Smoking, Cadmium Exposure, and Mortality

Continuous and categorical variables of demographic characteristics, underlying diseases, anthropometric index, and blood test results are presented using mean or frequency (%) according to smoking status, respectively. Independent t-tests and Pearson’s chi-squared tests were used to compare results. We also used sampling weights to account for multistage and stratified sampling. Multiple Cox regression analysis was utilized to assess the hazard ratio (HR) of smoking status with or without passive smoking and smoking pack years for all-cause and disease-specific mortality by adjusting for age, sex, race/ethnicity, alcohol consumption, and comorbidities at baseline survey, such as cancer classification, hypertension, diabetes, hyperlipidemia, and previous CVD events according to smoking status. The follow-up duration was calculated as the time from the first anthropometric and clinical measurements to death or last follow-up (December 31, 2019).
We performed propensity score matching (PSM) with age, sex, race/ethnicity, current smoking status, and smoking pack-years, considering the heterogeneity of demographic, clinical, and laboratory characteristics according to passive smoking status. We utilized 1:1 matching according to passive smoking status by the nearest neighbor method with a caliber of 0.25 using the R package “MatchIt”³⁷ . The Pearson correlation coefficient was used to investigate the correlation between (smoking pack years and cotinine) and cadmium concentrations before mediation analysis.
Using the R package “Regmedint”³⁸ , we performed regression-based causal mediation analysis to examine the direct influence of smoking status and the indirect effect via cadmium exposure. This R package is equivalent to the SAS mediation macro^{39 (link),40 (link)}. The total natural indirect effect (TNIE), pure natural indirect effect, total natural direct effect (TNDE), pure natural direct effect, and cumulative effect of smoking exposure and smoking status on mortality were calculated. R version 3.1.0 (R Foundation for Statistical Computing, Vienna, Austria; www.r-project.org) and the Statistical Package for the Social Sciences Statistics (version 24.0; IBM, Armonk, NY) were used for statistical analysis. Statistical significance was defined as p < 0.05.