Development and validation of a screening model for lung cancer using machine learning: A large-scale, multi-center study of biomarkers in breath.

OBJECTIVES

Lung cancer (LC) is the largest single cause of death from cancer worldwide, and the lack of effective screening methods for early detection currently results in unsatisfactory curative treatments. We herein aimed to use breath analysis, a noninvasive and very simple method, to identify and validate biomarkers in breath for the screening of lung cancer.

MATERIALS AND METHODS

We enrolled a total of 2308 participants from two centers for online breath analyses using proton transfer reaction time-of-flight mass spectrometry (PTR-TOF-MS). The derivation cohort included 1007 patients with primary LC and 1036 healthy controls, and the external validation cohort included 158 LC patients and 107 healthy controls. We used eXtreme Gradient Boosting (XGBoost) to create a panel of predictive features and derived a prediction model to identify LC. The optimal number of features was determined by the greatest area under the receiver-operating characteristic (ROC) curve (AUC).

RESULTS

Six features were defined as a breath-biomarkers panel for the detection of LC. In the training dataset, the model had an AUC of 0.963 (95% CI, 0.941-0.982), and a sensitivity of 87.1% and specificity of 93.5% at a positivity threshold of 0.5. Our model was tested on the independent validation dataset and achieved an AUC of 0.771 (0.718-0.823), and sensitivity of 67.7% and specificity of 73.0%.

CONCLUSION

Our results suggested that breath analysis may serve as a valid method in screening lung cancer in a borderline population prior to hospital visits. Although our breath-biomarker panel is noninvasive, quick, and simple to use, it will require further calibration and validation in a prospective study within a primary care setting.