Individualized Prediction for Risk of Recurrence in Stage I/II Melanoma Patients With Negative Sentinel Lymph Node

Abstract

Background: Despite the favorable prognosis of AJCC stage I/II melanoma patients, up to 20%-30% will develop metastases. Our objective is to predict long-term risk (probability) of recurrence in early-stage melanoma patients.

Methods: A Risk Score to predict long-term recurrence was developed using Cox regression based on 2668 patients. Five clinicopathological risk factors were included. The association of the Risk Score with the risk of recurrence was evaluated using parametric models (exponential, Weibull, and Gompertz models) and compared to the Cox model using the Akaike information criterion. The discrimination of the model was measured by time-dependent ROC analyses. A calibration curve was used to evaluate the agreement between predicted and observed recurrence probabilities.

Results: The bootstrap adjusted C-index was 0.76 (95% CI, 0.74-0.79) overall and 0.87 (0.83-0.90) and 0.82 (0.78-0.85) at one and two years, respectively, and then remained above 0.70 up to 20 years. The Gompertz model for prediction of survival from the Risk Score showed the best performance and displayed good agreement with the Kaplan-Meier curves. The calibration curve of the Gompertz model showed a good agreement between predicted and observed 2-, 5-, and 10-year risk of recurrence. Population-level analysis demonstrated a significant association of Risk Score with risk of recurrence, with 10-year risks of recurrence of 4.5%, 13.0%, and 33.7% in the first, second, and third tertiles, respectively.

Conclusion: We developed a Risk Score to predict long-term risk of recurrence for early-stage melanoma patients. A Gompertz survival model fit to the Risk Score allows for individualized prediction of time-dependent recurrence risk.

Keywords: Gompertz model; negative sentinel lymph node; risk of recurrence; stage I/II melanoma.