8289189 Beyond multiplicity: strategies for assessing work-related risks in the surveillance of occupational cancers in Italy

<h3>Objective</h3> To enhance the interpretation of occupational cancer risks within the BEST project (‘Big data and deep learning for the surveillance of occupational cancers’) by applying statistical methods that address multiple and selective inference, improving the prioritization of work sectors and cancer sites. <h3>Material and Methods</h3> Mortality data from the Italian National Statistics Institute (2005–2018) were linked to employment histories from the National Social Insurance Agency (1974–2018). Each individual was assigned the sector with the longest employment duration as a proxy for occupational exposure, applying a minimum latency of five years. Cancer Mortality Odds Ratios (CMORs) were estimated by cancer site and employment sector, adjusting for age, place of residence, year of death, and education level. Analyses were stratified by sex and occupational class, using the service sector as the reference. To address multiplicity and selective inference, synthesis methods included QQ-plots with guide rails, control of the Positive False Discovery Rate (q-values), and multivariate hierarchical Bayesian models for rank estimation and uncertainty quantification. These approaches allowed the generation of league tables of sector rankings with credibility intervals. This study is funded by the Italian Workers’ Compensation Authority (contract ID 56/2022), within the framework of the BEST project. <h3>Results</h3> In male blue-collar workers, CMORs were computed for 43 sectors and 10 cancer sites. Lung cancer showed excess mortality in transport and construction and a deficit in agriculture; highlighted heterogeneity across sectors. Additional sector-site differences emerged from q-value analyses. Bayesian rankings placed lung cancer in construction consistently at the top, with narrow 80% credibility intervals. <h3>Conclusion</h3> Rigorous synthesis methods that address multiplicity and selective reporting improve the interpretation of occupational cancer risks. Graphical and Bayesian approaches offer robust tools for identifying priority sectors and support evidence-based surveillance strategies.