Fifty-seven studies were included (number of patients not reported).
Many studies suffered from selective patient sampling, suboptimal interpretation of the results, incomplete verification methods, and poor description of the study population and reference test.
Parametrical invasion (MRI, n=52; CT, n=9).
Both the sensitivity and specificity were heterogeneous for MRI, and there were fewer studies for CT. The pooled sensitivity was 74% (95% confidence interval, CI: 68, 79) for MRI and 55% (95% CI: 44, 66) for CT. The difference in sensitivity between MRI and CT was statistically significant (P=0.0027). The specificities of the two techniques were comparable.
Lymph node involvement (MRI, n=25; CT, n=17).
Heterogeneity was mainly observed in the sensitivity for both CT and MRI. The pooled sensitivity was 60% (95% CI: 52, 68) for MRI and 43% (95% CI: 37, 57) for CT. This difference was statistically significant (P=0.047). The specificities of the two techniques were comparable.
Bladder invasion (MRI, n=16; CT, n=3).
The pooled sensitivity was 75% (95% CI: 66, 83) for MRI and 64% (95% CI: 39, 82) for CT. This difference was not statistically significant. The pooled specificity was 91% (95% CI: 83, 95) for MRI and 73% (95% CI: 52, 87) for CT. This difference was statistically significant (P=0.032).
Rectum invasion (MRI, n=9; CT, n=2).
The pooled sensitivity was 71% (95% CI: 53, 83) for MRI and 45% (95% CI: 20, 73) for CT. This difference was not statistically significant. The specificities of the two techniques were comparable.
Investigation of heterogeneity.
The study results were plotted in receiver operating characteristic curve space, which gives a visual assessment of heterogeneity between studies. There were only sufficient data to carry out covariate adjustment and subgroup analysis for MRI datasets obtained for parametrial invasion and lymph node involvement. Population size, publication period and shortcomings in methodology had no influence on the estimates of sensitivity and specificity. No differences were observed in subgroups comparing MRI techniques.