Variation of a test's sensitivity and specificity with disease prevalence

BACKGROUND Anecdotal evidence suggests that the sensitivity and specificity of a diagnostic test may vary with disease prevalence. Our objective was to investigate the associations between disease prevalence and test sensitivity and specificity using studies of diagnostic accuracy. METHODS We used data from 23 meta-analyses, each of which included 10-39 studies (416 total). The median prevalence per review ranged from 1% to 77%. We evaluated the effects of prevalence on sensitivity and specificity using a bivariate random-effects model for each meta-analysis, with prevalence as a covariate. We estimated the overall effect of prevalence by pooling the effects using the inverse variance method. RESULTS Within a given review, a change in prevalence from the lowest to highest value resulted in a corresponding change in sensitivity or specificity from 0 to 40 percentage points. This effect was statistically significant (p < 0.05) for either sensitivity or specificity in 8 meta-analyses (35%). Overall, specificity tended to be lower with higher disease prevalence; there was no such systematic effect for sensitivity. INTERPRETATION The sensitivity and specificity of a test often vary with disease prevalence; this effect is likely to be the result of mechanisms, such as patient spectrum, that affect prevalence, sensitivity and specificity. Because it may be difficult to identify such mechanisms, clinicians should use prevalence as a guide when selecting studies that most closely match their situation.

Diagnostic value of immunoassays for heparin-induced thrombocytopenia: a systematic review and meta-analysis.

Immunoassays are essential in the workup of patients with suspected heparin-induced thrombocytopenia. However, the diagnostic accuracy is uncertain with regard to different classes of assays, antibody specificities, thresholds, test variations, and manufacturers. We aimed to assess diagnostic accuracy measures of available immunoassays and to explore sources of heterogeneity. We performed comprehensive literature searches and applied strict inclusion criteria. Finally, 49 publications comprising 128 test evaluations in 15 199 patients were included in the analysis. Methodological quality according to the revised tool for quality assessment of diagnostic accuracy studies was moderate. Diagnostic accuracy measures were calculated with the unified model (comprising a bivariate random-effects model and a hierarchical summary receiver operating characteristics model). Important differences were observed between classes of immunoassays, type of antibody specificity, thresholds, application of confirmation step, and manufacturers. Combination of high sensitivity (>95%) and high specificity (>90%) was found in 5 tests only: polyspecific enzyme-linked immunosorbent assay (ELISA) with intermediate threshold (Genetic Testing Institute, Asserachrom), particle gel immunoassay, lateral flow immunoassay, polyspecific chemiluminescent immunoassay (CLIA) with a high threshold, and immunoglobulin G (IgG)-specific CLIA with low threshold. Borderline results (sensitivity, 99.6%; specificity, 89.9%) were observed for IgG-specific Genetic Testing Institute-ELISA with low threshold. Diagnostic accuracy appears to be inadequate in tests with high thresholds (ELISA; IgG-specific CLIA), combination of IgG specificity and intermediate thresholds (ELISA, CLIA), high-dose heparin confirmation step (ELISA), and particle immunofiltration assay. When making treatment decisions, clinicians should be a aware of diagnostic characteristics of the tests used and it is recommended they estimate posttest probabilities according to likelihood ratios as well as pretest probabilities using clinical scoring tools.

Bivariate meta-analysis of predictive values of diagnostic tests can be an alternative to bivariate meta-analysis of sensitivity and specificity

Meta-analysis of predictive values is usually discouraged because these values are directly affected by disease prevalence, but sensitivity and specificity sometimes show substantial heterogeneity as well. We propose a bivariate random-effects logitnormal model for the meta-analysis of the positive predictive value (PPV) and negative predictive value (NPV) of diagnostic tests.