Background: Large record-linkage surveillance systems can screen many cancer–industry associations, but interpretation is made difficult by multiplicity and selective reporting, as well as the challenge of translating relative measures into public health impact (e.g., attributable deaths).
Objectives: To improve prioritization of occupational sectors and cancer sites within the BEST project (“Big data and deep learning for the surveillance of occupational cancers”) by integrating synthesis methods that address multiple and selective inference and attributable cases/fractions to support impact-oriented league tables.
Methods: Mortality data from the Italian National Institute of Statistics (ISTAT, 2005–2018) were linked to employment histories from the National Social Insurance Agency (INPS, 1974–2018). Each individual was assigned the economic sector with the longest employment duration as the exposure proxy, with a 5-year latency. Cancer Mortality Odds Ratios (CMORs) were estimated for selected cancer sites and sector using other cancer sites as controls and the service sector as reference (unexposed) through logistic regression, adjusting for age, region of residence, calendar year of death, and education; analyses were stratified by sex. Multiplicity and selective inference were addressed through: Q–Q plots with guide rails; q-values (positive FDR control); and multivariate hierarchical Bayesian models to obtain shrinkage estimates, posterior ranks, and 80% credibility intervals for league tables. To quantify burden, we derived absolute risk differences from CMORs using an internal baseline risk in the reference group, then estimated attributable deaths by multiplying risk differences by sector-specific person-years. Person-years were obtained from ISTAT workforce denominators, provided as annual averages, and converted to person-years by scaling to the observation window (2005–2018). Attributable fractions were calculated as attributable deaths divided by observed deaths in the corresponding sector-site group.
Results: CMORs were calculated for 43 sectors and 9 cancer types among male workers. Lung cancer showed higher mortality in the construction and transport sectors and lower in the agriculture sector, with marked variation across sectors. q-value analyses revealed additional sector–site patterns that need further investigation. Bayesian rankings consistently identified lung cancer in construction as a top priority, with relatively narrow 80% credibility intervals, supporting strong prioritization despite multiple comparisons. Attributable metrics enhanced prioritization by considering both effect size and sector size: for lung cancer, construction contributed approximately 1,981 attributable deaths (attributable fraction 5.5%). For pleural mesothelioma, shipyards had a high attributable fraction (around 16.8%), demonstrating how attributable fractions can reveal sector-specific risks even when absolute numbers vary.
Conclusions/Implications: Combining multiplicity-aware synthesis with attributable cases or fractions enhances the interpretation of occupational cancer surveillance results, reducing the overinterpretation of noisy extremes and highlighting potentially preventable burden. This integrated framework supports transparent, evidence-based priority setting for occupational surveillance and for planning sector-specific etiologic studies. This study is funded by INAIL (contract ID 56/2022) under the BEST project.