Association between lung cancer/multiple myeloma mortality and exposure to oncogenic viruses -- statistical analyses using non-model- and model-based statistical methods and various control sampling schemes for cancer mortality in occupational cohorts.

摘要

This study was designed to compare non-model- and model-based statistical techniques typically applied in cohort mortality analyses, and various schemes for selecting controls in nested case-control studies to document risk for lung cancer and multiple myeloma mortality, among workers in poultry slaughtering/processing plants. These workers are conceived to have a high exposure to oncogenic viruses compared to the general public. Data from the ongoing Cancer Risk in Workers Exposed to Oncogenic Viruses (CRIWETOV) project for members in a local Union Pension Fund belonging to the United Food &Commercial Workers (UFCW) international union, and followed--up for mortality from January 1, 1972 to December 31, 2003 were used for analyses. This cohort comprised of two large groups: poultry slaughtering/processing and non-poultry workers. The statistical methods applied were direct and indirect standardizations, Poisson, Cox proportional hazards, and binary/multiple logistic regression models and the sampling schemes for selecting controls were the cumulative survival, cumulative incidence, case-cohort, and incidence density sampling schemes. The entire cohort and sub groups of poultry and non-poultry separately had higher risks of mortality from both malignant diseases (statistically significant for lung cancer) compared to the United States' general population, but slightly lower (statistically not significant) risks among poultry compared to non-poultry workers. Results of comparative effect measures from the various statistical methods under consideration were similar with a very slight difference in variability/precision within the cohort analyses. The effect measures were also similar for nested case-control analyses that applied the cumulative survival, cumulative incidence and case-base sampling schemes in selecting controls. However, the incidence density sampling scheme led to markedly different results (both in magnitude and statistical significance), that were more profound with the Cox regression model. Where the Cox model was not appropriate the interval Poisson (exponential) model was used and predictions were similar to those obtained using other methods.