population. Because no denominator data are available, VAERS cannot be used to evaluate causality. The VAERS data are useful, however, for the development of adverse event signals and the formation of related hypotheses that can be further tested and validated by more robust methods.
Vaccine Safety Datalink
One system better suited to the testing of hypotheses about vaccine safety is VSD. The VSD project was formed in the 1990s as a collaborative effort between CDC and a group of managed care organizations (MCOs) to maintain a large linked database for monitoring immunization safety and studying potential rare and serious adverse events. The number of VSD member sites has increased over the years and now includes nine MCOs that enroll approximately 9.5 million children and adults, or about 3 percent of the U.S. population. VSD sites are located at geographically diverse locations in California, Colorado, Georgia, Hawaii, Massachusetts, Michigan, Minnesota, Oregon, and Washington (Frank DeStefano, CDC, personal communication, October 18, 2012). Because the data in the database are generated as a by-product of the routine administration of health care and the system does not rely on voluntary adverse event reporting (as VAERS does), the problems of underreporting and recall bias are reduced.
VSD is a useful system that includes demographic data and information on the medical services that have been provided to those enrolled in the health plans, such as age and gender; vaccinations; hospitalizations; outpatient clinic, emergency department, and urgent care visits; mortality data; and additional birth information (e.g., birth weight) (Baggs et al., 2011). Automated pharmacy and laboratory data as well as information on diagnostic procedures (e.g., radiography and electroencephalography) that the patient has undergone are also included (Chen et al., 2000). Data on adverse events, including deaths (from probabilistic matching of death files), are routinely collected (Chen et al., 1997). Covariates used to control for potential confounders include birth certificates and variables from the decennial census at the zip code level, in addition to demographic data from the health plans.
Each site collects data on vaccinations (the type, date, and concurrent vaccinations), medical outcomes (diagnoses and procedures associated with outpatient, inpatient, and urgent care visits), and birth and census data. To ensure compliance of federal regulations and to protect confidentiality, each person within the VSD is assigned a unique random VSD study identification number which is not linked to their MCO member identification number. These VSD study identification numbers can be used to link data on demographics and medical services (Baggs et al., 2011).
Since 2001, VSD has used a distributed data model whereby each MCO