Development of a model to predict the age at breast cancer diagnosis in a global population.

Knowing the mean age at diagnosis of breast cancer (BC) in a country is important for setting up an efficient BC screening program. The aim of this study was to develop and validate a model to predict the mean age at diagnosis of BC at the country level. To develop the model, we used the CI5plus database from the IARC, which contains incidence data for 122 selected populations for a minimum of 15 consecutive years from 1993 to 2012 and data from the World Bank. The standard model was fitted with a generalized linear model with the age of the population, growth domestic product per capita (GDPPC) and fertility rate as fixed effects and continent as a random effect. The model was validated in registries of the Cancer Incidence in Five Continents that are not included in the CI5plus database (1st validation set: 1950-2012) and in the most recently released volume (2nd validation set: 2013-2017). The intercept of the model was 30.9 (27.8-34.1), and the regression coefficients for population age, GDPPC and fertility rate were 0.55 (95% CI: 0.53-0.58, p < 0.001), 0.46 (95% CI: 0.26-0.67, p < 0.001) and 1.62 (95% CI: 1.42-1.88, p < 0.001), respectively. The marginal R and conditional R were 0.22 and 0.81, respectively, suggesting that 81% percent of the variance in the mean age at diagnosis of BC was explained by the variance in population age, GDPPC and fertility rate through linear relationships. The model was highly accurate, as the correlations between the predicted age from the model and the observed mean age at diagnosis of BC were 0.64 and 0.89, respectively, and the mean relative error percentage errors were 5.2 and 3.1% for the 1st and 2nd validation sets, respectively. We developed a robust model based on population age and continent to predict the mean age at diagnosis of BC in populations. This tool could be used to implement BC screening in countries without prevention programs.