A prognostic model for predicting 10-year survival in patients with primary melanoma

Objective: To develop a prognostic model, based on clinical and pathologic data that are routinely available to the clinician, that would estimate the chance for survival of a patient with primary cutaneous melanoma after definitive surgical therapy. Design: Cohort analytical study. Setting: University medical center. Patients: 488 patients with primary cutaneous melanoma who had no apparent metastatic disease. Patients were followed prospectively for at least 10 years. An independent validation sample of 142 patients was used to assess the stability of the model. Measurements: Six clinical and pathologic variables that predict survival and are readily available to the clinician were used to develop a prediction model. The variables were tested for their association with death by using a univariate logistic regression model. Point estimates were generated for the probability of surviving melanoma at 10 years. Variables that were statistically significantly associated with survival were retained for testing in a logistic regression model. Results: 488 patients were followed prospectively for a median of 13.5 years (minimum, 10.0 years; maximum, 20.5 years). The overall 10-year survival of the study group was 78%. Four variables were found to be independent predictors of survival. Presented as adjusted odds ratios, from strongest to weakest relative predictive strength, these variables were tumor thickness (odds ratio, 50.8), site of primary melanoma (odds ratio, 4.4), age of the patient (odds ratio, 3.0), and sex of the patient (odds ratio, 2.0). The four-variable model was significantly more accurate than tumor thickness alone, particularly for predicting death. Overall, use of the model reduced the error rate of the prediction of death by 50%. Conclusions: A prognostic model that uses four readily accessible variables more accurately predicts outcome in patients with primary melanoma than does tumor thickness alone. This four-variable model can identify patients at high risk for the recurrence of disease, an identification that becomes increasingly important as adjuvant therapies are developed for treatment of melanoma.