Data mining in the US using the Vaccine Adverse Event Reporting System.

Abstract

The US Vaccine Adverse Event Reporting System (VAERS), which is charged with vigilance for detecting vaccine-related safety issues, faces an increasingly complex immunisation environment. Since 1990, steady increases in vaccine licensing and distribution have resulted in increasing numbers of reports to VAERS. Prominent features of current reports include more routine vaccine co-administration and frequent reports of new postvaccination clinical syndromes. Data-mining methods, based on disproportionality analyses, are one strategy being pursued by VAERS researchers to increase the utility of its complex database. The types of analyses used include proportional reporting ratios, association rule discovery, and various 'historic limits' methods that compare observed versus expected event counts. The use of such strategies in VAERS has been primarily supplemental and retrospective. Signals for inactivated influenza, typhoid and tetanus toxoid-containing vaccines have been successfully identified. Concerns flagged through data mining should always be subject to clinical case review as a first evaluation step. Persistent issues should be subject to formal hypothesis testing in large linked databases or other controlled-study settings. Automated data-mining techniques for prospective use are currently undergoing development and evaluation within VAERS. Their use (as one signal-detection tool among many) by trained medical evaluators who are aware of system limitations is one legitimate approach to improving the ability of VAERS to generate vaccine-safety hypotheses. Such approaches are needed as more new vaccines continue to be licensed.