Identifying modifiable risk factors of lung cancer: Indications from Mendelian randomization

Abstract

Background: Lung cancer is the major cause of mortality in tumor patients. While its incidence rate has recently declined, it is still far from satisfactory and its potential modifiable risk factors should be explored.

Methods: We performed a two-sample Mendelian randomization (MR) study to investigate the causal relationship between potentially modifiable risk factors (namely smoking behavior, alcohol intake, anthropometric traits, blood pressure, lipidemic traits, glycemic traits, and fasting insulin) and lung cancer. Besides, a bi-directional MR analysis was carried out to disentangle the complex relationship between different risk factors. Inverse-variance weighted (IVW) was utilized to combine the estimation for each SNP. Cochrane's Q value was used to evaluate heterogeneity and two methods, including MR-Egger intercept and MR-PRESSO, were adopted to detect horizontal pleiotropy.

Results: Three kinds of smoking behavior were all causally associated with lung cancer. Overall, smokers were more likely to suffer from lung cancer compared with non-smokers (OR = 2.58 [1.95, 3.40], p-value = 2.07 x 10-11), and quitting smoking could reduce the risk (OR = 4.29[2.60, 7.07], p-value = 1.23 x 10-8). Furthermore, we found a dose-response relationship between the number of cigarettes and lung cancer (OR = 6.10 [5.35, 6.96], p-value = 4.43x10-161). Lower HDL cholesterol could marginally increase the risk of lung cancer, but become insignificant after Bonferroni correction (OR = 0.82 [0.68, 1.00], p-value = 0.045). In addition, we noted no direct causal relationship between other risk factors and lung cancer. Neither heterogeneity nor pleiotropy was observed in this study. However, when treating the smoking behavior as the outcome, we found the increased BMI could elevate the number of cigarettes per day (beta = 0.139[0.104, 0.175], p-value = 1.99x10-14) and a similar effect was observed for the waist circumference and hip circumference. Additionally, the elevation of SBP could also marginally increase the number of cigarettes per day (beta = 0.001 [0.0002, 0.002], p-value = 0.018).

Conclusion: Smoking behavior might be the most direct and effective modifiable way to reduce the risk of lung cancer. Meanwhile, smoking behavior can be affected by other risk factors, especially obesity.

Fig 1. The basic principles underlying Mendelian randomization.

Fig 1A is the three assumptions for Mendelian randomization analysis. Fig 1B is the basic principles of bi-directional Mendelian randomization. SNP is the single nucleotide polymorphism and IV is the instrumental variable.

Fig 2. Forest plot of main MR results with lung cancer as the outcome.

Exposure represents risk factors; NSNP is the number of SNPs used to estimate the causal effect size; OR is the odds ratio; 95%LCI is the lower limit of 95% confidence interval; 95%UCL is the upper limit of 95% confidence interval. BMI is the body mass index; WHR is the waist-to-hip ratio; DBP is the diastolic blood pressure; SBP is the systolic blood pressure; HDL is the high-density lipoprotein; LDL is the low-density lipoprotein; 2-h glucose is the 2-hour glucose level of the oral glucose tolerance test. The units of effect measures are per 1-SD increase for continuous exposures and 1-unit increase in log OR of binary exposures.

Fig 3. Forest plot of MR results treating the number of cigarettes per day as the outcome.

Exposure represents risk factors; NSNP is the number of SNPs used to estimate the causal effect size; BETA is the effect size; 95%LCI is the lower limit of 95% confidence interval; 95%UCL is the upper limit of 95% confidence interval; BMI is the body mass index; WHR is the waist-to-hip ratio; DBP is the diastolic blood pressure; SBP is the systolic blood pressure; HDL is the high-density lipoprotein; LDL is the low-density lipoprotein; 2-h glucose is the 2-hour glucose level of the oral glucose tolerance test. The units of effect measures are per 1-SD increase for continuous exposures and 1-unit increase in log OR of binary exposures.

Fig 4. MR analysis of the effect of smoking initiation on lung cancer.

Fig 4A is the scatter plot of the MR result. Fig 4B is the forest plot of the leave-one-out sensitivity result.

Fig 5. MR analysis of the effect of smoking cessation on lung cancer.

Fig 5A is the scatter plot of the MR result. Fig 5B is the forest plot of the leave-one-out sensitivity result.

Fig 6. MR analysis of the effect of cigarettes per day of lung cancer.

Fig 6A is the scatter plot of the MR result. Fig 6B is the forest plot of the leave-one-out sensitivity result.