An application of machine learning based on real-world data: Mining features of fibrinogen in clinical stages of lung cancer between sexes

Fangtao Yin; Hongyu Zhu; Songlin Hong; Chen Sun; Jie Wang; Mengting Sun; Lin Xu; Xiaoxiao Wang; Rong Yin

doi:10.21037/atm-20-4704

An application of machine learning based on real-world data: Mining features of fibrinogen in clinical stages of lung cancer between sexes

Ann Transl Med. 2021 Apr;9(8):623. doi: 10.21037/atm-20-4704.

Authors

Fangtao Yin¹, Hongyu Zhu¹, Songlin Hong², Chen Sun², Jie Wang^{3

4}, Mengting Sun^{3

4}, Lin Xu¹, Xiaoxiao Wang⁵, Rong Yin^{1

3

4}

Affiliations

¹ Department of Thoracic Surgery, The Affiliated Cancer Hospital of Nanjing Medical University & Jiangsu Cancer Hospital & Jiangsu Institute of Cancer Research, Jiangsu Key Laboratory of Molecular and Translational Cancer Research, Collaborative Innovation Center for Cancer Personalized Medicine, Nanjing, China.
² F&E Data Technology (Tianjin) Corp., Tianjin, China.
³ Department of Scientific Research, The Affiliated Cancer Hospital of Nanjing Medical University & Jiangsu Cancer Hospital & Jiangsu Institute of Cancer Research, Jiangsu Key Laboratory of Molecular and Translational Cancer Research, Nanjing, China.
⁴ Biobank of Lung Cancer, Jiangsu Biobank of Clinical Resources, Nanjing, China.
⁵ GCP Research Center, Affiliated Hospital of Nanjing University of Chinese Medicine, Jiangsu Province Hospital of TCM, Nanjing, China.

Abstract

Background: Lung cancer is the most threatening malignant tumor to human health and life. Using a variety of machine learning algorithms and statistical analyses, this paper explores, discovers and demonstrates new indicators for the early diagnosis of lung cancer and their diagnostic performance from large samples of clinical data in the real world.

Methods: By applying machine learning methods, including minimum description length (MDL), naive Bayesian (NB), K-means (KM), nonnegative matrix factorization (NMF), and decision tree (DT), based on large sample data of 2,502 patients, we built a classification model and systematically explored differences in fibrinogen levels in different clinical stages of lung cancer between the sexes. We also validated the reliability of the model by testing it on a validation cohort of 447 patients. This report adheres to the "Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis" (TRIPOD) statement for the reporting of prediction models.

Results: The analysis revealed significant differences in fibrinogen levels, pleural effusion, chlorine levels, A-G ratio, glutamic-oxaloacetic transaminase and alkaline phosphatase levels as well as in sex composition between the early-stage lung cancer group and the middle-late-stage lung cancer group. The classification model created by the combination of fibrinogen, alkaline phosphatase and sex demonstrated good performance with an AUC of 73.5%. In addition, in males, a fibrinogen level of 2.94 g/L could initially serve as the upper limit for determining the early-stage lung cancer group, but a level of 3.91 g/L could be preliminarily used as a reference threshold for the lower limit for middle- to late-stage lung cancer. This latter level could also serve as the upper limit of the critical value for early-stage lung cancer in females.

Conclusions: An integrated application based on supervised and unsupervised machine learning algorithms could effectively explore the potential links contained in the clinical data and reveal the differences in fibrinogen levels in different clinical stages of lung cancer between the sexes, which could provide a new reference basis for lung cancer staging.

Keywords: Machine learning; early diagnosis; fibrinogen; lung cancer; sex.