Table 4.

Comparison of our model with previous models for identifying lung cancer risk factors.

Model	Population	Method	Risk factors	Accuracy	AUROCa
Our model	235,673	Deep neural network	As listed in the Results section	0.927	0.913
Panayiotis, 2016 [36]	25,486	Dynamic Bayesian network	Demographics, smoking status, family history of cancer, cancer history, comorbidities related to lung cancer, occupational exposures, and low-dose computed tomography screening outcomes	0.65	0.75
Wang, 2019 [37]	961	Conditional Gaussian Bayesian network	Age, sex, level of education, region, urbanization, diagnosis-based factors, prior utilization factors, prescription factors	0.67	N/Ab
Ankit, 2012 [38]	70,132	Decision tree	Age, birthplace, cancer grade, diagnostic confirmation, farthest extension of tumor, type of surgery performed, reason for no surgery, order of surgery and radiation therapy, scope of regional lymph node surgery	0.863	0.91
Xie, 2014 [39]	1703	Artificial neural network	41 risk factors: age, education level, marital status, income status, smoking, alcohol drinking, coffee intake, etc	0.838	N/A
Kaviarasi, 2019 [40]	321	Gaussian classifier	Age, sex, radiation sequence with surgery, first malignant primary indicator, radiation, etc	N/A	0.881

^aAUROC: area under the receiver operating characteristic curve.

^bNot available.