Development and validation of an improved pathologic nodal classification system for cutaneous melanoma.

Abstract

e21563 Background: Given recent therapeutic advances and evolving patterns of lymph node (LN) evaluation for cutaneous melanoma, accurate and precise LN staging is needed to guide adjuvant treatment and future investigations. Current staging was developed primarily for patients undergoing completion LN dissection (CLND) for node-positive disease and do not produce LN classification groups with continuously increasing mortality. Thus, we developed and validated an improved LN classification system for cutaneous melanoma. Methods: Retrospective cohort analysis of 105,785 patients with cutaneous melanoma who underwent surgery from 2004 to 2015 in the National Cancer Database. Extent of LN dissection (sentinel LN biopsy [SLNB] and/or CLND) was available for patients diagnosed 2012 onward. Multivariable models were generated with number of positive LNs modeled using a non-linear restricted cubic spline function. Recursive partitioning analysis (RPA) was used to derive a modified LN classification system based on LN variables independently associated with overall survival (OS). The proposed LN classification system was validated in 85,499 patients from SEER-18. Results: Number of positive LNs (1-2 LN+: hazard ratio [HR] 2.48 per LN, 95% CI, 2.37-2.61, P< 0.001; ≥3 LN+: HR 1.10 per LN, 95% CI, 1.07-1.13, P< 0.001), clinically detected metastases (HR 1.35, 95% CI 1.27-1.42; P< 0.001), and in-transit metastases (HR 1.48; 95% CI 1.34-1.65; P< 0.001) were associated with OS. An RPA-derived LN classification system using these variables demonstrated continuously increasing mortality risk for each proposed LN classification group (HR: 1.83, 2.72, 3.79, 4.56, 6.15, and 8.25 for the proposed N1a-N3b groups, Table, P< 0.001). By contrast, AJCC 8E produced a more haphazard mortality profile (HR: 1.83, 3.81, 2.59, 2.71, 4.51, 3.44, 6.06, 8.15, and 6.90 for N1a-N3c). As a sensitivity analysis, the proposed system continued to accurately predict outcomes when we excluded patients undergoing CLND for microscopic LN metastases. Lastly, we validated this system for OS and cause-specific mortality in SEER-18 ( P< 0.001). Conclusions: A modified and simplified LN classification system can accurately predict mortality in cutaneous melanoma in an era of increasing use of SLNB without CLND and should be considered for future staging systems.[Table: see text]