Model fit | ||||

logLik | AIC | BIC | ||

−432.8 | 871.6 | 878.3 | ||

Random effects | ||||

Factor | σ^{2} | σ | Cochran’s Q Test for heterogeneity | |

P value | ||||

Study | 167.8* | 13.0* | <0.001 | |

Fixed effects | ||||

Factor | Estimate | P value | Lower 95% CI | Upper 95% CI |

(intercept) | 91.7 | <0.001 | 84.5 | 98.9 |

Time to reperfusion, hours | −13.1 | <0.001 | −14.6 | −11.5 |

*Exploration of heterogeneity: Separate inclusion of the timing of MRI (χ

^{2}(1)=11.5, p<0.001), T2-weighted MRI interpretation method (χ^{2}(2)=7.8, p=0.020), T1-weighted late gadolinium enhancement MRI interpretation method (χ^{2}(5)=15.3, p=0.009) and gender distribution (χ^{2}(1)=11.3, p<0.001) resulted in a significantly better fit of the mixed effects model, while the inclusion of the T2-weighted MRI sequence (χ^{2}(3)=5.6, p=0.131), T1-weighted late gadolinium enhancement MRI sequence (χ^{2}(1)=0.5, p=0.468) and mean age (χ^{2}(1)<0.1, p=0.995) did not. Inclusion of all significant confounding variables reduced the between-study heterogeneity by 45.2% to σ^{2}=91.9 (see online supplemental table 4).AIC, Akaike information criterion; BIC, Bayesian information criterion; logLik, log-likelihood.