Prediction of Clinical Outcome for All Stages and Multiple Cell Types of Non-small Cell Lung Cancer in Five Countries Using Lung Cancer Prognostic Index()

Lung cancer is a commonly diagnosed cancer. In this era of personalized medicine, genetic predictive models are becoming increasingly important. However, many current predictive models fail verification tests due to small sample sizes and institutional biases. We collected 17 gene expression datasets from public databases to generate our largest training and testing cohorts. After successfully eliminating institutional variations and merging multiple datasets, we generated a training cohort of 1073 and a testing cohort of 659. Using Siggenes, univariate and multivariate analyses, we identified seven gene signatures, and combined them with the clinical parameter age and stage to design the lung cancer prognostic index (LCPI). Using LCPI, we could differentiate lung cancer patients into three risk groups and predict patient survival probabilities at 10 and 15 year post-surgical resection. We extensively verified the predictive ability of LCPI for overall and recurrence free survival using 6 other datasets from five different countries.