797Understanding bias in recruitment through population-based cancer registries

Focus of Presentation Participant recruitment through registries can provide population-based samples, but there are methodological challenges. We examine challenges and solutions in two studies recruiting participants through the Victorian Cancer Registry. Cross-sectional surveys of patients with colorectal, lung or ovarian cancer conducted in 2013-14 (study 1) and repeated in an ongoing study (study 2). The registry managed fast-track recruitment of eligible patients ( > =40yrs, confirmed diagnosis, approached within 6-months of diagnosis) Findings In study 1, registry rules for confirmed diagnosis required surgical histopathology. For cancers primarily treated non-surgically (e.g. lung), selection bias was common. In study 2, confirmation rules were extended to include biopsies with subsequent hospital notification. The registry does not receive imaging notifications and pathology labs/hospital reports can be delayed (>60days). Study eligibility criteria meant cases with advanced disease, common in lung cancer, or without a confirmed diagnosis within the study timeframe were excluded. However, colorectal and ovarian cases in study 1 were largely representative of the eligible population. In study 2, a different fast-track recruitment approach meant a longer time from diagnosis to response compared to study 1 (median 7 vs 5 months). Recruitment procedures are being improved through engagement between researchers and registry. Conclusions/Implications Local registry-based rules for recruitment and available data, study eligibility criteria and cancer type (particularly cancers without histopathological confirmation), may introduce bias. Researchers need to work closely with registry colleagues to understand and optimise recruitment. Key messages Registry-based recruitment can be biased. Early and ongoing communication between researchers and registry is strongly recommended.