A paper published in JAMA Network Open shows the potential for deep learning algorithms to assign individuals to two-year screening intervals, reducing the harm of screening and potentially making screening more cost-effective in some health care systems.
The paper, titled ‘Recalibration of a deep-learning model for low-dose CT images to choose lung cancer screening intervals’ was conceived and authored by clinicians at the National Cancer Institute (NCI) in the United States, with input from Dr Robbins at The International Agency for Research on Cancer (IARC) in France who developed the precursors to the comparator statistical models presented in this paper.
The work used data from the National Lung Screening Trial (NLST) and applying Optellum’s AI malignancy risk scoring. The authors demonstrated that by using recalibrated AI risk scores in their statistical model, they could predict lung cancer diagnosis with a two-year screen with good discrimination. The deep learning algorithm outperformed current American College of Radiology guidelines and comparator statistical models.
Over-utilization of screening by individuals increases potential harm to the patient – through increased radiation and increased risk of unnecessary invasive procedures – and cost to the healthcare system. Using AI to identify patients with a sufficiently low risk for a two-year screening interval is appropriate; benefits patients, providers, and payers.
Related links
