The role of ethnicity in predicting diabetes risk at the population level
Often there is a need to predict the probability that an individual with specific characteristics (risk factors) will suffer from a certain disease or a health condition. Sometimes the real interest is in providing the aggregate predicted risk estimates at the population level or at the level of a subpopulation (e.g., males aged 65+). A related problem is prediction of such a probability for an individual who does not belong to the surveyed population. While the estimation of these probabilities follows from fitting an appropriate model to the available data, the design-based standard error estimation of so-obtained estimates is not obvious. We are proposing a bootstrap method for estimation of the standard errors of predicted individual and aggregate risks. The method is illustrated using data from the Canadian Community Health Survey.