Related Experiment Video
Updated: Jun 25, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Comparing the i31-SLNB and the MIA Nomogram for Sentinel Lymph Node Biopsy Positivity Prediction in Cutaneous
Rohit Sharma1, Rajan P Kulkarni2, Richard Essner3
1Marshfield Clinic Health System, Marshfield, WI, USA.
Introduction:
Sentinel lymph node biopsy (SLNB) for cutaneous melanoma (CM) provides essential prognostic information and is a component of the American Joint Committee on Cancer (AJCC) staging; however, up to 88% of SLNBs performed are negative. This means most patients who undergo the procedure do not benefit but are exposed to additional costs and surgery-associated morbidities. Improved tools are needed to identify patients at low risk of sentinel lymph node (SLN) positivity for whom the procedure is unnecessary and can be avoided. Here, we compared the sentinel lymph node positivity predictive accuracy of the integrated 31-gene expression profile (GEP) for sentinel lymph node biopsy (i31-SLNB) and the Melanoma Institute Australia (MIA) nomogram in patients with CM enrolled in the prospective DECIDE study who underwent SLNB.
Methods:
Data from patients with T1-T4 tumors and i31-SLNB test results from a prospective, multicenter study were evaluated. Area under the curve (AUC) was used to compare the discriminative performance of the i31-SLNB and MIA nomogram in predicting SLN positivity. Observed SLN positivity rates in patients with conflicting risk predictions were examined.
Results:
The analysis included 912 patients, of whom 430 were SLN-assessed. The i31-SLNB predicted < 5% risk of SLN positivity more often than the MIA nomogram and observed positivity was below 5% for the i31-SLNB (2.6%, negative predictive value [NPV] = 97.4%, 95% CI 94.2-100.0%) but not the MIA nomogram (5.8%, NPV = 94.2%, 95% CI 88.1-98.7%). The AUC for the i31-SLNB was 0.74, significantly higher than that of the MIA nomogram (0.61; p = 0.001), demonstrating better discriminative performance. At 5% and 10% risk thresholds, when the tools conflicted in their risk predictions, the i31-SLNB was accurate, with observed SLN positivity rates of 1.9% and 2.8% below each threshold and 11.5% and 25.8% above each threshold.
Conclusions:
The i31-SLNB more accurately stratifies patients under consideration for SLNB into groups based on risk of SLN positivity than the clinicopathologic-only MIA nomogram.
