329 Impact of Ultraviolet Exposure on Merkel Cell Carcinoma Long-Term Survival

Annabelle Y. Guo,Indra Neil Sarkar,Elizabeth S. Chen,Joanna Walker,Paul Stey,Wen Qing Li,Eunyoung Cho,Abrar A. Qureshi
DOI: https://doi.org/10.1016/j.jid.2018.03.335
2018-01-01
Abstract:Annabelle Y. Guo, PhD1, Indra Neil Sarkar, PhD1,3, Elizabeth S. Chen, PhD1,3,Joanna Walker, MD2, Paul Stey, PhD3, Wen Qing Li, PhD2, Eunyoung Cho, ScD2, Abrar A. Qureshi, MD21The Warren Alpert Medical School; 2Department of Dermatology; 3Center for Biomedical Informatics, Brown University, Providence, RI. Merkel Cell Carcinoma (MCC) is an aggressive malignant skin cancer with a high mortality rate. The effect of ultraviolet (UV) exposure on MCC patients long-term survival is not fully understood. With 5064 deceased MCC patients data from the Surveillance, Epidemiology, and End Results Program (SEER), a random forest based machine learning model was used to predict MCC patients 5-year survival following diagnosis and identified UV exposure (estimated based on county of residence) as a significant prognostic factor among 19 other prognostic factors (year of diagnosis, age, primary tumor site, tumor size, regional lymph node involvement, marital status, tumor stage, etc.). UV exposure was inversely associated with number of survival months in a linear regression analysis. To build a prediction model for MCC long-term survival, the dataset with the prognostic factors and UV exposure was split into training (70%, n=3545) and testing (30%, n=1519) datasets. A survival prediction model was built from the training dataset using the random forest machine learning method. The model was evaluated with the testing dataset, which performed at a prediction accuracy of 71% and c-statistic (area under the receiver operating characteristic curve) of 0.74. Among the 20 prognostic factors, UV exposure, year of diagnosis, and age at diagnosis were the top three predictors based on the mean decrease in Gini coefficient. In summary, these data suggest an association between UV exposure and MCC survival, and provide foundation for prognostication using machine learning methods.
What problem does this paper attempt to address?