The Time Is Brain Review

Phenotypic determinants of atrial fibrillation in ischaemic stroke: a retrospective study

· DOI: 10.1136/openhrt-2026-004393

stroke deep-dive atrial-fibrillation ischemic-stroke

Reconstructed on October 14, 2026 for the missed October 6, 2026 publication slot. This is a replacement article, not a recovered original. Source selection: the paper appeared in the archived October 6 briefing; its selection is reconstructed from that briefing, not recovered from an original article draft.

Clinical Question (PICO)

Among consecutive patients clinically diagnosed with ischaemic stroke or transient ischaemic attack (TIA) at one Dutch university centre, which demographic, clinical and cardiac features were associated with documented atrial fibrillation (AF), and did data-driven phenotypes identify different AF prevalence?[1] There was no assigned intervention: comparisons were patients without documented AF versus those with previously known or newly detected AF.[1] The paper does not report a prespecified subgroup analysis; its multinomial model separates known from newly detected AF.[1]

Bottom Line

Among 1,836 patients, 30.5% had documented AF: 19.7% previously known and 10.8% detected after the index event.[1] In adjusted analyses, higher admission NIHSS, prior ischaemic stroke, heart failure, mitral regurgitation, left atrial volume and peak E-wave velocity were associated with AF; the highest-prevalence GTM phenotype had 47.9% AF.[1] These are retrospective associations under non-standardized monitoring, not a validated screening rule.[1]

Design

  • Study type: Single-centre retrospective cohort; observational, with no randomization or assigned treatment.[1]
  • N: 1,836 patients in the analytic cohort; no intervention arms.[1]
  • Setting: Radboud University Medical Center, Nijmegen, the Netherlands; emergency department, neurology ward and outpatient neurology clinic.[1]
  • Study period: January 2019–October 2024; data collection December 2024–February 2026.[1]
  • Follow-up: Median 23.8 months (IQR 6.4–45.8).[1]
  • Analysis: Descriptive comparisons, logistic and multinomial multivariable logistic regression with multiple imputation for multivariable analyses, plus generative topographic mapping (GTM) phenotyping.[1] No single primary outcome is designated; the study describes AF prevalence and associated factors.[1] The binary model combined known and newly detected AF; the multinomial model separated them.[1]

Population

Inclusion Criteria

  • Consecutive patients with a treating-physician clinical diagnosis of ischaemic stroke or TIA at the centre during the study period.[1] The abstract describes the cohort as patients with ischaemic stroke; the Methods also include TIA, and a separate TIA denominator is not reported.[1]

Exclusion Criteria

  • No formal exclusion criteria are enumerated for this cohort in the report.[1] Do not infer exclusions beyond the stated clinical diagnosis and available retrospective records.[1]

Baseline Characteristics

  • Median age 73 years (IQR 64–80); 810/1,836 (44.1%) were female; median admission NIHSS 5 (IQR 2–13).[1]
  • AF was previously known in 361 (19.7%) and newly identified after stroke in 199 (10.8%).[1] AF required documentation on ECG, Holter or clinical telemetry; smartwatch/self-reported episodes were excluded unless confirmed.[1]
  • ECG was recorded in 1,608 (87.6%), Holter monitoring in 422 (23.0%; median 1.5 days, IQR 1–7), and echocardiography in 446 (24.3%).[1] Implantable cardiac monitors were not used routinely.[1]

Interventions

  • Exposure/comparison: No treatment was assigned.[1] The investigators compared documented AF with no documented AF, then separately compared previously known and newly detected AF with no AF.[1] Clinical diagnostic work-up followed routine care and was not standardized by the study.[1]

Outcomes

The principal outcomes were AF prevalence and associations with documented AF; odds ratios describe association, not treatment effect or causation.[1] Estimates marked * were based on log-transformed variables in the original model.[1]

Finding Estimate 95% CI P value
Any documented AF, N=1,836 30.5% Not reported —
Previously known AF 19.7% Not reported —
Newly detected AF after stroke 10.8% Not reported —
Admission NIHSS* (adjusted OR) 1.32 1.12–1.55 0.001
Prior ischaemic stroke (adjusted OR) 2.18 1.28–3.70 0.005
Heart failure (adjusted OR) 3.18 1.37–7.41 0.008
Mitral regurgitation (adjusted OR) 2.34 1.40–3.91 0.001
Left atrial volume* (adjusted OR) 5.58 2.25–13.82 <0.001
Peak E-wave velocity* (adjusted OR) 6.41 2.20–18.70 0.001
Binary model discrimination, AUC 0.86 0.84–0.88 Not reported

The multinomial model’s AUC was 0.85 (0.83–0.88) for previously known AF versus none and 0.79 (0.76–0.81) for newly detected AF versus none; the report does not provide a single prespecified primary contrast or multiplicity-adjusted P values for these models.[1]

Among the 199 newly detected cases, 81.4% were identified by ECG, 7.0% by Holter, 7.6% by telemetry and 4.0% by other documented rhythm assessments.[1] These percentages describe detection routes in this care setting, not comparative test sensitivity.[1]

GTM identified four phenotypes; AF prevalence and reported univariable odds relative to phenotype 2 were:[1]

GTM phenotype N AF prevalence Univariable OR for AF (95% CI) vs phenotype 2
1: older, mostly male; vascular/cardiac burden 687 38.6% 5.26 (3.74–7.56)
2: younger, mostly male; lowest AF prevalence 403 10.7% Reference
3: oldest; greatest severity and cardiac burden 438 47.9% 7.71 (5.39–11.25)
4: youngest; high stroke severity, lower cardiac burden 308 13.6% 1.32 (0.84–2.08)

Safety and subgroup outcomes

This was not an intervention study; comparative treatment safety was not assessed.[1] No randomized or prespecified treatment subgroup effects are reported.[1]

Figures

Adjusted associations with AF in a forest plot.
Figure 1. Forest plot of multivariable associations with AF; the strongest reported associations include left atrial volume and peak E-wave velocity.[1]

Source: van Kempen et al., Open Heart (2026), PMC13646944, openhrt-13-2-g001.webp. Original figure, unmodified; CC BY-NC 4.0 (license). [1]

Adjusted associations for known and newly detected AF in two forest plots.
Figure 2. Multinomial regression contrasts previously known AF and newly detected AF separately against no AF.[1]

Source: van Kempen et al., Open Heart (2026), PMC13646944, openhrt-13-2-g002.webp. Original figure, unmodified; CC BY-NC 4.0 (license). [1]

Criticisms

  • Internal: Retrospective, non-standardized cardiac work-up and only 23.0% Holter use mean some patients classified as AF-free may have occult paroxysmal AF; monitoring intensity can create ascertainment bias.[1]
  • Internal: Echocardiography was performed selectively in routine care (24.3% of the cohort); associations involving echo markers may reflect selection, and mitral-regurgitation severity/mechanism was not consistently characterized.[1]
  • Methodological: The four GTM phenotypes and proposed monitoring implications are explicitly hypothesis-generating; external prospective validation is needed.[1] The full binary model’s cardiac-marker availability was limited, and the complete-work-up sensitivity analysis included only 217 patients.[1]
  • External commentary: No external editorial or letter is assessed in this source-grounded draft; no outside critique is presented as retrieved evidence.

Funding

The EU Horizon Europe TARGET project (Grant Agreement 101136244) supported the listed authors; the funding statement does not specify a funder role.[1] Gregory Y H Lip reported institutional consultancy/speaker roles for BMS/Pfizer, Boehringer Ingelheim, Anthos and Huawei, with no personal fees; the other authors declared none.[1]

The paper

  • Authors. Evi J van Kempen, Wouter V Elzinga, Ivan Olier, Sandra Ortega-Martorell, Gregory Y H Lip, Tim J F ten Cate and Anil M Tuladhar.[1]
  • Title. Phenotypic determinants of atrial fibrillation in ischaemic stroke: a retrospective study.[1]
  • Journal. Open Heart. 2026;13(2):e004393.[1]
  • DOI. 10.1136/openhrt-2026-004393
  • PMCID / PMID. PMC13646944 / 42833699.[1]
  • Source license. Creative Commons Attribution Non Commercial 4.0 (CC BY-NC 4.0); see the license terms.[1]
  • Primary source mapping: [1] https://www.ebi.ac.uk/europepmc/webservices/rest/PMC13646944/fullTextXML — Phenotypic determinants of atrial fibrillation in ischaemic stroke: a retrospective study.
Deep Dive — click to expand

What this is

This is a description of who had documented AF in one hospital’s retrospective stroke/TIA cohort and which measured features travelled with that diagnosis.[1] Its useful contribution is not a new detection device or a validated prediction score; it is the conjunction of routine-care AF prevalence, adjusted associations and data-driven phenotypes.[1] The distinction matters because monitoring was not uniform and the design cannot show that selecting patients by phenotype improves detection or outcomes.[1]

1. Shadow Audit

The headline “new AF” count is 199/1,836 (10.8%), but detection arose under the local monitoring pattern: 87.6% had ECG, 23.0% Holter and no routine implantable monitoring.[1] Of the newly detected cases, 81.4% were attributed to ECG and only 7.0% to Holter.[1] Those figures are not evidence that one modality is more sensitive; exposure to each modality differed and was selected in routine care.[1] Thus, 30.5% is the prevalence of documented AF under this center’s ascertainment, not a standardized estimate of all AF or an estimate of monitoring yield under equal surveillance.[1]

A second hidden boundary is the echocardiography subset: only 446/1,836 (24.3%) received echo.[1] Strong adjusted associations for left atrial volume and peak E-wave velocity therefore rest on incompletely observed cardiac data, with the authors warning of selective testing.[1] Multiple imputation addresses missingness under model assumptions; it does not erase selection into having the test.[1]

2. Inversion Engine

The phenotype contrast could weaken if prospective, uniform rhythm monitoring found similar rates of previously occult AF across groups, or if the observed differences were largely generated by who received cardiac testing.[1] The reported phenotype 3 odds ratio of 7.71 (95% CI 5.39–11.25) compares its combined documented-AF prevalence with phenotype 2; it is not a relative probability of newly detected AF and does not establish a clinical cut point.[1] Confidence would rise if an external cohort reproduced the phenotype structure and calibrated its AF detection under a common monitoring protocol.[1]

3. Second-Order Catalyst

The defensible next move is a validation study, not an automatic change to bedside monitoring.[1] A prospective design could prespecify the phenotype variables, apply the same rhythm-monitoring schedule across strata, and report new AF separately from previously known AF.[1] It should test whether phenotype-guided prioritization adds detection beyond simple clinical/cardiac features and whether that extra detection changes management or patient-centered outcomes.[1] Those are proposed tests, not outcomes established here.[1]

4. Asymmetric Leverage

A widely available marker such as stroke severity may be easier to evaluate at scale than echocardiographic remodeling measures, which were observed in fewer than one-quarter of the cohort.[1] The authors’ binary model had AUC 0.86 (95% CI 0.84–0.88), while the model excluding echo factors had AUC 0.81 (0.78–0.84); these are model-specific apparent discrimination estimates, not a randomized comparison of screening strategies or proof of incremental clinical utility.[1] The smaller complete cardiac-work-up analysis (N=217) had AUC 0.94 but wider, less stable estimates, underscoring why a higher AUC in that selected subset should not be treated as better validated performance.[1]

5. Paradigm Destroyer

The paper challenges the idea that detected AF after stroke is distributed uniformly across clinical profiles; it does not overturn existing monitoring practice or prove that a particular phenotype should receive anticoagulation.[1] Its most useful reframing is to ask whether clusters of stroke severity, prior vascular events and cardiac remodeling can help design better-tested surveillance strategies.[1] Any protocol-level change would require external validation and evidence that the strategy improves meaningful outcomes, neither of which this retrospective analysis supplies.[1]

MVP — Minimum Viable Proof

In this cohort, 30.5% had documented AF, including 10.8% first identified after the event, and phenotype 3 had the highest combined AF prevalence at 47.9%.[1] These numbers support prospective testing of targeted detection hypotheses, but the variable work-up and single-center design do not establish a screening algorithm or causal benefit.[1]

Best Combination

Use this study as a hypothesis generator alongside a future validation protocol that keeps AF ascertainment comparable across risk groups and separates known from newly detected AF.[1] The source itself points to individualized AF detection as a research direction; this draft does not import uncited external trial or guideline estimates.[1] Clinically, the study’s association results should be read with the treating team’s usual stroke-mechanism assessment rather than as a stand-alone rule.[1]

Overvalue Warning

Do not equate a high prevalence of documented AF in a phenotype with evidence that phenotype-based screening finds more treatable AF under equal monitoring.[1] Do not interpret adjusted odds ratios as causal effects or as a validated bedside score: the cohort was retrospective, single-centre, selectively monitored, and the phenotypes require prospective external validation.[1]

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.