Role of Weights in Investigating Health Outcomes: An Example Using the Kentucky Women's Health Registry

Health registries are often used to better understand health issues in real world settings. Unfortunately, registries may not represent an accurate picture of the population since many registries are comprised of volunteers. Hence, the recruitment of participants to health registries often leads to a biased sample. In order to correct for potential discrepancies between the sample and the population of interest, it is possible to construct sample weights and make adjustments for unequal distributions in the sample on important variables. The KY Women's Health Registry (KWHR) was initiated in 2006 to better understand health issues particular to women. In addition to serving as a patient pool for clinical studies, it also serves as an epidemiological database for various health outcomes. Currently more than 10,000 women in KY have participated. However, this sample of women tend to be more educated (KWHR:58 % vs KY:16% attaining a college degree or higher), older (KWHR: 42% vs KY: 29% older than 55), and less racially diverse (KWHR: 93% vs KY: 85% white race alone, non-hispanic) than the general KY population of women (18 years and older). The construction and use of weights in this health registry will be discussed. Additionally, a comparison of weighted and non-weighted data will be presented as well as comparisons to a weighted multi-stage probability sample (BRFSS).