Study population
904 workers previously exposed to asbestos - already examined at the Occupational Health Service (OHS) of the Primary Care Trust (PCT) 12, Venice, Veneto Region (Italy) - were stratified by cumulative exposure and PP and 30% of them were extracted from each stratum with a random sampling procedure. The final sample included 263 subjects, 54 with and 209 without PP. These subjects received a letter of invitation and respondents were phoned one or two weeks later to fix an appointment for a medical examination. One hundred and ninety two out of 263 workers (73%) agreed to be examined. Since the original number of subjects with PP was only 54, in order to avoid small numbers, 8 additional newly examined workers with PP were included in the study, totalling 200 examined subjects.
After providing written informed consent, subjects were examined by occupational physicians, using the same protocol for collecting clinical and occupational histories and taking blood samples. Incidental findings were discussed with the patients and their primary care physicians and, where appropriate, referred for specialist evaluation. Information on smoking habits was collected and smoking cessation was recommended and facilitated for all patients.
Plasma samples and Osteopontin ELISA assay
Blood samples (12 ml) were collected by forearm venipuncture in BD Vacutainer sterile tubes containing K2-EDTA and processed within maximum 2 hours. The blood was centrifuged at 1,500 g for 20 minutes at room temperature. Plasma was then collected, aliquoted in sterile Safe Lock tubes, and stored at -25°C until further analysis. Repeated freezing and thawing cycles were avoided.
Plasma OPN was analyzed using an Osteopontin TiterZyme ELISA kit (Assay Designs, Ann Arbor, MI) according to the manufacturer's instructions and results were expressed in ng/ml. All samples were coded for a blinded analysis, and each plasma was determined in duplicate. Quality controls were analyzed in every plate. The analytical validity of the assay was also independently confirmed, and optimal results in terms of sensitivity, precision and recovery were obtained. The intra-assay precision was 6.5% at 24 ng/ml and 3% at 56.7 ng/ml, with analytical sensitivity of 1.78 ng/ml. Moreover, test of linearity (four dilutions from 1:2 to 1:16) and recovery from spiked samples gave acceptable results (mean recovery: 102% and 95.5%, respectively). All assays were carried out in a single session in the same laboratory (Regional Centre for the Study of Biological Markers of Malignancy, PCT 12, Venice, Veneto Region, Italy).
Statistical analysis
The mean, standard deviation (SD), median, 25% and 75% percentiles, and the results of Shapiro-Wilk test for normality were calculated in two groups of subjects (with or without PP) for OPN, age, time elapsed since first exposure (TSFE), time elapsed since last exposure (TSLE), length of exposure, peak asbestos level (highest asbestos exposure for any job held), and cumulative asbestos exposure. The two groups were then compared by calculating the non-parametric Wilcoxon rank-sum test statistics and the corresponding p-values.
Smoking was coded in three classes: non-smokers, ex-smokers (those who had given up smoking for at least one year), and current-smokers (if he/she regularly smoked at least one cigarette per day or a pipe or one cigar a week for at least a year or for shorter periods that added up to 12 months). The comparison of smoking classes between the two groups (with or without PP) was performed using the chi-square (χ2) test.
Several models of linear regression (and several bivariate scatter diagrams) were built, separately in workers with and without PP, where the dependent variable was always OPN and the independent variable was age or TSFE, TSLE, duration of asbestos exposure, peak exposure, cumulative exposure, smoking classes. To test whether the coefficient estimated over the group without PP was equal to the coefficient estimated over the group with PP, the Chow test statistics and the corresponding p-values were calculated.
Data of both groups were pooled together and OPN was regressed against age and asbestos exposure indices at univariable and multivariable linear regression analysis.
To convert OPN into a normally distributed variable, the inverse transformation (1/x) was chosen because it minimized the χ2 test for departure from normal distribution. The normal distribution of the transformed variable was then checked using a Q-Q plot (quantiles of 1/OPN against the quantiles of the normal distribution), a graph of kernel density estimates (Epanechnikov kernel function) against normal density, and the Shapiro-Wilk normality test.
At multivariable analysis, two models of linear regression were fitted where 1/OPN was the dependent variable and the above indicators acted as predictors. In Model 1, the associated risk factors were identified with the backward stepwise selection of predictors using 0.05 as criterion. In Model 2, age was introduced into the former model. In order to obtain more interpretable results, other models of multivariable linear regression were fitted where the dependent variable was OPN and the predictors were the same as the above. Normality of the residuals was checked using a Q-Q plot, a graph of kernel density estimates, and the Shapiro-Wilk normality test.
According to Checkoway [
8], the matrix of correlation coefficients "r" (and the scatter plot matrix) between all possible pairs of the independent variables was used to estimate collinearity (linear relationship between two predictors). Furthermore, when comparing a crude (univariable regression) with an adjusted (multivariable regression) estimate, the inflation of the standard error of a regression coefficient was used to check on the degree of multicollinearity (when more than two variables are involved), while the change in the regression coefficient was used to evidence a confounding effect.
All statistical analyses were carried out with STATA 10 (Stata Corporation, College Station, Texas, USA).