Diagnostic test accuracy reviews

Background: Systematic reviews of diagnostic test accuracy

The main steps of a Cochrane review of diagnostic test accuracy studies are the same as the main steps of a Cochrane review of the effects of interventions:

  1. Framing the question and definition of the objectives

  2. Search and selection of studies

  3. Quality assessment

  4. Data extraction

  5. Statistical analysis  

  6. Interpretation and presentation of results

However, there are some differences between intervention and diagnostic test accuracy research that impose special challenges to each of these six steps:

Many different designs are likely to have been used in diagnostic test accuracy studies, mostly non-randomized observational studies. There is no single, dominant design, such as the randomized trial for the effects of interventions, which is least prone to bias. This also complicates the indexing of diagnostic test accuracy studies in electronic bibliographic databases, such as MEDLINE.

The quality of the diagnostic test accuracy studies is likely to vary very widely, even among studies that have very similar designs. The relationship between study quality and study outcome is unclear, although several researchers have studied these relationships (Lijmer, Rutjes). Study quality is also difficult to assess because of the generally poor quality of the reporting of study methods.

The outcomes of diagnostic test accuracy research are very dependent on setting and patient population. Sensitivity, specificity, likelihood ratios and odds ratios depend very much on the characteristics of the people in whom the test is used, including any  prior testing. These data are also reported poorly in most studies.

There is likely to be significant heterogeneity across diagnostic test accuracy studies, with many sources of heterogeneity being unique to this type of research, notably the role of the positivity threshold and the presence of various biases.

The methodology for quality assessment, meta-analysis and investigation of heterogeneity in diagnostic test accuracy are underdeveloped compared to the methodology for intervention research.

Sensitivity analysis

It is not possible to conduct sensitivity analyses in RevMan by omitting individual diagnostic studies. These analyses need also to be done outside RevMan.

See also Assessment of methodological quality (QUADAS-2), Data and analyses in diagnostic reviews, Constructing data tables, Covariates, Entering data, Analysis graphs in diagnostic reviews