Results of fitting the models
| Covariable name (unit) | β2, ERR/unit exposure relative to external background (ie, with β1, fixed at unity) | β1, free, internal-to-external background ratio | β2, ERR/unit exposure relative to internal background (ie, with β1, free) | α, ERR/unit exposure | α1 and α2, ERR/unit exposure |
| Gamma (Gy) | NC | 0.90 (0.78 to 1.02) (p<0.001) | −1.27 (−2.4 to −0.14) (p=0.03) | −1.18 (−2.4 to 0.02) (p=0.055) | |
| Medium PA (years) | −0.010 (−0.019 to −0.001) (p=0.04) | 0.87 (0.74 to 0.99) (p<0.001) | −0.003 (−0.17 to 0.01) (p>0.5) | −0.003 (−0.016 to 0.011) (p>0.5) | |
| High PA (years) | −0.003 (−0.016 to 0.010) (p>0.5) | 0.81 (0.69 to 0.93) (p<0.001) | 0.01 (−0.01 to 0.03) (p=0.26) | 0.013 (−0.008 to 0.033) (p=0.24) | α1=0.021 (0.001 to 0.040) (p=0.04) |
| Underground work (105days) | −4.44 (−7.11 to −1.76) (p=0.001) | 0.93 (0.78 to 1.08) (p<0.001) | −3.30 (−7.20 to 0.06) (p=0.097) | −3.07 (−7.12 to 0.99) (p=0.14) | α2=−5.59 (−9.81 to −1.36) (p=0.01) |
The first three numerical columns are for the standardized mortality ratios (SMR) (from equation 2). The interpretation of β1 is that it represents the overall ratio of the observed background number of prostate cancers deaths in the cohort to the number of prostate cancer deaths expected in the comparison German Democratic Republic population, and β2 is the additional incremental SMR that is linearly dependent on the covariable of interest listed in the first column of the table, that is, an excess relative risk (ERR) per unit exposure of the covariable of interest. The fourth and fifth numerical columns are for the ERR internal regression models with background rates stratification on age and calendar year (from equations 3–5) for the univariate option in equation 5 (parameter α) and the multivariate option in equation 5 (parameters α1 and α2) models, respectively. Values given in parentheses represent 95% Wald type CIs, and the p values represent the statistical significance of the parameter values (and not the statistical significance of model improvement by their inclusion in the model). Models that did not converge are identified with NC. PA, physical activity.