Classification of localized melanoma by the exponential survival trees method.

BACKGROUND

Over the past 2 decades, remarkable progress has been made in the identification of clinical and pathologic factors that affect the survival of patients with melanoma. Through the use of multivariate regression methods, key prognostic factors, such as tumor thickness, tumor ulceration, invasion level, and lesion location, have been identified. Clinical investigators are often interested in developing criteria to classify melanoma patients into different risk groups based on the key prognostic factors identified. However, classical multivariate regression models are generally less efficient in accomplishing this task than newly developed tree-based methods.

METHODS

In this study, the authors applied the exponential survival trees method to analyze a combined data set (n = 4568) from the University of Alabama at Birmingham and the Sydney Melanoma Unit in Camperdown, Australia. A survival tree was created according to prognostic factors that classified patients into homogeneous subgroups by survival. Six clinical and pathologic factors were included in the analysis. This tree-based method provided a superior means of prognostic classification and was shown to have greater ability to detect interactions among the variables than regression models.

RESULTS

Tumor thickness was found to be the most important prognostic factor, followed by tumor ulceration and primary lesion site. Some important interactions among these prognostic factors were identified. Five distinct risk groups, defined by tumor thickness, ulceration, and primary lesion site, were created. Patients who had tumor thickness less than or equal to 0.75 mm and lesions on their arms or legs had the best prognosis. Patients who had ulcerated tumors with thickness greater than 4.50 mm had the poorest prognosis.

CONCLUSIONS

The authors' analysis, based on exponential survival trees, provides a comprehensive, easy-to-use risk grouping system for classifying patients with localized melanoma. This grouping system would be useful in the clinical management of melanoma patients and in designing and analyzing clinical trials.