← Back
NeuroTrials.ai
Neurology Clinical Trial Database

CATCH-AF Score

Atrial fibrillation detection after embolic stroke of undetermined source: Development and validation of the CATCH-AF score

Year of Publication: 2026

Authors: D'Anna L, Favruzzo F, Baracchini C, et al.

Journal: International Journal of Stroke

Citation: D'Anna L, et al. Int J Stroke. 2026. doi:10.1177/17474930261428118.

Link: https://doi.org/10.1177/17474930261428118


Clinical Question

Can a simple clinical score predict which ESUS patients will develop atrial fibrillation on long-term implantable cardiac monitoring?

Bottom Line

In 543 ESUS patients monitored with ICMs, the CATCH-AF score (range 0-9) based on age, coronary artery disease, heart failure, and prior TIA/stroke stratified AF risk with excellent discrimination (AUC 0.85) that remained stable over 4.7 years. High-risk patients (≥5 points, 17.1% of the cohort) had a 19-fold higher hazard of AF detection than low-risk patients and lost 918 more AF-free days. The score supports targeted and cost-effective rhythm monitoring after ESUS but requires external validation before widespread clinical adoption.

Major Points

  • First prediction score specifically designed for AF detection in ESUS patients with long-term ICM monitoring. Addresses a key clinical gap: which ESUS patients benefit most from prolonged cardiac monitoring.
  • 543 consecutive ESUS patients across 7 stroke centres (UK and Italy, 2021-2024), with 1558.5 patient-years of ICM follow-up (median 1104 days, IQR 695-1454.5). 118 patients (22%) developed new AF during follow-up.
  • CATCH-AF components derived via LASSO-penalized Cox regression. Age: <60 years=0, 60-69=1, 70-79=2, ≥80=3; Coronary artery disease=2, Previous TIA/Stroke=2, Congestive Heart Failure=2. Total range 0-9.
  • Discrimination: AUC 0.85 (95% CI 0.82-0.89) overall. Time-dependent AUC values actually INCREASED over time: 0.849 at 6 months, 0.844 at 12 months, 0.867 at 24 months, 0.872 at 36 months, 0.858 at 1710 days (~4.7 years) — performance did not decline.
  • Risk stratification: Low risk (0-2 points) — 226 patients (41.6%), <10% AF at 1710 days; Medium risk (3-4 points) — 224 patients (41.3%), ~20% AF plateau at 3 years, HR 6.1 (95% CI 3.0-12.4, p<0.001) vs low risk; High risk (≥5 points) — 93 patients (17.1%), approaching 60% AF at long-term follow-up, HR 19.2 (95% CI 9.4-39.4, p<0.001) vs low risk.
  • RMST analyses quantified absolute disease burden: high-risk patients lost 918 fewer AF-free days vs low-risk; 614 fewer days vs medium-risk; medium-risk lost 305 fewer AF-free days vs low-risk (all p<0.001).
  • Internal validation via 10-fold cross-validation and internal-external cross-validation across the 7 participating centres demonstrated consistent discrimination and calibration across geographically distinct sites.
  • Key clinical insight: LASSO regression shrank CHA2DS2-VASc, hypertension, diabetes, and sex to zero — they were NOT independent predictors once age, coronary artery disease, heart failure, and prior stroke were in the model.
  • Methodological innovation vs prior scores (HAVOC, STAF, Brown ESUS-AF, AF-ESUS, ACTEL, Decryptoring): time-dependent ROC curves and restricted mean survival time (RMST) analysis provide longitudinal predictive performance and absolute time-based quantification of risk, respectively.
  • ICM device used: Reveal LINQ (Medtronic Inc, Minneapolis, MN, USA). Study conducted per STROBE guidelines.

Design

Study Type: Retrospective, multicentre, observational cohort

Randomization:

Blinding: Not applicable (observational)

Enrollment Period: 2021-2024

Follow-up Duration: Median 1104 days (1558.5 patient-years)

Centers: 7

Countries: United Kingdom, Italy

Sample Size: 543

Analysis: LASSO-penalized Cox proportional hazards regression with internal-external cross-validation


Inclusion Criteria

  • Adults (≥18 years) with a diagnosis of ESUS stroke or TIA, supported by consistency between symptoms and findings on brain MRI or CT
  • ESUS defined per Cryptogenic Stroke/ESUS International Working Group criteria: (a) non-lacunar stroke on MRI/CT; (b) absence of extracranial stenosis ≥50% in arteries supplying the ischemia area; (c) no major-risk cardioembolic source; (d) no other specific cause identified
  • Systematic ICM implantation (Reveal LINQ, Medtronic) for AF monitoring
  • Enrolled consecutively between 2021 and 2024 at 7 stroke centres in the UK and Italy
  • Pre-implantation workup: 12-lead ECG, transthoracic or transoesophageal echocardiography, and ≥24h cardiac monitoring

Exclusion Criteria

  • Life expectancy less than 6 months
  • Prosthetic mechanical valve
  • Pacemaker
  • Hepatic disease associated with coagulopathy (prothrombin time prolonged beyond normal range)
  • Clinically relevant bleeding risk, including cirrhotic patients with Child-Pugh B or C
  • Estimated glomerular filtration rate (eGFR) <15 mL/min/1.73 m²

Arms

FieldSingle observational cohort
InterventionAll patients received ICM insertion after ESUS diagnosis with protocolized follow-up and remote monitoring for AF detection
DurationMedian 1104 days

Outcomes

OutcomeTypeControlInterventionHR / OR / RRP-value
New-onset atrial fibrillation detected during ICM follow-up (AF defined as irregular heart rhythm without detectable P waves lasting >30 seconds)Primary<0.001 for all pairwise hazard comparisons
Model discrimination (AUC) | Overall AUC: 0.85 (95% CI 0.82-0.89); 1-year AUC: 0.86; 3-year AUC: 0.80; Cross-validated AUC: 0.84Secondary
Medium vs Low risk | 95% CI: 2.6-8.4Secondary4.7<0.001
AF-free survival difference (High vs Low) | Difference: 918 fewer AF-free days in high-risk groupSecondary

Subgroup Analysis

Internal validation via 10-fold cross-validation and internal-external cross-validation across the 7 participating centres demonstrated consistent discrimination and calibration across geographically distinct sites (detailed results in Supplementary Appendix Figures S6-S11, Tables S9-S10).


Criticisms

  • Retrospective design — inherent risk of selection and information bias despite consecutive enrollment from prospective registries.
  • Moderate sample size (N=543) limits power for rare subgroup analyses.
  • Cohort limited to UK and Italy — external generalizability to non-European populations is unknown.
  • The score has NOT been validated in a fully independent external cohort. Internal-external cross-validation across participating centres represents an intermediate strategy but does not replace true external validation.
  • No head-to-head comparison with existing AF prediction scores (HAVOC, STAF, AF-ESUS, Brown ESUS-AF, ACTEL, Decryptoring) in the same cohort.
  • ICM monitoring protocols may vary across centres despite standardized device (Reveal LINQ).
  • Follow-up duration (median 1104 days, max 4.7 years) may be insufficient to refine long-term predictive accuracy beyond 5 years.
  • The score was developed and validated exclusively in ESUS patients systematically monitored with ICMs — applicability to non-ICM monitoring modalities is uncertain.
  • Residual confounding from unmeasured factors cannot be excluded despite comprehensive adjustment for clinical covariates.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article (investigator-initiated, non-industry-funded study).

Based on: CATCH-AF Score (International Journal of Stroke, 2026)

Authors: D'Anna L, Favruzzo F, Baracchini C, et al.

Citation: D'Anna L, et al. Int J Stroke. 2026. doi:10.1177/17474930261428118.

Content summarized and formatted by NeuroTrials.ai.