Janakiraman Subramanian

MD, MPH

Inova Health System

Director Thoracic Oncology

PLCO (Learn more about this study)

PLCO-1154

Feb 8, 2023

A Machine Learning Model to Identify Patients at Risk for Developing Lung Cancer

Lung cancer is the most common cause of cancer related death both globally and in the United States. Despite improvement in systemic treatment for cancer, identifying patients at an earlier stage when they are eligible for curative treatment is essential to improving patient outcomes. Significant progress has been made in lung cancer screening by the use of low dose CT (LDCT) based on the findings from the National Lung Screening Trial (NLST) and NELSON trials. But only small proportion of patients with lung cancer are eligible for current screening criteria for LDCT. Further recent analysis identified PLCOm2012 was more effective than currently approved screening criteria in identifying patients at risk for lung cancer. However, PLCOm2012 model was based on traditional statistical modeling using logistic regression. We will apply machine learning inquiry to develop a model and compare its accuracy against currently approved LDCT screening criteria as well as the PLCOm2012 model.

Aim 1: To develop a machine learning based risk model that would predict for lung cancer risk in the control cohort of the PLCO lung dataset.

Aim 2: To test the machine learning based risk model in the intervention cohort of the PLCO lung dataset. Compare detection of lung cancer with the PLCOm2012 model. Further identify if combining with PLCO xray data would enhance the accuracy of lung cancer detection.

Aim 3: To validate the deep learning risk based model in an external dataset; the national lung screening trial (NLST) data, and compare with PLCOm2012 model.