Comparison of two prognostic models predicting survival in patients with malignant melanoma

Accurate predictions of prognosis are important in the clinical management of patients with malignant melanoma. Primary lesions from 55 patients with cutaneous melanoma, having 10 or more years of clinical follow-up, were evaluated by two models predicting patient survival. One model was simple and relied solely on tumor thickness. The other model was complex and considered stage of tumor progression, and six clinical and histological variables. Accuracy of the two models was determined by retrospective review of the medical record, and the predictive power of each model was compared by receiver operating characteristic (ROC) curve analysis. The area under the ROC curve (a measure of the "goodness" of the model) for the single variable model was 0.70 +/- 0.0775 standard error (SE), and the area under the ROC curve for the multiple variable model was 0.77 +/- 0.0779 SE. Although a modest improvement in predictive power is suggested for the multiple variable model, the SEs for the two models overlap, and the difference is not statistically significant. Further study using a larger database may be required to determine definitively if the multiple variable model significantly increases the ability to predict patient survival or death, and make better clinical decisions.