Using logistic regression to improve the prognostic value of microarray gene expression data sets: application to early-stage squamous cell carcinoma of the lung and triple negative breast carcinoma

摘要: Numerous microarray-based prognostic gene expression signatures of primary neoplasms have been published but often with little concurrence between studies, thus limiting their clinical utility. We describe a methodology using logistic regression, which circumvents limitations conventional Kaplan Meier analysis. applied this approach to thrice-analyzed and squamous cell carcinoma (SQCC) the lung data set, objective identifying expressions predictive early death versus long survival in early-stage disease. A similar analysis was set triple negative breast cases, present challenges. Important our is selection homogenous patient groups for comparison. In study, we selected two (including only stages I II), equal size, earliest deaths longest survivors. Genes varying at least four-fold were tested by regression accuracy prediction (area under ROC plot). The list refined applying sliding-window analyses validations leave–one-out model building validation subsets. used after selecting appropriate cases total 8594 variable genes predicting survivors SQCC. After sliding window leave-one-out analyses, 24 identified; most them B-cell related. When same stage II analyzed (KM) approach, identified fewer immune-related among statistically significant hits; when III included, missed. Interestingly, cancer many outcome. Stratification based on data, careful comparison, application substantially improved comparison KM approaches. B cell-related dominated SQCC cancer.