Predictive performance of the clinicopathologic gene expression profile (CP-GEP) in identifying cutaneous melanoma patients for whom sentinel lymph node biopsy is unnecessary: A systematic review and meta-analysis

Abstract

Context & aim

Sentinel lymph node biopsy (SLNB) is an invasive procedure that detects microscopic nodal metastasis, crucial for accurate staging and optimal management. In melanoma, most patients who undergo the procedure have no sentinel lymph node (SLN) metastasis detected. The CP-GEP model (Merlin Assay) was developed to identify patients who do not have SLN metastases and who may therefore safely forgo SLNB, based upon clinicopathologic and gene expression features of the primary tumour. While the Merlin Assay has been validated by independent cohorts with relatively moderate sample sizes, this meta-analysis aims to assess the overall predictive performance of the model and examine potential heterogeneity between the external validation cohorts.

Method

We conducted a literature search in MEDLINE and Embase for studies that externally validated the CP-GEP model and were published between January 2019 and June 2024. Studies that reported the model’s sensitivity and specificity were included. Quality assessment was conducted by two reviewers using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. The outcomes investigated were sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and SLNB reduction rate, defined as the proportion of patients that can forgo SLNB based on the Merlin Assay outcome. Individual estimates were pooled using random effects meta-analysis model. Subgroup analysis was performed by T category.

Findings (Impact)

Four external validation studies (three retrospective and one prospective) were identified and included, involving 1099 participants. The pooled median age was 60 years and median Breslow thickness was 1.8 mm. Across all primary tumour classification groups (pT1-pT4), the pooled sensitivity was 93 % (95 % CI: 88 %-96 %), specificity was 32 % (95 % CI: 23 %-41 %), PPV was 24 % (95 % CI: 18 %-31 %), NPV was 95 %, (95 % CI: 92 %-97 %) and SLNB reduction rate was 27 % (95 % CI: 20 %-35). The subgroup analysis revealed that the model performed best in the pT2 group; pooled sensitivity was 91 % (95 % CI: 82 %-96 %), specificity was 35 % (95 % CI: 30 %-39 %), PPV was 20 % (95 % CI: 14 %-27 %), NPV was 96 %, (95 % CI: 91 %-98 %) and SLNB reduction rate was 31 % (95 % CI: 27 %-35 %).

Conclusion

This meta-analysis suggests that the CP-GEP model can reduce the number of unnecessary SLNBs performed, especially in pT2 patients. It can improve clinical decision making and assist in patients’ informed consent. Broader implications such as potential reductions in healthcare costs and risks of surgical complications should be explored further.

PROSPERO registration

CRD42024547893