Thirty-five studies were included in the meta-analysis (n=2,456 participants, range 13 to 229). Thirty-two studies recruited a representative patient sample and 27 avoided disease progression bias. All studies avoided partial and differential verification bias. Twenty-eight studies blinded interpreters of the index test and reference standard to the results of the other test. Most studies reported uninterpretable results and withdrawals.

Overall per-patient analysis (26 studies, 2,125 patients) showed pooled sensitivity of 89% (95% CI 88% to 91%), specificity of 80% (95% CI 78% to 83%), positive likelihood ratio of 4.18 (95% CI 3.31 to 5.27), negative likelihood ratio of 0.15 (95% CI 0.11 to 0.20) and diagnostic odds ratio of 33.65 (95% CI 22.09 to 51.27). The area under the curve was 0.92. Significant statistical heterogeneity was observed for all analyses.

Similar results were seen when studies that used adenosine as the stressor were analysed separately (20 studies, 1,658 patients). The positive likelihood ratio and diagnostic odds ratio were lower when studies that used dipyridamole as the stressor were analysed separately, but these were far fewer in number (five studies, 417 patients).

Results for 3.0T (four studies, 282 patients) and 1.5T (23 studies, 1,904 patients) technologies were similar to the overall results.

Overall prevalence of coronary artery disease was 57% (1,205 of 2,125 patients). Results for per-territory and per-artery analyses were presented.