Abstract
Dear Editor,
We thank Aashish and colleagues for their thoughtful and constructive comments on our systematic review and meta-analysis comparing early versus later anticoagulation after ischemic stroke in patients with atrial fibrillation. We appreciate their careful reading of our work and welcome the opportunity to clarify the rationale for our methodological approach and interpretation of the findings.
Aashish and colleagues correctly highlighted that low I2 values should not be interpreted as evidence of complete clinical homogeneity. We agree with this important distinction. In our study, the low I2 values reflected statistical consistency in the observed treatment effects, rather than the absence of clinical variation across trials. The included randomized controlled trials differed in their timing windows, stroke severity distributions, comparator strategies, and follow-up durations. These differences were acknowledged in our article and were considered when interpreting the findings.
However, clinical diversity does not automatically preclude meta-analysis. A quantitative synthesis remains appropriate when studies address the same underlying clinical question, evaluate broadly comparable strategies, and report clinically aligned outcomes. In our review, all included randomized trials examined the timing of anticoagulation initiation after ischemic stroke in patients with atrial fibrillation. The trials also reported outcomes that are central to this clinical decision, including recurrent ischemic stroke, symptomatic intracerebral hemorrhage, and mortality. Therefore, the rationale for pooling was not based solely on low I2 values, but on the shared clinical decision problem faced by clinicians: whether earlier initiation of anticoagulation is associated with a different balance of recurrent ischemic and bleeding outcomes compared with later initiation.
The purpose of the meta-analysis was not to imply that all timing strategies, infarct profiles, or patient populations were interchangeable. Rather, pooling was used to summarize the totality of randomized evidence and to improve the precision of effect estimates for relatively infrequent outcomes. This is particularly relevant because symptomatic intracerebral hemorrhage and recurrent ischemic stroke occurred at low rates in the individual trials, making single-study estimates unstable and underpowered. A narrative synthesis alone would have been less able to quantify the direction, magnitude, and uncertainty of the available randomized evidence.
We also took steps to address clinical heterogeneity in both the analysis and interpretation. A random-effects model was used because some degree of clinical and methodological diversity was expected. Differences in timing definitions and trial characteristics were described in the article, and we performed a sensitivity analysis excluding the AREST trial because of its distinct comparator arm involving warfarin rather than delayed DOAC initiation. Importantly, exclusion of AREST did not materially alter the direction, magnitude, or statistical significance of the pooled estimates for recurrent ischemic stroke, symptomatic intracerebral hemorrhage, or all-cause mortality. This finding supports the robustness of the main results and suggests that the overall conclusions were not driven by this structurally different trial.
We also appreciate the concern regarding the interpretation of the pooled estimate for recurrent ischemic stroke. The pooled odds ratio of 0.80, with a 95% confidence interval of 0.59 to 1.09, did not reach statistical significance. We agree that this result should not be interpreted as definitive evidence that early anticoagulation reduces recurrent ischemic stroke. Our description of a “non-significant reduction” was intended to describe the numerical direction of the point estimate, not to imply a statistically proven benefit. A more precise interpretation is that early anticoagulation showed a numerically lower but statistically non-significant estimate for recurrent ischemic stroke, with the confidence interval remaining compatible with no effect and a small increase in risk.
We further agree that infarct volume is a clinically important determinant of hemorrhagic transformation risk and may influence the optimal timing of anticoagulation. However, our analysis was limited by the availability of aggregate trial-level data. The included randomized controlled trials did not provide sufficiently consistent or extractable subgroup data stratified by infarct volume to permit a meaningful pooled subgroup analysis. For this reason, we emphasized individualized decision-making based on stroke severity, infarct characteristics, bleeding risk, and clinical judgment. We agree that future individual patient data meta-analyses would be valuable to determine whether the effect of early versus later anticoagulation differs according to infarct volume, imaging features, stroke severity, and other patient-level risk factors.
Therefore, we respectfully maintain that a meta-analysis was methodologically justified as a summary of the current randomized evidence, while also recognizing that the pooled estimate should not be applied rigidly to every clinical scenario. The presence of clinical variation among trials is precisely why our conclusion emphasized individualized decision-making rather than a uniform time-based recommendation. Our findings should be interpreted as an aggregate estimate of the available randomized evidence, not as a substitute for bedside assessment of infarct characteristics, neurological severity, and bleeding risk.
In conclusion, we thank Aashish and colleagues for emphasizing the distinction between statistical and clinical heterogeneity and for highlighting the need for more granular patient-level evidence. Their comments reinforce the central message of our review: early anticoagulation after ischemic stroke in atrial fibrillation appears safe in selected patients, without clear evidence of increased symptomatic intracerebral hemorrhage or mortality, but the timing of initiation should remain individualized according to clinical and imaging characteristics.
