Introduction Mesothelioma is a malignant deadly disease primarily caused by exposure to asbestos. Previous studies have used simple modelling strategies to evaluate the burden of unreported mesothelioma. We built on previous work by creating a predictive multivariate regression model that incorporates novel data sources to improve the accuracy of estimates.

Methods As dependent variable, we used country-specific annual number of mesothelioma deaths from the WHO Mortality Database. As independent variables we initially used the following: 10-year average asbestos consumption, an indicator variable for asbestos-producer countries (data from the US Geological Survey), the proportion of the population aged 40 years or older, and the GDP per capita (Data from United Nations Statistics). Countries with non-missing data on mesothelioma mortality were used to fit a negative binomial regression model weighted by population size. Due to the latency between exposure to asbestos and onset of mesothelioma, we considered different lag times for the asbestos variables that ranged from 20 to 40 years.

Results Forty-seven countries were used to fit the model and 37 were used to predict the number of mesothelioma deaths. The latency period that resulted in the more robust model was 1975–1985 which corresponds to between 28 and 38 years between exposure to asbestos and disease onset. The Spearman correlation coefficient between mesothelioma deaths and annual asbestos consumption in 1975–1985 was 0.72. After model performance evaluation, only asbestos consumption and the proportion of the population aged 40 years or older were used for predictions. Based on our model, we estimated 5404 annual deaths from mesothelioma (95% CI: 3228–9235) in addition to the 15229 mesotheliomas that were reported to WHO.

Conclusions Findings from our predictive model suggest that a substantial proportion of mesothelioma deaths may be unreported in official aggregated data.