Discussion
The major finding of this study is that VA-ECMO cannulation location did not appear associated with stroke risk. Ischemic stroke was the most common stroke subtype, followed by hemorrhagic transformation, and intracranial hemorrhage. Lesions were most common in the right hemisphere, followed by the left hemisphere lesions, bilateral lesions, and vertebrobasilar lesions.
Current literature suggests that 10-50% of VA-ECMO patients suffer neurological complications 6,8,12. Variable incidence rates may be attributed to inconsistent definition of what constitutes a neurological complication, with some studies including brain death, diffuse anoxic brain injury, and coma under the scope of neurological complications while other studies do not. Neurological complications include subclinical cognitive impairment, seizures, paraplegia, peripheral neuropathy, and ischemic and hemorrhagic strokes. Ischemic stroke and hemorrhagic stroke are catastrophic complications associated with ECMO. Incidence rates for stroke range from 4.2-7.8% among VA-ECMO patients6-8. The true stroke incidence rate may be higher as stroke events are likely unrecognized. Neurological diagnosis is often difficult due to deep sedation and systolic and metabolic derangements encountered in intensive care unit patients. In a prospective study, Mateen et al. found that despite a lack of clinical diagnosis of stroke, autopsy studies of 9 of 10 brains showed hypoxic-ischemic and hemorrhagic lesions of vascular origin12.
The specific pathophysiology leading to ischemic or hemorrhagic stroke during VA-ECMO is poorly understood. There are multiple proposed stroke risk factors in patients undergoing VA-ECMO. Such risk factors may include cardiac arrest, cardiac surgery, myocardial infarction, heart failure, need for renal replacement therapy, sepsis, duration of ECMO, presence of bleeding at other sites, and non-pulsatile cerebral perfusion7, 13, 14. Small studies have suggested that certain risk factors are independently associated with stroke: Omar et al. found that high pre-ECMO blood lactic acid levels (lactic acid > 10 mmol/L) were independently associated with ischemic stroke9. In a retrospective, single-center study, Saeed et al. found a strong association between early low-level hemolysis (i.e. 48-hour plasma-free hemoglobin 11-50 mg/dL) with subsequent non-hemorrhagic stroke15. However, both studies were limited by small number of cases with the main outcome variable. As such, assessing the relative impacts of the underlying risk factor and ECMO itself is challenging.
Cannulation site is a potentially modifiable risk factor that has not been extensively studied. The association between outflow cannulation site and neurological complications has been studied among pediatric VA-ECMO patients with mixed results. Incidence rate of neurological complications in neonatal VA-ECMO patients has been estimated to be 20%16. A study of pediatric VA-ECMO patients by Werho et al.17 did not find association between stroke risk and cannulation site in neonates. Another study by Pinto et al.19 found that peripheral cannulation was associated with higher incidence of stroke in pediatric patients. Few studies have studied such associations among adult VA-ECMO patients. Laterality of lesions have not been extensively studied in adult VA-ECMO patients, and reports in pediatric patients have been conflicting19, 21.
This study is subject to limitations particular to non-randomized studies. Due to the small number of cases with the main outcome variable, a main study limitation is the low statistical power of the analysis. In order for our study to have 80% power at alpha = 0.05, the effect size, as demonstrated by the odds ratio, would have to be at least 6 (compared to our findings, where OR=1.05 for both axillary vs. femoral cannulation and central vs. femoral cannulation, and OR=1 for axillary vs. central cannulation). Therefore, our study was underpowered to detect a difference in stroke rates based on cannulation site, if such a difference existed. Second, selection bias may have been introduced in terms of choice of cannulation strategy due to accessibility. For example, most patients who received central cannulation had underlying etiology of postcardiotomy shock for ease of access, and patients with femoral cannulation were more likely to have acute myocardial infarction as underlying cardiogenic shock etiology, as shown by pairwise comparisons with p<0.01 of Bonferroni correction. Larger, multi-center studies are needed to confirm these findings. In order to lower the risk of information bias, all documented stroke outcomes were verified with a CT scan of the head. Therefore, overall incidence of stroke is likely underestimated. Furthermore, the long study period (12 years) introduces technological advances that may have changed patient management during the study period (see Supplementary Table S2).