RFP3. Accuracy of Melanoma Institute of Australia and Memorial Sloan Kettering Cancer Center Sentinel Node Metastasis Risk Prediction Tools applied to a single-institution cohort

Introduction: Sentinel lymph node (SLN) status is the most important prognostic factor in patients with primary cutaneous melanoma. Rates of SLN positivity vary greatly and risk prediction tools have been developed to guide clinical decision-making. In this study, we use our single-institution population to validate and compare two SLN risk calculators for melanoma developed by the Melanoma Institute of Australia (MIA) and Memorial Sloan Kettering Cancer Center (MSKCC).


Methods: Patients older than 18 years of age undergoing a SLN biopsy for cutaneous melanoma between January 1, 2018 and December 31, 2022 were reviewed. Patients were excluded if they had evidence of metastasis or clinically positive lymph nodes. Missing imputation of variables was not performed. Decision curve analysis compared each model to one another and to treat-all and treat-none SLN biopsy strategies. Statistical analysis was performed using R.


Results: 491 patients were included in our study, of which 317 and 167 had sufficient data for MIA and MSKCC risk score calculation, respectively. Median MIA and MSKCC estimated SLN risk scores were 9.0% and 11.8%, respectively. The actual SLN positivity rate in our patient cohort is 20.6%. Established prognostic factors were increased in SLN positive patients as expected (Table 1). The MIA model had significant discriminative ability between SLN positive and negative patients, whereas the MSKCC model did not (AUC: 0.67, p = 0.01 vs. 0.60, p = 0.1). In paired samples, the MIA model trended towards better discriminative ability than MSKCC. The MIA model demonstrated good calibration across clinically relevant ranges, whereas the MSKCC model tended to underestimate probability of SLN positivity. Both models had significant over-dispersion. The net-benefit analysis suggests that beyond the 12.5% threshold, the MIA model reduced the number of unnecessary SLN biopsies without missing additional SLN positive patients compared to performing a SLN biopsy on all patients. The MSKCC model only did so beyond 17.5%. The MIA model demonstrated greatest net-benefit over the MSKCC model at the 20% threshold. Using this threshold, the MIA model would avoid 28.6% of unnecessary SLN biopsies without missing SLN positive patients, compared to 9.2% with the MSKCC model.


Conclusions: In our patient cohort, the MIA model outperforms the MSKCC model, with the latter tending to underestimate SLN positivity. Additionally, the MIA model demonstrated clinical utility at lower thresholds and had greater net-benefit compared to the MSKCC model.