Grant: Catalyst: Methods Post-Market DSE - Bayesian methods and statistical models

The occurrence of a disease in a population is often inferred from health services administrative data. This method of disease ascertainment may result in large proportions of false negative and false positive cases. To correct for these errors in disease classification, a subsample of the study subjects may be contacted and asked to report their disease status and/or invited for a clinical assessment. Statistical methods using Bayesian approaches are then used to combine the information obtained on the disease from the full sample with that obtained from the subsamples to infer the true state of the disease for all study individuals. In this project, we will assess the performance of the Bayesian method in correcting for disease misclassification. First, we will generate data for the true state of a disease for a sample of individuals. Second, we will hide the true data for some of these individuals and assume that their disease status was wrongly recorded in the database. Third, we will apply the Bayesian method on the misclassified data to investigate whether it is able to reduce the misclassification proportions. Finally, we will apply the Bayesian method to the study of the association between statin use and risk of new-onset diabetes using a random sample of 6,247 Quebec individuals that we obtained from the Quebec health services administrative databases. Among these individuals, about half responded to a survey on diabetes status and about a quarter also provided fasting capillary blood samples for glucose testing.