Eleven studies met the inclusion criteria (n= 4,387, range 157 to 1,080). The post hoc assessment using QUADAS showed all studies recruiting a representative patient sample; all but one study clearly reported selection criteria. Eight studies reported the analysis of death being blinded to the index test results. None of the studies reported on uninterpretable results or withdrawals. Given the prognostic nature of the studies and the use of mortality as the reference standard, verification and incorporation bias were avoided, and progression bias was not a relevant criteria.
The summary negative likelihood ratio was 0.62 (95% CI: 0.55 to 0.69) and diagnostic odds ratio was 3.03 (95% CI: 2.44 to 3.76) for non-sustained ventricular tachycardia as a predictor of sudden cardiac death was; there was no statistically significant heterogeneity for either analysis.
The summary positive likelihood ratio was 1.86 (95% CI: 1.56 to 2.21); statistically significant heterogeneity was observed.
There was evidence of a threshold effect, with a Spearman's correlation coefficient between sensitivity and specificity of -0.818 (p=0.002); the area under the curve was 0.675 (standard error 0.0156). Meta-regression showed no significant relationship between mean left ventricular ejection fraction and log diagnostic odds ratio.
There was no evidence of publication bias.