Make / diagnostic research
Preparing a diagnostic-accuracy study report
A diagnostic-accuracy report should show which clinical question was asked, who was tested, how the comparison was made and what the estimates leave uncertain. Clear reporting enables assessment; it does not certify a test or prove patient benefit.
Define the question and the role of the test
Name the target condition, the test being evaluated and its proposed role in the care pathway. Is it intended to replace an existing test, help triage people to another test or add information after earlier testing? Describe the setting and intended users. STARD asks for the scientific and clinical background, intended use, objectives and hypotheses. Keep a study of diagnostic accuracy distinct from a study of what happens to patients when a testing strategy is used.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / STARD 2015: explanation and elaboration
Choose reporting guidance for the actual study
EQUATOR lists STARD 2015 for diagnostic-accuracy reports and relevant extensions. Its current resources include STARD-AI, published in 2025, for studies using artificial intelligence. Check the relevant version and extension rather than treating every diagnostic study as interchangeable. A checklist is a way to expose necessary information, not a substitute for a protocol, ethical review or appropriate statistical methods. This guide is an introduction to general reporting, not a complete AI-reporting checklist.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / EQUATOR: STARD versions and extensions
Make recruitment and participant flow visible
Describe eligibility, how potentially eligible people were identified, where and when they were recruited, and whether recruitment was consecutive, random or based on convenience. Report the flow from eligibility through the index test and reference standard to the analysis, with reasons for exclusions. A group assembled from obvious cases and healthy controls may not represent the difficult cases in the proposed clinical setting. Do not let missing tests or unavailable verification disappear between the methods and final denominators.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / STARD 2015: explanation and elaboration / FDA: reporting diagnostic test performance
Describe the test and reference standard separately
Give enough detail about both methods for a reader to understand how they were performed and interpreted. Explain why the reference standard was chosen and what information was available to the people interpreting each test. State positivity cut-offs and distinguish prespecified choices from exploratory ones. FDA explains that agreement with a comparator that is not a reference standard is not the same as establishing sensitivity and specificity. An analysis of only discrepant results must not be presented as though it removes every reference-standard problem.
Evidence: STARD 2015: explanation and elaboration / FDA: reporting diagnostic test performance
Report counts, uncertainty and incomplete findings
For binary results, present the cross-tabulation of test and reference-standard results so readers can see the denominators. Report the accuracy estimates with their uncertainty, and explain how indeterminate, failed and missing results were handled. State the intended sample size and how it was determined. Do not select only favorable thresholds or turn an exploratory analysis into an apparently prespecified one. Describe variability analyses and limitations so that a percentage can be interpreted in the population and procedure that produced it.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / STARD 2015: explanation and elaboration / FDA: reporting diagnostic test performance
Keep reporting, appraisal and ethics distinct
STARD supports complete reporting. QUADAS-3 is a different tool: the current Bristol resources recommend it for systematic reviewers assessing risk of bias and applicability of primary test-accuracy evidence, with judgments at the selected-estimate level. Filling a STARD checklist does not produce a QUADAS judgment or establish patient benefit. Separately, ICMJE asks authors to report appropriate independent ethics review and protect identifiable participant information. Do not publish names, images or records merely because they make the methods easier to illustrate.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / University of Bristol: QUADAS-3 / ICMJE: protection of research participants
Close with claims the study can support
Explain the study limits and how the results fit the intended use. Accuracy evidence alone does not establish clinical benefit, cost-effectiveness, regulatory authorization or safety of an entire pathway. Check that the abstract matches the full report, and disclose protocol access, registration where applicable, support and the role of funders. This is guidance for preparing a checkable report, not proof that Mynd has conducted a study, assessed a device or opened a research or manuscript-review service.
Evidence: STARD 2015: essential diagnostic-accuracy reporting items / STARD 2015: explanation and elaboration
Source note
This is general health-science education. No clinical review has been performed.
General reporting education, not a protocol, an ethics or regulatory decision, a risk-of-bias assessment, a complete STARD-AI checklist or a Mynd study/submission service. No clinical review has been performed.