Authors of diagnostic test accuracy systematic reviews need to define the kind of meta-analysis to perform. Questions to consider are:
Should we estimate summary sensitivity and specificity?
Should we compute a summary ROC curve?
The answer to these questions lies in the kind of data available and more exactly whether the diagnostic threshold is the same across the selected primary studies. Sometimes retrieved studies may rigorously use the same diagnostic threshold, but, on other occasions variations in the threshold may exist. This is often the case when there is no explicit numerical cut-off-point or when the index test is based on an observer’s judgement.
The basic strategy is as follows:
When the same threshold is used through the primary studies, then:
estimate the summary sensitivity/specificity.
When different thresholds are used, then:
produce a SROC curve; and
estimate the summary sensitivity/specificity for each threshold provided in the articles.
If a study has referred to sensitivity/specificity values for several threshold, it can contribute to several estimations of summary sensitivity/specificity.