work in this uncommon melanoma sub-type [6].

The present analysis aimed to compare the OS outcomes of patients with MUP and MKP under novel therapies, particularly focusing on ICIs and targeted regimens. The results suggest that, while patients with MUP exhibit a comparable prognosis to those with MKP when treated with contemporary therapies, some important nuances were observed. The pooled data from nine study arms indicated that patients with MUP exhibit comparable OS when treated with these therapies. This is consistent with previous studies suggesting that MUP patients tend to have a relatively favorable prognosis, particularly when matched for stage and metastatic burden. However, despite the improvement in outcomes with modern therapies, particularly including immune-based regimens, no clear survival advantage was observed between the two groups, even though MUP patients are usually diagnosed at more advanced stages, involving lymph nodes and visceral organs [1,2,3,4]. However, the lack of large-scale data specifically for MUP patients limits the definitive understanding of their effectiveness in this population.

Our analysis revealed significant heterogeneity in treatment outcomes, as indicated by an I² value of 66.9%. This variability could be attributed to differences in study design, patient characteristics, and the diverse treatment regimens used across studies. Moreover, since only one study in the analysis, by Tarhini et al., was an RCT [15], the majority of the data were derived from retrospective studies, which may introduce biases due to differences in patient selection, treatment protocols, and outcome reporting. Notably, many observational cohort studies could not be included into the analysis since there was no standard data reported, such as HR and 95% CI.

While the funnel plot and Egger’s test for publication bias did not suggest significant asymmetry, the limited sample size and predominance of retrospective data remain limitations. The underrepresentation of small or negative studies could result in subtle publication bias, which may have influenced the results. Future studies should ensure comprehensive data collection and avoid the potential biases inherent in retrospective studies. Additionally, meta-regression did not reveal any significant moderators explaining the heterogeneity, with sample size not significantly impacting the variation in treatment outcomes. This suggests that other factors, such as treatment regimens and patient demographics, may play a more critical role in explaining the variability in survival outcomes. The lack of significant moderators further highlights the need for more detailed investigations into the clinical and biological characteristics that may impact the treatment response in MUP.

The present meta-analysis also compared the results from fixed effect and random effects models. While the fixed effect model indicated a statistically significant survival benefit for MUP patients, this result was heavily influenced by the assumption of homogeneity across studies, which might not be a realistic reflection of the variability in treatment response. In contrast, the random effects model, which accounts for heterogeneity across studies, did not reveal a statistically significant difference in OS between the two groups (p = 0.15). The random effects model is more conservative and suggests that, when considering the substantial variability between studies, there is no clear survival advantage for MUP patients. This discrepancy highlights the importance of using appropriate statistical models in meta-analysis and suggests that the fixed effect model may have overstated the treatment benefit for MUP.

Study limitations were as follows: Our findings rely mainly on retrospective cohorts (with just one RCT), which may introduce selection and reporting biases, and many potential studies lacked, for example, HR/CI reporting for inclusion. High heterogeneity in design, patient characteristics, and treatments (I² = 66.9%) limits the precision of pooled estimates, while potential underreporting of null results could mask publication bias. Meta-regression did not identify clear sources of variability, highlighting gaps in clinical and biological data. Hence, the unknown sources of variability among the studies included weaken the robustness of this meta-analysis. Prospective, uniformly reported trials are needed to validate these results and clarify factors influencing outcomes in MUP vs. MKP.

5. Conclusions

While the borderline significant HR observed in the fixed effect model suggests a potential benefit of novel therapies for MUP, the random effects analysis indicates that there is no significant difference between MUP and MKP patients in terms of OS. The findings underscore the importance of using rigorous statistical models, such as the random effects model, to account for heterogeneity. Future research should focus on well-designed, prospective randomized controlled trials reporting standard measures, including HR and 95% CI, to better understand the clinical outcomes for MUP patients and to identify specific factors influencing treatment response. By addressing these gaps in the literature, we can optimize treatment strategies for both MUP and MKP patients and improve survival outcomes across both groups.