Ask Mallard

Diagnostic accuracy study

A diagnostic accuracy study evaluates how well an index test identifies a target condition compared with an appropriate reference standard.
NHMRC EVIDENCE LEVEL · 2009Depends on how the study is designed · Diagnostic accuracyProvisional design classification. Study quality and certainty of findings require separate appraisal.
Why this level?Specify the reference standard. Accuracy cannot be classified from diagnostic yield alone. NHMRC 2009 source

Diagnostic accuracy needs its own hierarchy. The level depends on the reference standard, independence, blinding, clinical presentation and recruitment. A design title alone does not establish those features.

Where the NHMRC 2009 hierarchy names this design, it places it here:

Research questionLevelWhen
Diagnostic accuracyIIConsecutive patients with a defined clinical presentation undergo an independent blinded comparison against a valid reference standard.
Diagnostic accuracyIII-1Nonconsecutive patients with a defined clinical presentation undergo an independent blinded comparison against a valid reference standard.
Diagnostic accuracyIII-2The reference-standard comparison does not meet all criteria for Levels II or III-1.
Diagnostic accuracyIII-3Known cases and a separate healthy control group are assembled for the diagnostic comparison; the clinical spectrum may be unrepresentative.
Diagnostic accuracyIVDiagnostic yield is measured without a reference standard to confirm accuracy.

What question does this help answer?

“How accurately does this test classify people as having or not having the target condition?”

When to use it

Use this design when evaluating a diagnostic test, screening tool, algorithm, imaging method, laboratory assay, or other classifier against a credible reference standard.

When not to use it

Do not frame a study as diagnostic accuracy when there is no meaningful target condition or reference standard, or when the main objective is prognosis rather than diagnosis.

How it works in plain language

Participants receive the index test and the reference-standard assessment. Results are compared to estimate measures such as sensitivity, specificity, predictive values, likelihood ratios, or discrimination. Threshold choice and disease prevalence affect interpretation.

What the study's flow looks like

A diagnostic accuracy study. Consecutive eligible people all receive both the index test and the reference standard, and the results form a two-by-two table. A dashed path shows partial verification, where only people with a positive index test get the reference standard, which distorts the estimates.Diagnostic accuracy: everyone gets both testsConsecutive eligible peopleIndex testReference standardTwo-by-two table: index test result against reference standard resultPartial verification:reference only ifpositiveSampling obvious cases and healthy volunteers instead of consecutive eligible people exaggerates apparentaccuracy.
Conceptual diagram. It shows the structure of the idea, not data from any study.

Why Mallard may recommend it

Mallard may recommend this design when the research question asks whether a clinical test or algorithm correctly identifies a current disease or condition.

Key assumptions and design requirements

  • The participant spectrum reflects the intended use population.
  • The reference standard is appropriate.
  • Index-test interpretation is blinded to reference-standard results when feasible, and vice versa.
  • Verification is not selectively applied based on the index result.
  • Thresholds are prespecified when confirmatory accuracy is the objective.

Intervals for sensitivity and specificity

Sensitivity is a proportion among the people with the condition and specificity a proportion among those without, so each is reported with an interval for a single proportion: a Wilson or Clopper-Pearson interval, not the simple normal approximation, which behaves badly when accuracy is close to 100 percent and can give an upper limit above it. The number of people with the condition, not the total enrolled, governs how precise sensitivity is, which is why a study with few cases can report a high sensitivity with a wide interval. Predictive values are single proportions too, but they hold only at the prevalence of the study sample.

What to watch out for

Case-control sampling of obvious diseased and healthy participants can exaggerate apparent accuracy. Predictive values depend on prevalence. Accuracy alone does not establish clinical utility.

Clinical example

Evaluate whether a new EHR-based screening score identifies current sepsis compared with an adjudicated clinical reference standard.

Code for this method
Loading…
Recent studies that used this
Searching PubMed…
Work it out
Related Learn pages
Sources & further reading
  1. STARD 2015: An Updated List of Essential Items for Reporting Diagnostic Accuracy Studies (opens in a new tab) · EQUATOR Network
  2. STARD 2015 guidelines for reporting diagnostic accuracy studies: explanation and elaboration (opens in a new tab) · BMJ Open · doi 10.1136/bmjopen-2016-012799
  3. Two-sided confidence intervals for the single proportion: comparison of seven methods (opens in a new tab) · Stat Med · doi 10.1002/(sici)1097-0258(19980430)17:8<857::aid-sim777>3.0.co;2-e
STARD is a reporting guideline: it makes a diagnostic accuracy study transparent to read, and it is not a recipe for designing a valid one. The explanation and elaboration paper is the one to read for why each item matters.
Last source review: 2026-09-05
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.