(B-207) Random Patient Selection in Retrospective Chart Review Studies: Cases, Considerations and Insights

Background: A common challenge in real-world data abstraction studies is the burden of manually reviewing patient medical records, particularly in retrospective chart review studies where sites are required to look across multiple timepoints. To reduce the burden for studies with large pools of potential patients or where a representative sample is required, random selection can be implemented. However, this method should be carefully designed not to introduce bias or additional site burden.

Objectives: To describe random selection processes from recent retrospective chart review studies and outline key considerations for selection of appropriate random selection methodologies.

Methods: A review was conducted of 3 global retrospective chart review studies: considerations, steps taken and learnings are provided for the random selection strategies.

Results: Three studies were conducted in gastroenterology and hepatology, across 13 countries targeting enrollment of approximately 2,900 patients total. To determine appropriate patient selection processes, the following were considered: suitability of non-random approaches (e.g., selecting patients with longest follow-up, or with highest quality data), level of selection (e.g., site-level, country-level, etc.), expected size of the population the sample would be drawn from (sampling frame), ability to filter for key eligibility criteria in electronic medical record (EMR) systems. For all 3 studies, there was the potential risk of non-random approaches introducing selection biases incongruent with study aims. Site-level random selection was used in all studies due to highly varied sample sizes among sites and ability to set patient cap per site. The size of the sampling frame varied compared to target sample size: Study 1 and Study 3- slightly larger; Study 2- greatly exceeds. Study-specific considerations for electronic data capture (EDC) systems used, site EMR capabilities and data abstraction timelines led to differences in screening methods. Study 1 had sites upload screening logs into the EDC system, which was programmed to randomly selected patients until the prespecified patient cap was met. In studies 2 and 3, sites used their EMR systems to filter and extract key enrollment criteria for patients into a screening log, to avoid manually entering large numbers of patients. Statistical programs were then utilized to generate a randomly ordered list which sites used for screening until the target sample size was met.

Conclusions: Random patient selection can be utilized in retrospective chart review studies that have larger than necessary potential sampling frames, however, methodologies need to be uniquely tailored to each study design/population and EMR system capabilities.