A large, multinational cohort study confirmed that the Melanoma Institute Australia sentinel node metastasis risk calculator could accurately predict positivity in patients with primary cutaneous melanoma and performed consistently across diverse populations.
In the study, recently published in JAMA Dermatology, researchers included 15,731 patients from Europe, North and South America, and Australasia.
The calculator was designed to estimate individual risk using six clinicopathologic parameters: age, Breslow thickness, melanoma subtype (mandatory), mitotic rate, ulceration, and lymphovascular invasion (LVI) (optional). When all parameters were available (n = 4,989), the area under the curve (AUC) was 73.0% (95% confidence interval [CI] = 70.6%–75.3%). When one, two, or three optional parameters were missing, AUCs ranged from 70.1% to 71.5%.
AUCs by country ranged from 69.5% in the United Kingdom to 73.7% in New Zealand. The U.S. cohort showed an AUC of 74.0% in patients with complete data. Despite incomplete data in some cohorts—Sweden lacked mitotic rate data; Denmark lacked LVI—the model’s performance remained consistent.
Sentinel node positivity across the full cohort was 18.8%. Rates increased with American Joint Committee on Cancer T category: 6.5% for T1, 15.3% for T2, 30.8% for T3, and 40.7% for T4.
Calibration was high. The model’s intercept was 0.01 (95% CI = −0.02 to 0.03), and the slope was 1.03 (95% CI = 0.90–1.16). The Hosmer-Lemeshow goodness-of-fit test yielded a P value of .26. Decision curve analysis showed the Melanoma Institute Australia model offered greater net clinical benefit compared with performing biopsy on all patients or using simpler models with just Breslow thickness and ulceration, particularly at decision thresholds above 8%.
To improve precision, the researchers combined patient-level data from the development and validation cohorts to recalculate the variance-covariance matrix of the model coefficients. This led to a median reduction of more than 75% in 95% CI widths. For instance, a 64-year-old patient with a superficial spreading melanoma (1.1 mm thickness, mitotic rate 1/mm², no ulceration or LVI) would now have a risk estimate with a 95% CI of 8% to 9% compared with 4% to 13% previously.
Among all of the patients involved in the study, 31.7% of them had complete data; limited models were used for the remainder of the patients. Each model version corresponded to the number of parameters available, ensuring proper adjustment of coefficient weights.
The updated calculator, available at melanomarisk.org.au, provides the same point estimates but with narrower 95% CIs.
“These findings should provide users globally with greater confidence when using the tool in day-to-day clinical practice,” lead study author Serigne N. Lo, PhD, of the Melanoma Institute Australia in Sydney, Australia, and colleagues concluded.
Full disclosure can be found in the study.