Pooled estimates of the overall sensitivity and specificity, with 95% CIs, for any foetal fibronectin assay were determined from a summary receiver operating characteristic (SROC) curve, fitted using a bivariate random-effects model. Pooled estimates included data for both singleton and multiple pregnancies, except one study for which separate data were provided; only data on singleton pregnancies from this study were included in the analysis. Post-test probabilities were calculated from pooled estimates of likelihood ratios and pre-test odds of delivery within seven days of testing (7.7%).
Between-study heterogeneity was assessed using Cochrane Q and I2 tests.
Subgroup analyses were conducted for different types of foetal fibronectin test and cut-off values.
The effect of study characteristics (multifoetal gestation, method of testing (ELISA versus others), blinding of test results, high quality studies vs. others, prevalence of delivery within seven days, USA studies versus others, studies published before 2002 versus others and language of publication) on test accuracy was assessed using a series of univariate regression models. Data were considered insufficient to support multivariate regression modelling.
Publication bias was assessed using a funnel plot.