Study design and ethical approval

This retrospective cohort study was conducted among railway workers in the China Railway Shanghai Bureau Group through the Shanghai Railway Center for Disease Control and Prevention, China Railway Shanghai Bureau Group Co., Ltd. All railway workers in this group were enrolled in the study, totaling 64090 individuals. Their demographic and lifestyle characteristics were collected through self-reported electronic questionnaires, while clinical and biochemical health indicators were obtained through routine physical examinations. Railway workers were categorized into non-users, e-cigarette users, cigarette smokers, and dual users based on their responses to questions regarding smoking habits. The study adhered to the Declaration of Helsinki and was approved by the Ethics Committee of Xuzhou Medical University (Approval number: XZHMU-2023626).

Questionnaire survey and physical examination

The self-reported questionnaire addressed various aspects including: age, marital status (unmarried, married), education level (junior high school or below, high/vocational school, university/technical college, Master’s degree or above), working age, weekly working hours (18–78 hours), night work shifts per week (0, ≤1, 2–3, ≥4) and income per month (0–2999, 3000–4999, 5000–7999, 8000–10999, 11000–29999, ≥30000 RMB). The questionnaire regarding smoking habits was derived from the questions: ‘Do you smoke?’ (smoking refers to taking one or more cigarettes per day for more than a year), with four response options ‘yes’, ‘sometimes’, ‘never’, ‘have already quit’; if the first two options were selected, a subsequent question was asked, ‘Have you ever used traditional tobaccos or e-cigarettes?’ with three options ‘only traditional tobaccos’ (usually traditional tobaccos and took e-cigarettes less than twice a year on average), only ‘e-cigarettes’ (usually e-cigarettes and took two or less traditional tobaccos a year on average), both ‘e-cigarettes and traditional tobaccos’ (except for the above two situations). Answers to these questions defined the four subgroups used in the present analysis. The accumulated smoking years were also collected from the questionnaire. In addition, information regarding railway workers lifestyle habits were collected including: self-reported sleep quality (very bad, poor, good, very good), lack of energy last month (no, occasionally, sometimes, often have), alcohol consumption (never, occasionally, sometimes, quit), smokers among 5 best friends (0, 1, 2, 3, 4, 5 persons), cognition of smoking hazard (harmless, hard to say, slightly, relatively, very harmful), and self-reported health quality (poor, average, good, very good).

Lastly, clinical and biochemical health indicators were measured by a Grade-A tertiary hospital as part of the participants’ routine physical examinations, following standardized procedures. Body mass index (BMI, kg/m²) was obtained by dividing weight by the square of height. Cardiovascular indicators included systolic blood pressure (SBP), diastolic blood pressure (DBP), and heart rate (HR), all measured using an automated and calibrated digital sphygmomanometer while participants were in a seated and rested position. Hematological parameters were assessed via complete blood count (CBC) tests, using venous blood samples collected under sterile conditions. These included red blood cell count (RBC), hemoglobin concentration (HGB), platelet count, and white blood cell count (WBC). Metabolic indicators comprised fasting blood glucose (FBG, mmol/L), which was measured after an overnight fast using standardized diabetic testing protocols. Lipid profile variables included serum triglycerides (mmol/L), total cholesterol (μmol/L), high-density lipoprotein (HDL-C, mmol/L), and low-density lipoprotein (LDL-C, mmol/L). These were analyzed using enzymatic colorimetric assays in certified clinical laboratories. Liver function and carcinogenic indicators were assessed through blood chemistry analysis and urinalysis. These included alanine aminotransferase (ALT, U/L), aspartate aminotransferase (AST, U/L), γ-glutamyl transferase (GGT, U/L), total bilirubin (μmol/L), direct bilirubin (μmol/L), total protein (g/L), albumin (g/L), urea nitrogen (mmol/L), uric acid (μmol/L), and creatinine (μmol/L). Tumor markers, including alpha-fetoprotein (AFP, u/mL) and carcinoembryonic antigen (CEA, ng/mL), were measured using immunoassay methods. Urine pH was assessed via standard urinalysis.

Participants and data

In this study, all females and individuals who had quit smoking (n=5618) were excluded as the number of female smokers was very low (n=32), and smoking cessation could confound the effects of current cigarette exposure. Besides, we included female smokers in the screening process and found that only male smokers remained in the sample. Therefore, we excluded females in the first step. Outliers in continuous variables were identified using the interquartile range method and set as missing values. Variables with over 99% missing data (urinary glucose, urine protein, and urine ketones) were excluded. All remaining variables had <20% missing data. Individuals who started smoking before junior high school (when age minus accumulated smoking years was greater than 13 years, n=937) were eliminated as Chinese primary schools have strict smoking regulations for students, and it was unlikely for them to continue smoking during their primary school years. E-cigarettes were invented in China in 2003 and have been around for 20 years until 2022. Thus, e-cigarette users with accumulated smoking years ≥20 years were eliminated (n=147). Individuals with missing values were excluded from the non-user, cigarette smoker, and dual user subgroups. For e-cigarette users, Little’s test showed data were missing completely at random; therefore, missing values were imputed using multiple imputation by chained equations (MICE), following Rubin’s rules. To improve model interpretability, selected continuous clinical and biochemical indicators were discretized into ordinal categories using clinically meaningful bin widths. Most variables (e.g. blood pressure, heart rate, liver enzymes, hemoglobin, and creatinine) were binned into 10-unit intervals, while platelet count and uric acid were grouped into 100-unit intervals due to wider ranges. Each bin was labeled according to its upper bound, and numeric codes were assigned for regression modeling. All binned variables were checked and converted to a numeric format before analysis. After this rigorous data cleaning and imputation process, the final analytical sample consisted of 7719 participants: 4825 non-users, 1863 cigarette smokers, 524 e-cigarette users, and 507 dual users. The detailed flow for sample screening is shown in Figure 1 and the Supplementary file Section S1.

Figure 1

Flowchart illustrating the sample screening process for the study conducted among railway workers in Shanghai, China, March 2022 (N=7719)

Statistical analysis

All statistical analyses were done using R version 4.3.1 software. The continuous variables are described by the mean and standard deviation (SD), and the difference among the four subgroups was compared by analysis of variance (ANOVA) when homoscedasticity was satisfied; otherwise, the Kruskal-Wallis test was used. Categorical variables are summarized using frequencies (n) and percentages (%), and differences between subgroups are assessed using the chi-squared test (χ²). A multinomial logistic regression (MNLR) analysis was conducted to determine the potential health impacts associated with different types of cigarette use among male railway workers (see more details about MNLR in the Supplementary file Section S1). Sample size adequacy was determined according to the method of 10 events per variable (EPV) for a reliable regression analysis¹⁹. Given that 39 independent variables (including dummy variables) were included in the MNLR model, a minimum of 390 observations per outcome subgroup was required. The actual sample size in each subgroup exceeded this threshold, thus meeting the EPV criterion for model stability and validity. Before model fitting, multicollinearity was detected by running the corresponding linear regression model and calculating the variance inflation factor (VIF) for all independent variables included in the MNLR²⁰. As shown in Supplementary file Table S1, all VIF values were below 10, indicating the absence of significant multicollinearity. The findings of the MNLR are presented in the form of adjusted odds ratios (AORs) together with their 95% confidence intervals (95% CIs). Akaike information criterion (AIC) and receiver operating characteristic (ROC) curve were used to determine how well the model classified different types of smokers correctly. Additional analysis assessed the effects of accumulated smoking years on clinical and biochemical health indicators through multiple linear regression analyses. Each health indicator was modeled as an independent variable. All statistical tests were two-tailed, and a p<0.05 was considered statistically significant.

Comparison of baseline characteristics

The study included 7719 male individuals (62.5% non-users, 24.1% cigarette smokers, 6.8% e-cigarette users, and 6.6% dual users) with an average age of 33.0 ± 7.0 years (Table 1). A significantly higher proportion of unmarried individuals was observed across all four subgroups compared to married individuals (p<0.001). Significant differences were also noted among the four subgroups in terms of education level (p<0.001), night work shifts (p=0.010), self-reported energy levels in the past month (p=0.013), and self-rated health quality (p<0.001). Specifically, education level varied significantly among non-users, cigarette smokers, e-cigarette users, and dual users. Differences in night work shifts were evident between non-users and dual users, while reports of reduced energy in the past month were more frequent among e-cigarette users and dual users. Lastly, self-rated health quality differed significantly across all four subgroups.

For non-users, individuals with 2 smokers among their 5 best friends and those who had never consumed alcohol had the highest proportion, while for cigarette smokers, those with 3 smokers among their 5 best friends who occasionally consumed alcohol had the highest proportion. Perception of smoking hazards also differed: over 50% of non-users recognized smoking as very harmful, while only 40% of cigarette smokers and dual users considered it relatively harmful, and over 30% of e-cigarette users found the risk uncertain (‘hard to say’) (Table 1).

Regarding clinical and biochemical health indicators, e-cigarette users had the highest SBP and DBP, HR, HGB, ALT, AST, and CEA, while albumin, AFP, and urine pH were the lowest. Non-users had the lowest values of triglyceride, WBC, platelet, and GGT, but the highest values of HDL-C and total protein. Cigarette smokers had the lowest values of total bilirubin and direct bilirubin, but no difference was observed between non-users and e-cigarette users. Although platelet levels varied significantly among subgroups (p=0.02), pairwise comparisons were not statistically significant, likely due to sample variability or the conservative nature of Tamhane’s test. Further subgroup characteristics are detailed in Table 2.

Effects of different cigarette exposure

All MNLR models were adjusted for potential confounders, including age, body mass index, marital status, education level, monthly income, night work shifts, self-rated sleep quality, and alcohol consumption. The results of the MNLR, using non-users as the reference group, are presented in Table 3, and the corresponding results with cigarette smokers and e-cigarette users as references were demonstrated in Figure 2 and Supplementary file Table S2 to facilitate comparison across all subgroups.

Figure 2

Forest plots comparing the effects of e-cigarette exposure versus other product exposures on a range of clinical and biochemical health indicators among male railway workers in Shanghai, China, March 2022 (N=7719): A) Adjusted odds ratios and 95% confidence intervals (CIs) for various health indicators when comparing e-cigarette users with cigarette smokers; B) Comparison between dual users (individuals who use both e-cigarettes and traditional cigarettes) and e-cigarette users. Each forest plot illustrates the direction and strength of associations between different types of product exposure and the selected biochemical health indicators, with models adjusted for relevant covariates

The findings indicated that age was positively associated with dual users compared to both non-users and cigarette smokers (AOR=1.04; 95% CI: 1.00–1.08; AOR=1.05; 95% CI: 1.01–1.10) among male railway personnel. Being married was significantly and positively associated with cigarette smoking (AOR=1.20; 95% CI: 1.02–1.40) compared to non-use, while it was negatively associated with e-cigarette use and dual use compared to both non-use (AOR=0.71; 95% CI: 0.54–0.93; AOR=0.69; 95% CI: 0.53–0.90) and cigarette smoking (AOR=0.59; 95% CI: 0.44–0.78; AOR=0.57; 95% CI: 0.43–0.75). Participants who reported a higher frequency of alcohol consumption were more likely to be users of cigarettes, e-cigarettes, or dual cigarettes compared to non-users, even among those who indicated that they had quit drinking alcohol. Individuals who recognized smoking as a hazard exhibited diminished odds of being cigarette smokers, e-cigarette users, and dual users (AOR=0.64; 95% CI: 0.61–0.68; AOR=0.50; 95% CI: 0.46–0.54; AOR=0.65; 95% CI: 0.60–0.70) relative to non-users; it also holds for e-cigarette users relative to cigarette smokers (AOR=0.77; 95% CI: 0.71–0.84). Conversely, individuals who recognized smoking as a hazard exhibited elevated odds of being dual users relative to e-cigarette users (AOR=1.31; 95% CI: 1.18–1.45). E-cigarette users reported a higher self-rated health quality compared to non-users and cigarette smokers (AOR=1.24; 95% CI: 1.11–1.40; AOR=1.27; 95% CI: 1.13–1.44).

Compared to non-users, results from the MNLR indicated that cigarette smokers, e-cigarette users, and dual users had increased odds of elevated hemoglobin levels, WBC counts, and CEA levels. Specifically, for every 10 g/L increase in hemoglobin, the odds of use were significantly higher among cigarette smokers (AOR=1.22; 95% CI: 1.12–1.34), e-cigarette users (AOR=1.50; 95% CI: 1.29–1.73), and dual users (AOR=1.41; 95% CI: 1.22–1.63). Similarly, the odds of elevated WBC counts were higher among cigarette smokers (AOR=1.21; 95% CI: 1.16–1.27), e-cigarette users (AOR=1.27; 95% CI: 1.18–1.37), and dual users (AOR=1.16; 95% CI: 1.07–1.25). For CEA levels, the odds were also significantly increased per one-unit rise among cigarette smokers (AOR=1.12; 95% CI: 1.06–1.19), e-cigarette users (AOR=1.20; 95% CI: 1.09–1.33), and dual users (AOR=1.13; 95% CI: 1.02–1.24). Relative to cigarette smokers, e-cigarette users exhibited increased odds of elevated hemoglobin levels (AOR=1.22; 95% CI: 1.05–1.42) per 10 g/L increase.

Compared to cigarette smokers, e-cigarette users showed significantly increased odds of elevated heart rate (AOR=1.18; 95% CI: 1.06–1.33) per 10 beats per minute increase (Figure 2). Additionally, e-cigarette users demonstrated increased odds of elevated SBP (AOR=1.12; 95% CI: 1.01–1.24 vs non-users; AOR=1.18; 95% CI: 1.06–1.31 vs cigarette smokers) and elevated AST levels (AOR=1.27; 95% CI: 1.04–1.55 vs non-users; AOR=1.23; 95% CI: 1.00–1.51 vs cigarette smokers), per 10-unit increase. E-cigarette users exhibited reduced odds of higher urine pH levels (AOR=0.64; 95% CI: 0.52–0.80 vs non-users; AOR=0.70; 95% CI: 0.56–0.88 vs cigarette smokers), while dual users showed increased odds of higher urine pH levels compared to e-cigarette users (AOR=1.63; 95% CI: 1.23–2.15).

The ROC curve for MNLR is presented in Supplementary file Figure S1. The model showed varying levels of discrimination between non-users and the three smoking subgroups. Specifically, the model’s ability to differentiate between non-users and e-cigarette users was the highest (area under curve, AUC=0.794), followed by non-users and dual users (AUC=0.777), and non-users and cigarette smokers (AUC=0.752). These AUC scores were all greater than 70%, suggesting acceptable discrimination, and the corresponding results were reliable²¹. The corresponding pseudo coefficient of determination is 0.26, representing that 26.0% of the total variation was explained by the model.

The effects of accumulated smoking years

The results of the multiple linear regression analysis evaluating the impact of accumulated smoking years on clinical and biochemical indicators are presented in Figure 3 and Supplementary file Table S3. For cigarette smokers, the accumulated smoking years exhibited a significant and negative effect on SBP and AST (β= -0.1281, p<0.001; β= -0.0502, p=0.001), while no significant effects were observed in the other two smoking subgroups. The accumulated smoking years had negative effects on fasting blood glucose in both cigarette smokers and dual users (β= -0.0053, p=0.011; β= -0.0079, p=0.029), while it holds conversely true for alpha-fetoprotein (β=0.0217, p<0.001; β=0.0240, p=0.022). For e-cigarette users, the accumulated smoking years exhibited a positive effect on total cholesterol and CEA (β=0.0075, p=0.001; β=0.0212, p=0.001) and a negative effect on low-density lipoprotein (β= -0.0049, p=0.017), while no significant effects were observed in the other two smoking subgroups. Both high-density lipoprotein (β= -0.0019, p=0.017; β= -0.0053, p<0.001; β= -0.0039, p=0.004) and red blood cell (β= -0.0031, p=0.001; β= -0.0039, p=0.006; β= -0.0054, p=0.001) were negatively affected by the accumulated smoking years in all the three smoking subgroups; while it holds conversely true for hemoglobin (β=0.0587, p=0.018; β=0.1018, p=0.006; β=0.1212, p=0.005). WBC was positively affected by the accumulated years in both cigarette smokers and e-cigarette users (β=0.0278, p<0.001; β=0.0304, p<0.001); while it holds conversely true for total protein (β= -0.0436, p=0.002; β= -0.0797, p<0.001). For dual users, the accumulated smoking years showed a positive impact on urine pH (β=0.0061, p=0.046).

Figure 3

Unadjusted estimates from a linear regression model assessing the effect of accumulated smoking years on selected clinical health indicators across the three smoking subgroups (e-cigarette users, cigarette smokers, and dual users) of male railway workers in Shanghai, China, March 2022 (N=2894)

Limitations

Several limitations in this study should be acknowledged. First, the retrospective cohort design may introduce recall bias, particularly in self-reported variables such as cumulative smoking duration, which may differ from actual smoking histories. Second, the study did not differentiate between specific e-cigarette device types (e.g. vapor-based vs pod-based systems), limiting our ability to evaluate the potentially varied health impacts of different product designs. Third, the study sample consisted exclusively of male railway workers in Shanghai, China, and excluded females and certain smoking subgroups. This occupational and gender-specific sampling may introduce selection bias and limit the generalizability of the findings to broader or more diverse populations. Fourth, although alcohol intake was adjusted for in the analysis, AST levels, which were found to be elevated among e-cigarette users, may also be influenced by alcohol consumption. Given that alcohol use was self-reported and detailed drinking patterns were not captured, the potential for residual confounding remains. Fifth, clinical and biochemical health indicators were obtained during routine health examinations, which may be affected by unmeasured lifestyle factors such as diet, physical activity, or occupational exposures, further limiting external validity. Lastly, the findings are based on data from the Shanghai Railway Bureau, which may not reflect the broader working population or international contexts. Future studies should employ longitudinal designs, include female participants and non-occupational cohorts, and incorporate more detailed lifestyle and exposure assessments to validate and expand upon the observed associations.