Meta-analysis topic ideas in diagnostic accuracy and imaging

A diagnostic test accuracy review asks a different question from a treatment review. Instead of one effect measure there are two, sensitivity and specificity, and they move in opposite directions as the threshold changes, so the review has to model them together. The rules for a good topic differ accordingly.

New tests, imaging methods and artificial intelligence tools appear faster than they are evaluated, so there are often accuracy studies with no review. The challenge is heterogeneity: patient mix, reference standard, threshold and device vary between studies far more than they do in treatment trials.

Find a diagnostic accuracy and imaging topic that has not been done

SynthGap searches the published literature, checks each candidate against existing systematic reviews, and proposes topics with the primary studies behind them. Start on the Free plan: no card needed, and every run shows every topic it finds.

Where gaps tend to hide in diagnostic accuracy research

Artificial intelligence and software tools

Algorithms for imaging, screening and triage are evaluated against different reference standards and on different datasets. Reviews that separate external validation from development data are informative.

Point-of-care and rapid tests

Ultrasound, rapid antigen tests and biomarker tests used outside the laboratory are studied in small accuracy studies with different case mixes.

New biomarkers and combinations

Blood-based and imaging biomarkers for cancer, dementia and cardiovascular disease are reported in single studies. Comparisons between tests or combinations are scarce.

Test comparison in the same population

Reviews that compare two tests directly, using only studies that applied both to the same patients, give stronger evidence than comparing results across different studies.

Screening and referral pathways

Accuracy in primary care differs from accuracy in hospital settings, and reviews that separate settings are useful.

Outcomes and effect measures that pool well in diagnostic accuracy reviews

Sensitivity and specificity: pooled with a bivariate or hierarchical summary ROC model that accounts for the correlation between them and for threshold differences, not by averaging each separately.
Positive and negative predictive values: not pooled directly, because they depend on prevalence. Calculate them for stated prevalences.
Likelihood ratios and diagnostic odds ratios: derived from the pooled sensitivity and specificity or modelled as a summary, and interpreted with care.
Area under the curve: only as a supplementary summary, since it hides the thresholds at which the test works.
Impact outcomes (change in management, time to diagnosis): usually narrative, since few studies measure them.

Common traps in diagnostic accuracy reviews

Case-control sampling (healthy controls and clear disease), which inflates accuracy compared with a real clinical population.
Using different reference standards, or a reference standard that depends on the test.
Excluding uninterpretable results from the analysis, which overstates accuracy.
Pooling studies with different thresholds as if they were one estimate.
Testing a tool on the same data it was developed on.

Check before you commit

Before committing months to a topic, check that nobody has beaten you to it. Search PubMed with its diagnostic accuracy and systematic review filters, the Cochrane Library (which has its own series of diagnostic test accuracy reviews), PROSPERO for registered reviews, and Epistemonikos. Then check whether the tests have changed since the last review, because a new version of an assay or a new algorithm can make an old review obsolete. Step-by-step guide.

Starting points to try in SynthGap

These are areas to explore, not claims that they are open. Some may already be well covered, and finding that out quickly is the point.

Point-of-care ultrasound for a specific diagnosis in a specific setting, compared with the usual reference test
Artificial intelligence for detecting a specific cancer on imaging: external validation studies only
Blood biomarkers for the early detection of cognitive decline in primary care

Questions

Which risk of bias tool should I use?

QUADAS-2 is the standard for primary diagnostic accuracy studies, and QUADAS-C extends it to comparisons between tests. For prediction models, which are a different design, PROBAST is used.

How do I check whether a diagnostic accuracy review already exists?

Search PubMed with the diagnostic accuracy and systematic review filters, the Cochrane Library, PROSPERO and Epistemonikos, and look at the date of the last search of the newest review.

How many studies do I need?

Bivariate models need enough studies to estimate the between-study variation in sensitivity and specificity. With fewer than about four or five they are unstable, and a narrative summary or a simpler approach may be safer.

Other specialties

This page is general methodological guidance, not medical advice, and it does not list specific unreviewed topics. SynthGap's suggestions are AI-generated from the published literature, so verify them before registering a protocol.