Background: Melanoma is a common cancer that causes a severe socioeconomic burden. Patients usually turn to plastic surgeons to determine their prognosis after surgery.
Methods: Data from hundreds of thousands of real-world patients were downloaded from the Surveillance, Epidemiology, and End Results database. Nine mainstream machine learning models were applied to predict 5-year survival probability and three survival analysis models for overall survival prediction. Models that outperformed were deployed online.
Results: After manual review, 156,154 real-world patients were included. The deep learning model was chosen for predicting the probability of 5-year survival, based on its area under the receiver operating characteristic curve (0.915) and its accuracy (84.8%). The random survival forest model was chosen for predicting overall survival, with a concordance index of 0.894. These models were deployed at www.make-a-difference.top/melanoma.html as an online calculator with an interactive interface and an explicit outcome for everyone.
Conclusions: Users should make decisions based on not only this online prognostic application but also multidimensional information and consult with multidiscipline specialists.
Keywords: Big data; Machine learning; Melanoma; Prognosis; Survival analysis.
Copyright © 2022. Published by Elsevier Ltd.