← allmeta

Workflow · PRISMA-DTA

Diagnostic test accuracy

DTA reviews are not pairwise reviews with sensitivity and specificity glued on. The threshold effect, joint Se-Sp distribution, and bivariate model are mandatory — Cochrane DTA Handbook is explicit. This is the path.

  1. Frame the question (PIRD, not PICO)

    For DTA, use Population–Index test–Reference standard–Diagnosis. Define: who is the population (screening, triage, or confirmation?), what is the index test, what is the reference standard, what condition.

    Mixing populations across phases of diagnosis is the most common DTA failure mode.

    PICO / PIRD Formulator → Learn the theory: DTA Course (When the test lies)

  2. Search & screen

    DTA-specific filters (Cochrane DTA Handbook §6) — broader than RCT filters because DTA papers report poorly. Then screen with PIRD eligibility.

    Standard PRISMA approach with DTA-tuned search; filters are looser to compensate for inconsistent indexing.

    Search Translator → PRISMA Screening →

  3. Risk of bias & applicability — QUADAS-2

    QUADAS-2 has 4 domains × 2 judgements (RoB and applicability). Get the patient selection and reference standard domains right or the rest is moot.

    Whiting PF et al. Ann Intern Med 2011;155:529 (QUADAS-2 — the only validated DTA RoB tool; RoB 2 does not apply).

    QUADAS-2 →

  4. Threshold-effect check

    Spearman correlation of logit(Se) and logit(1−Sp) across studies. r > 0.6 is a threshold effect — pooling Se and Sp separately becomes meaningless and the SROC curve is the right summary.

    If you skip this and pool Se/Sp anyway, your readers (and reviewers) catch you. Per advanced-stats.md.

    DTA SROC →

  5. Fit the bivariate / HSROC model

    Bivariate (Reitsma 2005) or HSROC (Rutter & Gatsonis 2001). Bivariate convergence often fails for k<5 — constrain ρ to [−0.95, 0.95] or fix ρ=0.

    Reitsma JB et al. J Clin Epidemiol 2005;58:982 (bivariate); Rutter CM, Gatsonis CA. Stat Med 2001;20:2865 (HSROC). A pooled (Se, Sp, AUC) point with confidence and prediction ellipses is the headline output.

    HSROC → DTA SROC → WebR (mada) →

  6. Certainty & report (PRISMA-DTA)

    Use the GRADE-DTA framework — rate certainty separately for sensitivity, specificity, and downstream patient-important outcomes (true positives, false negatives). CINeMA is for network meta-analysis, not DTA.

    Schünemann HJ et al. J Clin Epidemiol 2020;122:129-141 (GRADE-DTA primary peer-reviewed paper, with subsequent companion notes through 2021). Reviewers will check that certainty was rated for sensitivity, specificity, and clinical consequences separately.

    GRADE SoF → PRISMA flow →

New to DTA? The DTA Course (When the test lies) is the deepest free resource on the topic. For RCT reviews, see systematic review; for treatment networks, see NMA.