|
Purpose: Broad-panel clinical sequencing is increasingly becoming a routine part of care for cancer patients. We aim to use statistical machine learning approach for assessing the utility of clinical sequencing data for survival risk stratification, accounting for biases and confounding factors in cohort studies.
Method: We developed OncoCast that incorporates various statistical learning methods including penalized Cox regression, random forest, and gradient boosting for predicting survival outcome. Prediction is based on an ensemble of models to boost performance.
Results: We assessed sensitivity and specificity of the method using various simulated data sets with a mix of discrete and continuous variables and varying effect sizes and noise levels. We will also demonstrate an application to a large clinical series of metastatic lung adenocarinomas with tumor samples sequenced by a broad-panel NGS platform (Shen, Martin, et al., JCO precision oncology, 2019, In Press).
Conclusion: OncoCast is a flexible tool with freely available software for genomic risk stratification.
|
