Estimating a test's accuracy using tailored meta-analysis-How setting-specific data may aid study selection

Abstract

Objectives: To determine a plausible estimate for a test's performance in a specific setting using a new method for selecting studies.

Study design and setting: It is shown how routine data from practice may be used to define an "applicable region" for studies in receiver operating characteristic space. After qualitative appraisal, studies are selected based on the probability that their study accuracy estimates arose from parameters lying in this applicable region. Three methods for calculating these probabilities are developed and used to tailor the selection of studies for meta-analysis. The Pap test applied to the UK National Health Service (NHS) Cervical Screening Programme provides a case example.

Results: The meta-analysis for the Pap test included 68 studies, but at most 17 studies were considered applicable to the NHS. For conventional meta-analysis, the sensitivity and specificity (with 95% confidence intervals) were estimated to be 72.8% (65.8, 78.8) and 75.4% (68.1, 81.5) compared with 50.9% (35.8, 66.0) and 98.0% (95.4, 99.1) from tailored meta-analysis using a binomial method for selection. Thus, for a cervical intraepithelial neoplasia (CIN) 1 prevalence of 2.2%, the post-test probability for CIN 1 would increase from 6.2% to 36.6% between the two methods of meta-analysis.

Conclusion: Tailored meta-analysis provides a method for augmenting study selection based on the study's applicability to a setting. As such, the summary estimate is more likely to be plausible for a setting and could improve diagnostic prediction in practice.

Keywords: Data interpretation, statistical; Decision making; Diagnosis tests, routine; Mass screening; Meta-analysis; Models, statistical.