311Effective prospective study sampling from an existing large-scale national survey using propensity score methods

Abstract

Abstract Focus of Presentation Cohort study recruitment can be complex, often requiring extensive pre-screening to recruit sufficient numbers of exposed and unexposed participants. We discuss a prospective study of the impact of racism on adult health in New Zealand (NZ), with emphasis on propensity-score based sampling and recruitment methods accessing participants from a national survey (NZ Health Survey 2017/18, n = 12,530 eligible adults). Discussion will cover sampling design, response rate, and confounder balance in the final sample. Key empirical results will be summarised. Findings The NZHS provided a sampling frame with complete baseline exposure and covariate data, giving n = 2,099 exposed individuals (reported racial discrimination on NZHS). A propensity-score model (stratified by ethnicity) allowed invitation of unexposed individuals balanced on key sociodemographic confounders. Recruitment used postal invitations with telephone follow-up: individuals could respond by paper survey, online questionnaire or telephone interview. Response rate was 54%, with comparable rates in exposed/unexposed individuals, with key sociodemographic factors well-balanced by exposure status. Conclusions/Implications Racism is an important determinant of health inequity, with limited prospective research in New Zealand. Our approach enabled appropriate recruitment from a sampling frame with baseline exposure status (NZHS), including allowance for exposure variability by ethnicity. Propensity-score matching on baseline covariates allowed for balance on key confounders at invitation, with balance maintained in the final sample. Key messages Secondary sampling from large national surveys can provide efficient recruitment for prospective studies. We achieved a highly satisfactory response rate, and propensity-score based sampling substantially balanced confounders between exposed and unexposed groups, enhancing study validity.