Comprehensive overview of methods and reporting of meta-analyses of test accuracy

2012 Auckland

Dahabreh IJ¹, Chung M¹, Kitsios GD¹, Terasawa T², Raman G¹, Tatsioni A³, Tobar A⁴, Lau J¹, Trikalinos TA⁵, Schmid CH⁶

¹Center for Clinical Evidence Synthesis, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Boston, MA

²Department of Internal Medicine, Fujita Health University Nanakuri Sanatorium, Fujita Health University School of Medicine, Tsu, Mie, Japan

³Department of Internal Medicine, University of Ioannina School of Medicine, Ioannina, Greece

⁴Division of Gastroenterology & Hepatology, Department of Medicine, University of Colorado, Aurora, CO

⁵Center for Evidence-based Medicine, Program in Public Health, Brown University, Providence, RI

⁶Biostatistics Research Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Boston, MA

Background:

Meta-analyses of test accuracy are increasingly performed and methods are continually evolving.

Objectives:

To survey the methods and reporting practices in meta-analyses of test accuracy.

Methods:

We searched MEDLINE (1966–2009), reference lists and relevant reviews for meta-analyses of test accuracy for diagnostic or predictive medical tests. Nine reviewers extracted information on clinical topics, literature review methods, quality assessment and statistical analyses performed.

Results:

We reviewed 760 meta-analyses of test accuracy, published 1987–2009. Eligible reviews averaged 18 primary studies and typically examined one index test against one reference standard. Most common clinical areas were cardiovascular disease (21%) and oncology (25%); most common test categories were imaging (44%) and biomarker tests (28%). 62% used at least two electronic databases to identify eligible studies; the number has increased over time. Assessment of verification bias, spectrum bias, blinding, prospective study design, and consecutive patient recruitment also became significantly more common over time (P < 0.001, comparing reviews published through 2004 vs. 2005 onwards). Improvements coincided with increasing use of quality-item checklists to guide assessment of methodological quality. Sensitivity (77%), specificity (74%) and odds ratios (34%) were the most commonly used metrics. Heterogeneity tests were used in 58% of meta-analyses and subgroup or regression analyses were used in 57%. Random effects models were employed in 57% of meta-analyses and increasingly over time (38% through 2004 vs. 72% 2004-onwards; P < 0.001). Use of bivariate models of sensitivity and specificity also increased in recent years (21% of reviews published in 2008–2009 vs. 5% in earlier years; P < 0.001).

Conclusions:

Methods employed in meta-analyses of test accuracy have improved with the introduction of quality assessment checklists and the development of more sophisticated statistical methods. However, multivariate methods for meta-analysis and reporting of study design characteristics are still lacking in a substantial number of diagnostic test meta-analyses.