Using administrative health care system records to recruit a community-based sample for population research

Epidemiologists often seek a representative sample of particular persons from geographically bounded areas. However, it has become increasingly difficult to identify a sample frame that truly represents the underlying target population. We assessed the degree to which a clinic-based sample represents a target community.

Our sample frame is from a large health care provider from the Minneapolis-Saint Paul, Minnesota, metropolitan area. We used U.S. Census data to examine the sociodemographic and geospatial distribution of the sampling frame and among those who did and did not respond.

Our study's overall response rate was 57%. The most impoverished areas of the target population were under-represented in our sample frame, but this under-representation was similar for both respondents and nonrespondents. In addition, our sampled population was slightly older compared to the target population. Using ecological-level census-derived markers of sociodemographic characteristics, members of the sample frame were similar to that of the target population except for being somewhat more highly educated. However, the distributions of available individual-level data such as race and education were different between respondents and the target population.

Although the use of health care administrative records for identifying a sampling frame that represents a target population has limitations, our findings suggest that this method had strengths. More comparisons of methods for identifying and recruiting target populations are needed.