The State of the Art in Diagnostic Test Meta-analysis

• Activity Number: 63
• Type: Invited
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Health Policy Statistics
• Author(s): Christopher H. Schmid*+
• Companies: Tufts-New England Medical Center
• Keywords: diagnostic tests ; meta-analysis ; ROC curves ; random effects ; multilevel model

Abstract

Diagnostic testing occupies an increasing share of the focus and costs of health care. This share will only increase as imaging and genetic-screening applications proliferate. Because each technique is usually evaluated in multiple studies, meta-analysis can help combine the results. Nevertheless, relative to reviews of treatment efficacy, new medical tests are infrequently reviewed. We believe that part of the reason is the lack of appropriate methods for combining diagnostic test data. We have surveyed the literature on meta-analysis of diagnostic tests of medical procedures and technologies, collecting every study published since 1990. We describe the methods used, their appropriateness, and whether use of more sophisticated analytic multilevel models might change the conclusions of the studies. Many of the methods employed are simple summaries of sensitivity and specificity or use the summary receiver operating curve. Most are fixed effects methods that do not properly account for between-study heterogeneity. We focus on aspects of study design that introduce heterogeneity between studies and explore whether certain factors may be able to explain this heterogeneity.