Grant: Doctoral: Vanier Canada Graduate Scholarships

Linking routinely collected data from different government programs (e.g., education and health data) to better understand the social determinants of health is challenging because different personal identifiers (e.g., student number vs. personal health identification number) are often used in different data sources. These data must instead be linked on non-unique and partial identifiers (e.g., given names, surnames, residential address, birthdate, sex/gender), which can lead to two types of errors: 1) missed matches, and 2) incorrectly matching records belonging to different individuals. When these errors are not random, but related to demographic factors such age or gender, they can introduce selection bias into the linked data. If there are significant differences between the distributed characteristics of individuals included in the linked sample compared to those in the underlying population, this reduces the accuracy and relevance of study findings. Currently, there is a need for recommendations on how to effectively reduce linkage errors. There is also a lack of understanding as to how linkage errors may affect study results. My research aims to minimize data linkage errors and ensure linked data samples are representative of the populations we wish to describe. I will test different methods for linking Manitoba administrative datasets and then examine how varying levels of linkage error influence results. To assess the practical implications of linkage errors on research, I will perform a case study that describes health disparities due to the social determinants of health across multiple Canadian regions. Additionally, I will assess the accuracy of new techniques that protect the privacy of study participants by first concealing personal information, before linking data sources. Lastly, I will evaluate whether using artificial intelligence to reduce the time and cost of linking data increases errors, adversely impacts study validity, or introduces bias.