Making Smart Clinical Decisions with Genomic Data and Machine Learning: An Example of Lung Cancer Diagnosis

• Activity Number: 398 - Statistical and Computational Methods to Tackle Complex Diagnostic Challenges
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Jianghan Qu and Shuyang Wu and Lori Lofaro and Joshua Babiarz and Sangeeta Bhorade and Sean Walsh and Giulia Kennedy and Jing Huang*
• Companies: Veracyte and Veracyte and Veracyte and Veracyte and Veracyte and Veracyte and Veracyte and Veracyte Inc
• Keywords: machine learning; cancer diagnosis; RNA sequencing; genomics; lung cancer; ensemble

Abstract

The management of lung nodules is challenging among patients with inconclusive bronchoscopy. The Percepta Genomic Sequencing Classifier (GSC) was developed to accurately up-classifying as well as down-classifying cancer risk among these patients by leveraging multiple cohorts and sequencing. To address demographic heterogeneity and interfering factors, we developed three strategies: 1) ensemble of clinical dominant and genomic dominant models; 2) development of hierarchical regression models; and 3) targeted placement of genomic and clinical interaction terms to stabilize interference. The final model uses 1,232 genes and four clinical covariates. In the validation set of 412 patients, the GSC down-classified low and intermediate pre-test risk subjects to very low and low risk with specificity=45%, sensitivity=91%, and NPV=95%. 12% of intermediate pre-test risk subjects were up-classified to high risk with PPV=65%, and 27% of high pre-test risk subjects were up-classified to very high risk with PPV=91%. The GSC provides physicians actionable information to make an early diagnosis of lung cancer in malignant nodules while decreasing invasive procedures in those with benign nodules.