Diagnostic accuracy of circulating blood biomarker families in acute ischemic stroke: a secondary grouped meta-analysis with primary-study verification

Abstract

Circulating biomarkers may complement clinical assessment and neuroimaging during the early evaluation of suspected acute ischemic stroke (AIS), but the evidence spans biologically and analytically heterogeneous test families. We revised the original analysis as a transparent secondary grouped diagnostic accuracy meta-analysis and verified all contributing primary reports. Candidate studies were obtained from two recent systematic reviews of acute-window protein and cell-free nucleic-acid biomarkers, backward citation checking, and a targeted PubMed/MEDLINE update through 31 July 2026. Primary reports were verified against predefined eligibility criteria. Biomarkers were reclassified as extracellular chromatin/cell-free DNA, hypoxia-response, non-coding RNA, or protein/multimarker tests. Where published 2 × 2 tables were unavailable, cells were reconstructed from sensitivity, specificity, and diseased/non-diseased sample sizes using prespecified rounding rules. A bivariate random-effects model was used only when at least four independent cohorts were available; families with fewer cohorts were summarized descriptively. QUADAS-2, comparator-spectrum analyses, one-estimate-per-cohort rules, and rounding-bound sensitivity analyses were added. Seventeen verified primary reports published from 2003 to 2025 contributed 24 original diagnostic test estimates. The cohort-level primary dataset contained 17 independent cohorts. Extracellular chromatin/cell-free DNA tests (four cohorts) yielded pooled sensitivity 0.712 (95% CI 0.454–0.880), specificity 0.796 (0.699–0.868), LR + 3.50 (2.27–5.40), LR − 0.36 (0.17–0.76), and DOR 9.68 (3.39–27.62). Non-coding RNA tests (10 cohorts) yielded sensitivity 0.866 (0.829–0.896) and specificity 0.922 (0.585–0.990), but the wide specificity and DOR intervals indicated substantial uncertainty. Hypoxia-response and protein/panel evidence comprised only one and two independent cohorts, respectively, and was not formally pooled. Most studies used healthy-control two-gate designs and data-driven thresholds; all diagnostic 2 × 2 tables required reconstruction. Comparator-, rounding-, dependence-, and leave-one-cohort-out analyses preserved the direction of the main findings but demonstrated substantial uncertainty and spectrum effects. Non-coding RNA and extracellular chromatin/cell-free DNA tests show promising but methodologically fragile diagnostic signals. The evidence is not sufficient for routine clinical implementation because realistic stroke-mimic cohorts are sparse, thresholds are usually data-driven, and reconstructed cells and within-cohort multiple testing remain important limitations. Prospective one-gate, mimic-enriched validation is required.