Forty-four studies met the inclusion criteria. All included studies used an appropriate reference standard, avoided differential and partial verification bias, and avoided incorporation bias. Of the 44 studies, 39 recruited a representative patient spectrum, 41 blinded interpreters of the index test, 28 blinded interpreters of the reference standard, 28 reported on uninterpretable results, 22 reported on withdrawals and 28 blinded interpreters of scans to clinical data. None of the studies provided clinical data that would be available in clinical practice.
Compared with coronary angiography MRI (28 studies; 2,970 patients; range 33 to 823) had a sensitivity of 91% (95% CI 88 to 93) and specificity of 80% (95% CI 76 to 83), perfusion ECHO (10 studies; 795 patients; range 35 to 132) had a sensitivity of 87% (95% CI 81 to 91) and specificity of 72% (95% CI 56 to 83), and SPECT (13 studies; 1,323 patients; range 30 to 357) a sensitivity of 83% (95% CI 73 to 89) and specificity of 77% (95% CI 64 to 86). The sensitivity of MRI was significantly higher than that of SPECT.
The positive likelihood ratio was 4.43 (95 % CI 3.64 to 5.23) for MRI, 3.08 (95 % CI 1.65 to 4.50) for ECHO and 3.56 (95 % CI 2.07 to 5.04) for SPECT. The negative likelihood ratio was 0.12 (95 % CI 0.08 to 0.15) for MRI, 0.18 (95 % CI 0.13 to 0.24) for ECHO, and 0.22 (95 % CI 0.14 to 0.31) for SPECT.
Results for a range of sensitivity analyses were reported. The funnel plots for MRI and SPECT suggested publication bias, but not for ECHO.