PRISMA-DTA

What it is

PRISMA-DTA (Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies) is the official PRISMA extension for systematic reviews of diagnostic test accuracy, published as the McInnes et al. statement (JAMA, 2018) with a companion explanation-and-elaboration paper (Salameh et al., BMJ, 2020). It is maintained under the EQUATOR Network and the PRISMA family alongside the parent PRISMA 2020 statement. The checklist contains 27 items adapted from PRISMA to the structure of accuracy reviews, plus a separate PRISMA-DTA for Abstracts (PRISMA-DTA-A) extension of 11 items for abstract reporting. Its purpose is reporting transparency, not study conduct: it tells authors, peer reviewers, editors, and HTA assessors what facts a DTA review must disclose so that a reader can judge how sensitivity, specificity, and other accuracy estimates were derived and how trustworthy they are. The defining feature versus generic PRISMA is the two-test structure of every accuracy question: an index test compared against a reference standard ("gold standard"), with paired estimates (sensitivity/specificity, or PPV/NPV at a defined prevalence), threshold effects, and the hierarchical/bivariate meta-analytic models those data require.

When to use

Use PRISMA-DTA whenever the report is a systematic review or meta-analysis whose primary objective is the accuracy of one or more tests — diagnostic, screening, staging, prognostic classification, or triage — against a reference standard. This is the correct extension for journal submission of a DTA review, for the evidence-synthesis section of an HTA/payer dossier on a diagnostic or companion-diagnostic technology (e.g., NICE Diagnostics Assessment Programme), and for the diagnostic-evidence component of an FDA or EMA submission that relies on a synthesis of accuracy studies. Decision rules for picking PRISMA-DTA over a sibling guideline: (1) if the review estimates test accuracy (sensitivity/specificity, ROC/AUC, likelihood ratios) → PRISMA-DTA; (2) if the review estimates intervention or exposure effects (risk ratios, hazard ratios) → parent PRISMA 2020, with RECORD/RECORD-PE conventions for the underlying routinely-collected-data studies; (3) if you are writing the protocol for the review rather than the completed review → PRISMA-P; (4) for the primary DTA study being included, the relevant reporting guideline is STARD 2015, not PRISMA-DTA, and the within-study risk of bias is appraised with QUADAS-2. PRISMA-DTA governs the review layer; STARD governs the primary-study layer; QUADAS-2 is the appraisal instrument the review must apply and report.

What it requires

The checklist enforces complete reporting of the substantive domains that make an accuracy synthesis interpretable: a structured title and abstract identifying it as a DTA review; an explicit review question framed by population, index test(s), comparator test(s) where relevant, reference standard(s), and target condition; pre-registration (e.g., PROSPERO) and any protocol deviations; eligibility criteria including the definition of the target condition and the reference standard used to confirm it; a reproducible search strategy; data items capturing test thresholds, reference-standard definition, and 2×2 counts (TP/FP/FN/TN); risk-of-bias and applicability assessment using QUADAS-2 across the patient-selection, index-test, reference-standard, and flow-and-timing domains; the synthesis methods (bivariate/HSROC hierarchical models, handling of threshold variability and partial/differential verification); results presented as paired accuracy estimates with confidence/credible regions and SROC plots; investigation of heterogeneity and publication/selective-reporting bias; and a discussion of applicability to the intended clinical pathway and prevalence setting. For real-world-data DTA reviews — where the "index test" is often a claims- or EHR-derived phenotype or computable algorithm benchmarked against chart review or a validated reference — these same items map onto data-fitness-for-use, algorithm/phenotype definition and validation, verification completeness (the analogue of attrition), and sensitivity analyses around the case definition.

When NOT to use — limitations and common misapplications

PRISMA-DTA is a reporting checklist: it certifies that information was disclosed, not that the review is unbiased or high quality. Completing all 27 items does not make a flawed synthesis valid, does not substitute for QUADAS-2 appraisal of the included studies, and is not a quality score — there is no total to tally, and ticking items must never be treated as a methodological pass/fail. Specific failure modes: (1) applying PRISMA-DTA to a review of intervention effects or an etiologic question — use parent PRISMA 2020 instead; (2) using generic PRISMA for an accuracy review, which omits the index/reference structure, threshold reporting, and hierarchical-model items reviewers will demand; (3) confusing it with STARD (a primary-study DTA guideline) or with QUADAS-2 (the risk-of-bias instrument); (4) "checklist-as-theater," where authors cite PRISMA-DTA and append a checklist but report no 2×2 data, no reference-standard definition, and no risk-of-bias assessment, leaving the accuracy estimates un-auditable; (5) in RWD validation reviews, reporting algorithm PPV alone without the verification scheme and spectrum, which makes pooled accuracy uninterpretable. The checklist also does not address clinical-utility or decision-impact questions, which need different frameworks.

How it maps to this catalog

PRISMA-DTA is the reporting wrapper around DTA-relevant concepts in this repo. The accuracy-estimation core is implemented by diagnostic-accuracy (sensitivity, specificity, ROC/AUC, bivariate/HSROC synthesis) and, for routinely-collected-data benchmarking, by claims-outcome-algorithm-ppv-sensitivity-rwe and diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe (defining and validating the computable case definition that plays the role of the "index test"). The reference-standard and target-condition definition items connect to procedure-identification-and-measurement-in-claims-ehr. Verification completeness and differential loss of patients between index and reference standard map to attrition-and-loss-to-follow-up-rwe and database-feasibility-attrition-funnel-rwe. Estimand clarity for what the accuracy contrast actually targets connects to estimands-ate-att-intercurrent-events-rwe and estimand-analysis-traceability-rwe; the evidence-synthesis machinery aligns with systematic-review, meta-analysis-obs, and network-meta-analysis (for comparative/network DTA). Cross-database accuracy differences are surfaced by medicare-ffs-ma-commercial-claims-differences-rwe. Applied note for claims/EHR/registry RWE: when a DTA "review" is in fact a synthesis of algorithm-validation studies, report each study's reference standard (chart review vs adjudication), the verification fraction and how unverified records were handled, the operating threshold (e.g., 1 inpatient OR 2 outpatient codes within a time window), and prevalence-dependent measures (PPV/NPV) separately from prevalence-independent ones (sensitivity/specificity) — exactly the items PRISMA-DTA forces into the open.