Results of diagnostic test accuracy systematic reviews can be graphically represented through two different major ways.
As for systematic reviews of effectiveness, forest plots can be performed. In the case of diagnostic test accuracy, two forest plots are presented side by side: one for sensitivity and the other for specificity. These graphs thus show the means and confidence intervals for sensitivity/specificity for each of the selected primary studies. These values are also listed in numerical form. Moreover, the number of true positives, false positives, true negatives and false negatives are also reported, as well as, where appropriate, any covariates (for instance the type of diagnostic test used). Figure 9.2 shows a paired forest plot made using mock data (Campbell et al. 2015).
Figure 9.2: An example paired forest plot generated using mock data in RevMan5
Numerical values for sensitivity and specificity are presented alongside graphical representations where the boxes mark the values and the horizontal lines show the confidence intervals.
It is also possible to create Summary ROC (SROC) curves. They are graphs with 1-specificity on the x-axis and sensitivity on the y-axis, in which each primary study contributes to a unique point defined by its sensitivity and specificity for a given threshold. If several thresholds are reported in a single primary study, only one sensitivity/specificity pair for that study can be plotted on the SROC graph. Point size may vary according to sample size. To indicate more precisely the precision of the estimates, point height is proportional to the number of diseased patients, while point width is proportional to the number of control patients.
Following a rigorous meta-analysis, a curve can be added in the graph. A Summary ROC curve represents the expected ROC curve at many different positivity threshold levels. If the same positivity threshold has been used in each of the primary studies, it is appropriate to calculate and plot the summary sensitivity and specificity, and their confidence region. A prediction region can also be provided, corresponding to the area where the true sensitivity/specificity of a future study should be found in 95% of the cases. Figure 9.3 shows a SROC curve from made using mock data in RevMan5 (Campbell et al. 2015).
Figure 9.3: An SROC curve generated using mock data in RevMan5. Sensitivity is shown on the y-axis, the x-axis shows inverted specificity (Campbell et al. 2015)