A model for pretest probability of lymph node metastasis from cutaneous melanoma

Although many statistical models have been developed to predict survival in cutaneous melanoma, few predict the end point of regional lymph node metas tasis shortly Lifter the diagnosis of melanoma. We used routine clinical an d histologic data from 573 patients referred to the Duke University Melanom a Clinic, Durham, NC, during the 1980s and 1990s who underwent lymph node r esections during the first year after the diagnosis of primary cutaneous me lanoma. The outcome we modeled (using the Logistic regression model) was th e probability of lymph node metastasis. We found that tumor thickness was t he variable most significantly associated with the probability of nodal met astasis, and the presence of ulceration and tumor location also were signif icant but age, sex, and mitotic rate were not. When the resulting logistic model predicted that the probability of nodal metastasis was more than .6, 93 of 115 patients hall nodal metastasis. When the model predicted that the probability was less than .3 just 32 of 88 patients had positive nodes. Fu rthermore, after the result of the node sampling was known, Cox model analy sis demonstrated that the pretest probability added significant information about subsequent survival.