The State of the Field This 2023 review from Temple and Jefferson Universities summarizes the state of melanoma epidemiology, classification, diagnosis, and prognosis. While melanoma incidence has risen dramatically over decades, the treatment of advanced disease has entered what the authors call a golden era, driven by targeted therapies and immunotherapies that have dramatically improved survival for stage III and IV disease.
The Coming Paradigm Shift The traditional clinicopathologic classification of melanoma - based on morphology and histologic interpretation - is being replaced by a more precise molecular-based one. The 2018 WHO classification expanded melanoma subtypes from 4 to 9, incorporating genomic driver mutations. Molecular ancillary tests including gene expression profiling (GEP), comparative genomic hybridization (CGH), and FISH are becoming standard tools in pathology.
AI and Molecular Diagnostics Artificial intelligence augmented clinical and histopathologic diagnosis of melanoma is expected to make the diagnostic process more streamlined, precise, and efficient. As databases of clinical, dermatoscopic, and histologic images grow, validated generalizable AI training sets are expected to emerge. However, significant barriers remain including dataset bias toward lighter skin types, lack of regulatory approval, and unresolved interobserver reliability issues in pathology.
Global and US Burden In 2020, the global melanoma burden reached 325,000 cases - a 41% increase from 230,000 in 2012. In the US, melanoma is the fifth most common cancer, with 99,780 new invasive cases and 97,920 in-situ cases expected in 2022. Despite its deadly reputation, the overall 5-year survival rate is 93.5% because 78% of diagnoses are localized (99.6% 5-year survival). Stage III and IV have 5-year survival rates of 73.9% and 35.1% respectively - dramatically improved from just 15% for Stage IV in 2015 due to new therapies.
The Overdiagnosis Problem A threefold increase in melanoma incidence over 40 years has been accompanied by stable mortality rates - a pattern consistent with overdiagnosis of clinically insignificant lesions. Melanoma in-situ diagnoses rose from 28,600 in 2000 to 101,280 in 2021 in the US without a concomitant decrease in late-stage disease. The epidemiologic evidence indicates that most early melanomas represent indolent or biologically benign forms, not obligate precursors to deadly melanoma.
Screening and the Dysplastic Nevus The USPSTF has consistently given an 'I' (insufficient evidence) grade for skin cancer screening, as no randomized trials have shown population-level benefit. Population screening in Germany showed no impact on melanoma-specific mortality. Over 40 years of biopsying and excising dysplastic nevi - once promoted as melanoma precursors by NIH consensus - has failed to reduce mortality, and many authors now conclude the dysplastic nevus is a phenotype marker rather than an obligate precursor.
Three Epidemiologic Forms Current evidence distinguishes three heterogeneous melanoma forms: slow-growing melanomas associated with intermittent sun exposure and nevi; slow-growing indolent melanomas associated with chronic sun exposure on the head and neck; and fast-growing aggressive melanomas minimally associated with sun exposure. This third type - including nodular melanoma - is not amenable to screening because it grows too rapidly and lacks predictable early features.
Phenotype and Genetic Risk Caucasians have 20 times the melanoma risk of Black people. High nevus count, giant congenital nevi, fair skin, and immunosuppression are risk factors. Key germline mutations include CDKN2A (most common hereditary melanoma mutation, accounting for 40% of familial cases), CDK4, BAP1, TERT, MITF, MC1R, and POT1. CDKN2A mutation carriers have a 28-76% lifetime melanoma risk versus 2.6% in the general White population.
TERT and BAP1 Mutations TERT promoter mutations - first identified in melanoma and now among the most common noncoding cancer mutations overall - allow premalignant cells to escape senescence. Somatic TERT mutations correlate with poor prognostic features including increased tumor thickness, ulceration, high mitotic rate, and lymph node metastasis. BAP1 germline mutations associate with cutaneous melanoma, ocular melanoma, mesothelioma, and renal cell carcinoma, and are recognizable in tissue by characteristic BAP1-deficient (Wiesner) nevi.
Clinical Tools The ABCDE criteria (asymmetry, border irregularity, color variability, diameter greater than 6 mm, evolution) has been used since the 1980s. With the 'E' evolution criterion, naked-eye diagnostic accuracy is approximately 65% overall. Dermatoscopy - using 10x magnification with polarization to visualize subsurface structures - improves sensitivity and specificity by up to 18% and 10% respectively per meta-analyses, though mainly in expert hands.
Limits of Dermatoscopy Evidence The 2023 Cochrane review concluded that the evidence base for dermatoscopy is limited, and that evidence was lacking to explicitly estimate sensitivity and specificity in real practice. Despite widespread adoption of dermatoscopy, evidence of decreased biopsy rates, cost savings, or improved patient outcomes remains unavailable. The tape-strip Pigmented Lesion Assay (DermTech) - analyzing RNA expression of Linc00518 and PRAME with greater than 99% negative predictive value and greater than 91% sensitivity - is a promising non-invasive alternative, but requires further independent validation.
AI in Clinical Dermatology One AI algorithm evaluated by Jaffe et al. achieved 100% sensitivity on 1,550 images, with 64.8% specificity compared to 69.9% for clinicians. Google's DermAssist (CE-marked in Europe) performed non-inferior to 6 dermatologists and superior to 12 primary care physicians for 26 skin conditions. However, 90% of DermAssist training images came from patients with lighter skin types, raising equity concerns. No FDA-approved AI device for pigmented lesion diagnosis currently exists in the US.
Interobserver Reliability Crisis Histopathological diagnosis remains the gold standard, but concordance rates are alarmingly low for ambiguous lesions. In the largest concordance study, only 25% agreement was observed for Spitz and atypical nevi, and 45% for atypical Spitz tumors, severely atypical nevi, and melanoma in-situ - rates that are unacceptably low for a diagnostic gold standard. This interobserver variability also creates a barrier to AI training and validation.
Immunohistochemistry Advances PRAME IHC - which detects the antigen overexpressed in melanoma - offers sensitivity of 67-94% for melanoma diagnosis. Loss of p16 (CDKN2A product) by IHC correlates strongly with malignancy. BRAF, BAP1, and cKit IHC stains provide guidance for targeted therapy selection. However, spindle cell desmoplastic melanomas have lower PRAME sensitivity, requiring continued reliance on S100 and SOX10 stains.
CGH, FISH, and GEP CGH identifies chromosomal copy number variations and has greater than 95% sensitivity for detecting abnormalities in melanoma versus only 13% in nevi. FISH - targeting 6p25, 6q23, 11q13, Cep6, CDKN2A, and MYC - achieves 94% sensitivity and 98% specificity with the 6-probe panel. The myPath 23-gene GEP test achieves 94-96% sensitivity and specificity in unambiguous lesions but drops to approximately 50% sensitivity for ambiguous melanocytic neoplasms - precisely the cases it was designed for.
AJCC 8th Edition Staging The 8th edition AJCC melanoma staging implemented in 2018 lowered the T1a/T1b cutoff from 1 mm to 0.8 mm Breslow depth. This directed more patients toward sentinel node biopsy, though the full clinical impact remains unclear and the sentinel node positivity rate appears unchanged in population studies. A more individualized staging approach incorporating molecular features has been proposed.
Gene Expression Profiling for Prognosis The DecisionDx-Melanoma 31-gene GEP test provides risk stratification scores (1A low risk, 1B/2A intermediate, 2B high risk) independent of AJCC staging. Multiple studies report it independently predicts metastatic risk for stage I-III melanoma. However, AAD and NCCN do not endorse routine use due to concerns about performance in stage I disease (over 70% of all new cases), poor performance in some meta-analyses, and the test's $7,193 cost with uncertain clinical benefit.
AI in Pathology AI augmented dermatopathology is in its nascent stage. Several studies suggest AI performs equal to or better than experienced pathologists for melanoma in artificial settings. However, barriers include the high cost of slide digitization, need for large validated training datasets, and the fundamental challenge that AI cannot resolve the diagnostic disagreement problem among dermatopathologists - the very disagreements that make training ground truth unreliable.
Overdiagnosis as the Most Urgent Problem The authors identify overdiagnosis of melanoma as the most pressing issue requiring attention. While debate continues about the degree of overdiagnosis, epidemiologic evidence makes a compelling case for a significant fraction of early melanomas being clinically insignificant. The immediate priority is identifying the small fraction of aggressive melanomas within the large pool of indolent early lesions currently being detected.
Shifting Detection Strategy Without evidence of mortality benefit, population-based skin cancer screening risks receiving a USPSTF grade of D (discourage screening), as occurred with thyroid cancer in 2017 - which resulted in decreased incidence without change in mortality. The dermatology community needs to generate evidence of screening benefit, particularly in high-risk populations, or risk a mandatory strategy shift imposed by regulatory bodies.
The Molecular Future The molecular classification of melanoma is expected to replace clinicopathologic classification as validated, convenient, and cost-effective molecular tests emerge. AI-augmented clinical and histopathologic diagnosis will streamline the diagnostic process. The next AJCC edition will incorporate molecular prognostic factors. Solving melanoma's overdiagnosis problem, however, will require cultural and regulatory shifts beyond technology - changes in biopsy thresholds, pathological criteria, and detection strategies.