Core design
Diagnostic accuracy research evaluates how well an index test identifies a target condition relative to a reference standard. The study must define the intended clinical use, spectrum of participants, thresholds and verification procedures before analysis.
Approval and governance
Usually needs formal review / authorisation
- Prospective diagnostic testing or collection of extra samples/procedures generally requires ethics review and consent.
- Retrospective use of identifiable records/images/samples generally requires ethics/governance and data-access approval or authorised waiver.
- Use of stored biospecimens may have additional consent and biobank requirements.
May follow a lighter or different route
- Fully anonymised public image/dataset resources may follow a different route, subject to dataset licence and institutional policy.
Do not do this
- Do not choose the diagnostic threshold after examining the same data and then report performance as if it were prespecified.
- Do not verify only index-test-positive patients with the reference standard without addressing verification bias.
- Do not let assessors interpret the reference standard with knowledge of the index test when blinding is feasible.
Mediclinic Middle East publicly states that research projects carried out at MCME are to receive approval from its internal Research and Ethics Committee and applicable local regulatory authorities before initiation. The exact route varies by project, facility and emirate. Dubai projects may involve DSREC depending on applicability. This hub must therefore route users to the Research Office and current local forms rather than declaring a project “ethics exempt.” Institution-specific forms and contacts will be inserted after verification.
Step-by-step workflow
Define the intended clinical role
Is the test for screening, triage, replacement, add-on diagnosis or monitoring? Accuracy requirements depend on use.
Define the target condition and reference standard
Specify how the true disease state is established and what happens when the reference standard is imperfect or unavailable.
Define the participant spectrum
Recruit a clinically relevant range of patients in whom the test would actually be used. Avoid artificial case-control designs with very obvious diseased cases and healthy controls unless that is truly the intended use.
Define index-test procedures and thresholds
Specify equipment, operators, preprocessing, interpretation, cutoffs and whether thresholds are prespecified or exploratory.
Plan blinding and test order
State whether index-test readers know reference-standard results and vice versa, and whether clinical information is available.
Calculate sample size
Base size on expected sensitivity/specificity and desired precision, accounting for target-condition prevalence and unusable results.
Obtain approvals and standardise data collection
Submit protocol, consent, imaging/sample/data handling and any device/intervention requirements. Train assessors and document indeterminate results.
Build the 2×2 data and participant flow
For a binary threshold, count true positives, false positives, true negatives and false negatives; also report excluded, indeterminate and missing results.
Estimate accuracy with precision
Report sensitivity, specificity and confidence intervals. Depending on the question, include predictive values, likelihood ratios, ROC/AUC and calibration or decision metrics where appropriate. Remember predictive values depend on prevalence.
Address multiple thresholds and optimism
If thresholds are derived in the same dataset, distinguish development from validation and consider internal/external validation. Avoid reporting the best cutoff without acknowledging overfitting.
Report with STARD
Describe participant selection, index and reference tests, blinding, thresholds, sample-size rationale, flow, cross-tabulation, estimates with precision, indeterminate/missing results, adverse events if applicable, registration and protocol access where relevant.
Final checklist
- Clinical role of test defined.
- Reference standard justified.
- Participant spectrum representative of intended use.
- Threshold prespecified or clearly exploratory.
- Blinding described.
- Sample size/precision justified.
- Indeterminate and missing results reported.
- Sensitivity/specificity include confidence intervals.
- STARD checklist completed.
Analysis details to plan before data collection
- Predefine how indeterminate index-test and reference-standard results will be handled. Excluding them can make the test look better than it is.
- For continuous tests, distinguish a prespecified clinical threshold from exploratory threshold finding. If deriving a cutoff, validate it in independent data when possible.
- Report the 2×2 counts so readers can reproduce sensitivity and specificity.
- Use confidence intervals for all key accuracy estimates and state the method when relevant.
- If comparing two tests, account for paired testing when the same participants receive both tests.
- When the reference standard is imperfect, discuss incorporation bias, differential verification and misclassification explicitly.
- If an AI/ML model produces the index test, separate model development from diagnostic-accuracy validation and use the relevant AI/prediction reporting guidance in addition to STARD where applicable.
Interpretation
Accuracy is not the same as clinical utility. A highly accurate test may still provide little benefit if it does not change management, is unavailable at the point of care, creates harmful downstream testing or is evaluated in a population unlike the intended users. State the intended clinical pathway and avoid turning sensitivity/specificity into a treatment recommendation.