Importance: The Melanoma Institute Australia (MIA) sentinel node (SN) metastasis risk calculator provides estimates of positivity for individual patients based on 6 standard clinicopathological parameters and the full 6-parameter model has been externally validated previously using US data. However, given its geographically widespread use, further validation is required to ensure its applicability to other populations.
Objective: To further externally validate the MIA SN metastasis risk calculator and increase its precision by refinement of the 95% CIs.
Design, setting, and participants: A retrospective multicenter cohort study was carried out using data from 4 continents, including the national Danish Melanoma Database and cancer centers in the UK (n = 3), US (n = 2), New Zealand (n = 1), Sweden (n = 1), and Brazil (n = 1). All patients aged 18 years or older who had an SN biopsy performed for an invasive primary cutaneous melanoma and data available on the following parameters: SN status, patient age at diagnosis, Breslow thickness, and melanoma subtype were included (n = 15 731). Available data were also collected on ulceration status, lymphovascular invasion, and the tumor mitotic rate. Data were collected between July 2021 and December 2023, and the analysis was conducted between January 2024 and June 2024.
Main outcomes and measures: The primary outcome was the area under the curve (AUC) of the receiver operating characteristics for the full (6-parameter) risk prediction model. Secondary outcomes were the AUCs for each country and for the limited models (3-5 parameters), the model calibration, and the recalculated 95% CIs for the models. Decision curve analysis was performed to assess the tool's clinical utility.
Results: The whole pooled cohort consisted of 15 731 patients; 4989 had all 6 parameters available. The AUC was 73.0% (95% CI, 70.6%-75.3%) in the subset with all 6 parameters available, and 70.8%, 71.5%, and 70.1% when 1, 2, or 3 optional parameters were missing, respectively. Calibration was excellent, with an intercept and calibration slope of 0.01 (95% CI, -0.02 to 0.03) and 1.03 (95% CI, 0.90-1.16), respectively. The updated 95% CI ranges were substantially tighter, with a median reduction of more than 75%.
Conclusions and relevance: This study found that the MIA SN-positivity calculator performed best with all 6 parameters and has been significantly improved (version 2), with the same risk point estimates but much tighter 95% CIs. These results demonstrated that the calculator was robust, precise, and applicable to geographically widespread melanoma populations.