3086 - LASSO Regression to Identify Significant Patient Factors and Develop Survival Model in Patients with Cutaneous Melanoma: Results of 20,981 Cases from SEER Database

H. Wang, F. M. Kong, and Y. Zhang; The University of Hong Kong-Shenzhen Hospital, Shenzhen, China

Purpose/Objective(s): Survival prediction is challenging in patients with cutaneous malignant melanoma (CMM). Current models are mostly AJCC staging system based and limited in accuracy. This study aimed to identify significant patient and tumor factors in addition to AJCC staging and develop a comprehensive model for survival. Materials/

Methods: This is a study based on Surveillance, Epidemiology, and End Results (SEER) database. Patients diagnosed with CMM between 2018 and 2020 were included. Factors of our interest included AJCC staging, age at diagnosis, sex, race, pathologic characteristics of the tumors, such as anatomical site, primary site, breslow thickness, Clark level, tumor size, total number of in situ or malignant tumors at the patient level, laterality, LDH pretreatment level, site of metastasis (bone, brain, liver, lung, and lymph nodes), mitotic rate (0.1-11 mm²) and treatment modality (surgery, chemotherapy, and radiotherapy). The study endpoint was overall survival. The predictive power was compared using the time-dependent concordance index (time -index) and area under the time-dependent receiver Operating Characteristic Curve (time ROC).
Results: The study cohort included 20,981 eligible patients; 14,687 were randomly assigned to the training cohort. Age at diagnosis (HR:1.05), Mitotic rate (HR:1.05), Male gender (HR=1.45), Breslow thickness (HR:1.09), Ulceration (HR:1.62), AJCC-TNM [T2 (HR:2.19), T3 (HR:4.41), T4 (HR:3.37); N1 (HR :1.95), N2 (HR:2.29), N3 (HR:2.80); M1 (HR: 3.97)], diagnosis of bone metastases (HR=2.96), and brain metastases (HR:14.05) were significant for overall survival. These were incorporated into a novel and simple-to-use nomogram and available online (https://qq2918473747.shinyapps.io/20240221melanoma/). The concordance index for the cancer-specific survival rate was 0.908 (95% CI: 0.887- 0.929) in the training cohort. The areas under the curve for survival rates at 12 and 24 months were 92.5% and 91.4%, respectively. Validated indices and calibration plots indicated that this nomogram had good discrimination ability. The time ROC curve and time C-index curve were 88.6-93.0% and 87.8%-94.1%, respectively.
Conclusion: This study developed a model for survival prediction with excellent performance in patients with CCM, numerically better than that of AJCC and a recent published model. Further validation study is warranted.