Response by de Havenon et al to Letters Regarding Article, “Results From DEFUSE 3: Good Collaterals Are Associated With Reduced Ischemic Core Growth but Not Neurologic Outcome”

We would like to thank Pirson et al. and Boulois et al. and for their letters in response to our article in Stroke entitled “Results from DEFUSE 3 Good Collaterals Are Associated with Reduced Ischemic Core Growth but Not Neurologic Outcome.”¹ We believe that the surprising results of our pre-specified analysis of the DEFUSE 3 trial² warrant further discussion and welcome the opportunity to clarify several aspects of our article.

In response to Pirson et al.’s comment that patients in this analysis were “highly selected,” we would like to highlight that we included 71% (130/182) of patients enrolled in DEFUSE 3, which comprises a large portion of the DEFUSE 3 cohort. As we described in the manuscript, the vast majority of excluded patients were those who could not have collaterals assessed due to imaging evaluation by time of flight MR angiography. Additionally, there were no significant differences between our cohort and the entire cohort. The rapid enrollment of DEFUSE 3 and the permissive maximum core size of 70 mL further suggest that the patients in our cohort are not highly selected beyond the novel aspect of being in the extended time window of 6–16 hours from last known normal.

Pirson et al. state that “the association between collaterals and neurologic outcome may have been diminished due to a dichotomized measure.” As reported in the article, both the dichotomized Tan and Maas collateral scales were not associated with neurological outcome (the ordinal Mass data are presented in the Supplementary data section, which is available online). Additional multivariate analyses that used the full range of ordinal values for the Tan or Maas collateral scales did not find an association with neurological outcome. These results strongly suggest that the unanticipated findings of our article are not due to the methodology of collateral measurement.

Pirson et al. are correct that the average baseline core size in DEFUSE 3 was small. When we examined the 33 patients with poor collaterals by the Tan measurement, the core was larger (median = 21.7 mL), but we failed to find an association with neurologic outcome and poor collateral status. We also failed to find an association with worse outcome for the patients with poor collaterals by the Maas measurement, who were a larger cohort of 70 patients. When we adjusted for baseline core infarction and hypoperfused volume in multivariate models, the conclusions of our article were not changed. In short, the findings of this analysis are robust, but unexpected, which is why we urged caution in interpreting the results and advocated for further study, rather than changing practice patterns.

In response to Boulois et al.’s comment that our analysis “merged together successfully EVT-treated patient … with both EVT failures and even medically-treated patients,” we encourage examination of Figure 2 in our article, which demonstrates no difference in clinical outcomes when patients were stratified by collateral status and EVT treatment. Boulois et al. also suggest that “good collaterals would be expected to protect the penumbra only for a limited time.” We would draw attention to our recent publication that examined imaging profiles 24 hours after randomization in the DEFUSE 3 study.³ This study found that approximately 20% of patients retain a Target Mismatch 24 hours after enrollment, which indicates that robust collaterals can persist for prolonged periods of time.³ As we said in our article, “until we better understand collateral evolution, we will not fully appreciate their effect on neurological outcome or comprehend how to therapeutically modify that effect.” Additional research is needed, with follow-up imaging at time points past 24 or 48 hours from last known normal, to validate or refute the findings of this analysis and, more importantly, to provide a viable pathway forward for translational collateral research.