|
Abstract:
|
High-dimensional regression models have been effectively applied to a wide array of genetic data. We applied these methods to study the survival of a sample of 460 patients with ovarian cancer from The Cancer Genome Atlas (TCGA). We investigated whether somatic mutations provided a benefit in predicting patient's survival. In addition to well established clinical predictors, 12,376 somatic mutations were included in a high-dimensional Cox regression model. This model was developed through 10-fold cross validation of a Cox proportional hazards model with an L1 penalty on the negative partial log likelihood (LASSO). The C-statistic was bootstrapped with 100 replications to quantify the discriminative ability of the Cox LASSO model on three sets of predictors: clinical features alone, somatic mutations alone, and clinical features along with somatic mutations. Due to the rare nature of somatic mutations, the overall improvement in predictive accuracy was modest. Nonetheless, the Cox LASSO model was able to identify several somatic mutations that meaningfully impact survival. These findings are clinically relevant for assessing prognosis and determining proper treatment.
|
