Individualized prognosis for melanoma patients

Abstract

The clinical course of malignant melanoma is notoriously variable. Current approaches to prognostication allow assignment to risk categories but do not permit accurate assessment of prognosis on an individual patient basis. We analyzed a melanoma histology database that comprises 1,042 sequential melanoma patients evaluated by A.J.C. at UCLA between 1980 and 1990 for 30 separate variables according to a standard protocol. After censoring for absent data, a univariate Cox model analysis was performed that showed 20 individual variables that were significantly linked to clinical outcome. A step-up multivariate analysis was then performed. The combined analysis shows 5 variables: gender, site of primary, age relative to 60 years, Breslow thickness, and presence and width of ulceration to be linked to survival. Probability of survival is calculated using a 2-step approach. The survival-linked variables are multiplied to give an individualized risk score. This is converted into probability of survival by the formula .987 (risk score) for 3-year survival, .975 (risk score) for 5-year survival, and .960 (risk score) for 10-year survival. Thus, a 55-year-old woman with a 1.8-mm nonulcerated melanoma on the leg would have a risk score of (1 x 1 x 1 x 2 x 1) = 2 and a predicted probability of survival at 5 years of .9752 (95%) and at 10 years of .9602 (92%). We used similar techniques to develop individualized risk scores for likelihood of recurrence. The significant variables in this case are anatomic site of the primary melanoma, melanoma subtype, Breslow thickness, and presence and width of ulceration. The formulae for likelihood of recurrence at different periods after initial surgical removal of the primary melanoma are at 3 years, .979(risk score); at 5 years, .971(risk score); and at 10 years, .957(risk score). This relatively simple approach to prognostication uses readily available demographic information and is likely to be more accurate than single-factor analysis.