Statistical Methods for Studies Using Respondent Driven Sampling with Applications to Urban Indigenous Health

Introduction: The health of Indigenous Peoples in Canada is, on average, poorer than the general population and in particular, the Indigenous community suffers from higher rates of cardiovascular disease. Respondent driven sampling (RDS) allows the health of urban Indigenous people to be studied using information about their connectedness. However, statistical methods for data arising from RDS studies are still being developed. The objective of this thesis was to evaluate the statistical aspects of RDS as a technique for studying the health of urban Indigenous communities. Methods: Four studies were completed : 1) A simulation study examining the validity of regression models in RDS data; 2) Development of a validated regression model to examine factors associated with cardiovascular disease among the urban Indigenous community living in Toronto and Hamilton, Ontario; 3) A survey of respondent driven sampling data sets from a variety of populations around the globe to describe their characteristics; and, 4) A simulation study to investigate the performance of estimators of disease prevalence using real-world RDS data. Results: Personal network degree in RDS studies is skewed, with a small proportion of people reporting many connections to others in their communities and is imprecisely reported by those with more than ten connections. Simulations studies indicated that the homophily configuration graph estimator is the preferred estimator for RDS data, and that weighted regression should be avoided because of the potential for inflated type I error rates. In addition to age, diabetes and hypertension, there is some evidence of a link between experiences of discrimination and cardiovascular disease in urban Indigenous communities. Conclusion: Respondent driven sampling is an effective tool for measuring the health of urban Indigenous communities in Canada. This work has identified important information regarding the distribution of RDS degrees, regression methods and best practices. These findings are important for validating analyses of RDS data. In addition to traditional risk factors, previous studies identified discrimination as a potential risk factor for cardiovascular disease and this work supports those findings. Discrimination is a modifiable exposure that must be addressed to improve cardiovascular health among Indigenous populations in Canada.