Advancing biomarker-based prognostication in melanoma

Abstract

Biomarker-driven risk stratification in cutaneous melanoma remains an area of intensive research, with the ultimate goal of refining prognostic accuracy and optimizing patient management. By identifying patients at higher risk for recurrence or melanoma-specific mortality, biomarkers have the potential to enhance patient counselling, tailor follow-up intensity and guide shared decision-making regarding adjuvant therapy.

In this issue, Dixon et al.¹ propose the BAUSSS algorithm (Breslow thickness, Age, Ulceration, Subtype, Sex and Site) as a novel tool to estimate 15-year melanoma-specific mortality percentage (MSMP) in patients with primary melanoma. Their approach integrates a nomogram-derived table including Breslow thickness, age and ulceration (from the Lifemath online tool) and a second table incorporating the independent relative risks of melanoma subtype, sex and anatomical site (extracted from the El Sharouni study). The final MSMP estimate is derived by multiplying the respective values from both tables. However, several methodological considerations warrant attention. The El Sharouni study, which serves as a key data source for BAUSSS, was conducted in 5644 patients from the Melanoma Institute Australia (MIA), all of whom underwent sentinel lymph node biopsy (SLNB). That study's prognostic model included multiple variables—Breslow thickness, age, ulceration, subtype, sex, site, mitotic rate and regression—and was derived from a relatively short median follow-up period (3.9 years). The El Sharouni study analyzed melanoma-specific survival (MSS) only in the MIA cohort, and thus external validation was not possible, limiting the model's generalizability.² Likewise, the development and validation of the new combined BAUSSS method to predict melanoma-specific mortality at 15 years was not described .¹

While prognostic models offer valuable insights, their clinical applicability should be interpreted with caution. External validation is crucial to ensure reproducibility across diverse patient populations, and model performance should be assessed using calibration plots and discrimination metrics (such as the concordance index or c-index). Current melanoma survival calculators vary in their predictions due to differences in included prognostic factors (e.g. categorical vs. continuous variables), statistical modelling approaches, patient cohorts used for model development. The AJCC online prediction tool, which was developed and validated for MSS for localized melanoma in US, Australian and European populations, focuses on thickness, ulceration, lesion site and age as key prognostic factors.³ The EORTC nomogram, designed for recurrence and melanoma-specific mortality in SLNB-negative patients, incorporates Breslow thickness, ulceration and primary site. Notably, external validation showed that adding sex, age, melanoma subtype and mitotic rate did not significantly improve its predictive power.⁴ The EORTC-DeCOG nomogram, tailored for SLNB-positive patients, integrates tumour thickness, ulceration, age and sentinel node tumour burden, underscoring that different risk factors may be relevant for localized versus metastatic patients.⁵ Additionally, histologic subtypes such as nodular melanoma (NM) are known to influence prognosis. NM has been incorporated into validated nomograms for predicting sentinel node positivity and recurrence risk, also in patients with thin melanomas.⁶

Beyond prognostic indicators, the biomarker landscape in melanoma is evolving toward predictors of treatment response, particularly in the era of immunotherapy. Adjuvant immunotherapy is now a standard of care for patients with high-risk resected stage III-IV melanoma, as well as for stage IIB/IIC patients. Furthermore, neoadjuvant immunotherapy is emerging as a promising strategy for macroscopic resectable stage IIIB+ disease, reshaping melanoma treatment paradigms. Current research initiatives focus on integrating genetic, transcriptomic and clinicopathological data to enhance risk prediction. For instance, gene expression profiling (GEP) has been employed in the NivoMela trial (NCT04309409) to prospectively stratify stage II melanoma patients for adjuvant nivolumab therapy. The journey toward personalized melanoma risk prediction is advancing rapidly, yet much remains to be explored. As we refine our prognostic models and integrate emerging biomarkers, the goal is to enhance individualized treatment strategies, ultimately improving patient outcomes.

None.

Clio Dessinioti: None reported. Alexander J Stratigos: served on Advisory Boards of Regeneron and Novartis; received honoraria from LeoPharma, Novartis and MSD; and received research support from Roche, Genesis Pharma, Janssen Cilag and Abbvie, unrelated to the content of this work.