Cancer: Machine-learning approach could improve lung cancer screening

A machine-learning method that can identify patients with early-stage lung cancer is described in Nature this week. The approach detects tumour-derived DNA in blood samples — so-called liquid biopsies — and may help to increase the adoption of screening for high-risk individuals.

The recommended mode of screening for lung cancer in high-risk individuals is CT scanning, which has been shown to reduce lung cancer-related deaths. However, uptake is low owing to factors such as high costs, limited screening programmes and concerns with false positives; around 5% of eligible individuals undergo such screening in the United States. Blood tests are an attractive alternative means of cancer detection, although most liquid biopsy studies focus on monitoring patients with advanced disease, who may have higher levels of tumour-related DNA markers than early-stage patients.

Maximilian Diehn and colleagues optimize an existing sequencing method for assessing circulating tumour DNA (ctDNA). They improve the recovery of DNA and identify alterations that may serve as useful markers of disease. Using this method the authors show that although ctDNA is present in only very low levels in early-stage lung cancers, it is a strong prognostic marker. They go on to use these data to refine a machine-learning method for predicting the presence of lung-cancer-derived DNA in blood samples. This technique discriminates early-stage lung cancer patients from risk-matched controls in an initial sample of 104 patients with early stage non-small cell lung cancer and 56 matched controls, which was confirmed in an independent validation cohort of 46 cases and 48 controls.

doi:10.1038/s41586-020-2140-0

Return to research highlights