2.1 Study design and setting
In order to investigate the influence of benzene exposure, fat content, and their interactions on erythroid-related hematologic parameters, we conducted a cross-sectional study in workers from two state-run petrochemical companies which are located in Guangzhou and Maoming in Guangdong Province, China.
2.2 Study participants
We firstly recruited a total of 1802 workers and then intentionally excluded: a) workers who had been working for less than one year in their petrochemical plants; b) workers with self-reported and/or diagnosed carcinomas, hematological diseases, and/or immune diseases; c) workers taking any medicine in the preceding two weeks; d) workers experiencing X-ray examination for any reason in the preceding 1 month; and e) workers unwilling to provide biological samples or doing so in insufficient volume. Eventually, a total of 1669 workers were recruited in this study.
After participants signed informed consent, we administered a structured questionnaire to collect their information on demographic characteristics, lifestyle (such as smoking and drinking habits), medical history, and occupational experience (such as working years and workplaces). Participants who had smoked ≥ 1 cigarette/day for ≥ 1 year were considered as smokers, and those who drank wine and/or other alcoholic beverages at least once a week for ≥ 1 year were classified as drinkers. Workers that had been working in workshops such as petroleum refining, chemical production, and petroleum processing were considered as benzene-exposed group, while administration staffs were considered as non-exposed subjects.
After the interview, we collected ~20mL morning urine sample and ~2mL ethylenediaminetetraacetic acid (EDTA)-anticoagulated venous blood sample from each participant after overnight fasting. The present study had received ethical approval from the Ethical Review Committee at School of Public Health, Sun Yat-Sen University.
2.3 Measurement of urinary SPMA concentration
Urinary SPMA concentration was determined with liquid chromatography/electrospray tandem mass spectrometry (LC-MS/MS) described in detail previously [18]. In brief, 5mL urine sample for each participant was centrifuged at 800g for 5 min. An aliquot of 500 μL supernatant was mixed with 500 μL of 10 mM sodium acetate buffer (pH = 6.3), and treated with 50 μL of 10 μg/L SPMA-d5 working solution. After solid-phase extraction (Waters Oasis ® MAX), the samples were analyzed with LC-MS/MS (Agilent, US). SPMA (purity 98%) is used for qualitative analysis, and standard curves were drawn for quantitative analysis. SPMA-d5 working solution as internal standard was used for quality control. The inter-assay coefficient of variation (CV) (10 different sample preparations, 10 measurements at different days by different persons) was 10.21%. The detection limit for urinary SPMA was 0.01 μg/L and the concentrations of samples with levels below detection limit were substituted to 0.005 μg/L. SPMA concentration was standardized by urinary creatinine and expressed as μg/g creatinine.
2.4 Measurement of fat content indices
We measured the height of our participants using a wall-mounted stadiometer. The participants stood upright on a firm surface, looking straight ahead, arms at sides, with shoes removed and feet together. Their shoulders, buttocks, and heels were required to be touching the wall. We also measured the weight for participants with a calibrated standing balance scale. Participants removed shoes, heavy outer clothing, and items from pockets. Height and weight were measured twice for each participant to ensure accurate measurement. The average height and weight were used to calculate body mass index (BMI) according to the following formula:
BMI = Weight (kg)/Height (m)2
Then, we calculated BF% for each participant according to the following formula [19]:
BF% = (1.20 × BMI) + (0.23 × Age) –10.8 × Gender – 5.4
In this formula, “Gender” =1 for male workers, while “Gender” =0 for female workers.
Plasma samples extracted from EDTA-anticoagulated blood were used for the quantitative analysis of TC and TG by automatic biochemical analyzer (Cobas, Switzerland). The inter-assay CVs for TC and TG (10 different sample preparations, 10 measurements at different days by different persons) were 8.62% and 12.97%, respectively. Two professional sonographers separately detected the occurrence of fatty liver disease for each participant by abdominal ultrasound examination based on clinical diagnostic criteria of fatty liver [20], and the concordance rate was 100%.
2.5 Measurement of erythroid-related hematologic parameters
In the present study, we used automatic hematology analyzer (Sysmex, Japan) to measure seven commonly-used erythroid-related hematologic parameters in EDTA-anticoagulated venous blood samples, including RBC count (×1012/L), Hb concentration (g/L), hematocrit (HCT) (%), mean corpuscular volume (MCV) (fL), mean corpuscular hemoglobin concentration (MCHC) (g/L), RDW-coefficient of variation (RDW-CV) (%), and RDW-standard deviation (RDW-SD) (fL). These hematologic parameters were measured within 2 hours after sample collection, and each blood sample was assayed in duplicate. The average levels for these erythroid-related hematologic parameters were included in the following statistical analyses.
2.6 Statistical analyses
We examined the distributions of all numerical variables by Kolmogorov-Smirnov tests. As the urinary SPMA concentration had a left-skewed distribution, it was normalized by natural logarithm (ln) transformation before statistical analyses [21]. Several variables were adjusted as confounders in our statistical analyses, including age (continuous), gender (male/female), smoking status (smokers/non-smokers), drinking status (drinkers/non-drinkers), working years (continuous), and workplace (exposed/control group). As BF% was calculated based on age and gender, we didn’t adjust age and/or gender in all following analyses involving BF%.
We evaluated the differences of general characteristics between the exposed workers and controls with Student's t-test for continuous variables and chi-square test for categorical variables. We analyzed the between-group differences of SPMA concentration, fat content indices, and erythroid-related hematologic parameters by multivariate analysis of covariance, with adjustment for the above-mentioned confounding variables except for workplace.
RCS models, which have been widely-used to represent nonlinear relationships for continuous independent variables, were used to characterize patterns of changes in erythroid-related hematologic parameters with the changes in continuous exposure (including SPMA, BF%, TC, and TG) in the total population, while adjusting for the above-mentioned covariates. Continuous exposure was included into RCS models with 4 default knots located at the 5th, 35th, 65th, and 95th percentiles. The reference values on RCS curves were set at the median values for continuous exposure, and the values on Y axis represented the differences in erythroid-related hematologic parameters between individuals with any value of continuous exposure with those with median levels. The outputted P-overall indicate P values for test of overall association, and P-nonlinear indicate P values for test of nonlinear association. We used covariate-adjusted generalized linear models (GLMs) to analyze the effects of occurrence of fatty liver disease (no =0, and yes = 1) on erythroid-related hematologic parameters (as dependent variables) in the total population, and those of SPMA and all fat content indices with dependent variables in workers with different general characteristics. We further explored the modification effects of general characteristics on these associations by adding an interaction term of independent variable (continuous) and stratified variable (categorical) in GLMs.
Then, we divided our workers into three subgroups (T1, T2, and T3) according to the tertiles of BF%, TC, and TG, respectively, and divided workers into two subgroups according to the occurrence of fatty liver (No/Yes). We analyzed the associations of SPMA with erythroid-related hematologic parameters in subgroups with different fat content indices by covariate-adjusted GLMs, and explored the modification effects of fat content indices on these associations by adding an interaction term of SPMA (continuous) and fat content categories (categorical) in GLMs. In order to evaluate the combined effects of benzene exposure and fat content, we also divided our participants into four subgroups based on the median SPMA concentration (low/high) and the median continuous fat content indices or the occurrence of fatty liver. Workers with low SPMA and with low BF%, TC, TG, or without fatty liver were defined as the reference groups, and the differences of erythroid-related hematologic parameters between other subgroups and controls were evaluated with covariate-adjusted GLMs.
RCS analyses were performed by rms package in R software (Version 3.5.2), and other statistical analyses were performed using SPSS (Version 22.0). Two-tailed P < 0.05 was considered statistically significant.