Our study demonstrated the lower EMR on admission was associated with higher risk of three-month poor functional outcome in patients with AIS. This association was independent of established risk factors for stroke prognosis, and had a segmental and different population-specific trend. On the left side of the inflection point (EMR<0.28), the effect size was 0.00 (95 % CI: 0.00–0.08, P=0.0003). On the right side of the inflection point (EMR≥ 0.28), the relationship between EMR and poor outcome cannot be observed (0.44, 95 % CI:0.10-2.02, P=0.2897). And the stronger association between EMR and poor outcome was detected in minor stroke populations by subgroup analysis. Besides, EMR had a bigger AUC than eosinophil or monocyte count, and showed good calibration. Furthermore, adding EMR to conventional risk factors could improve risk prediction for poor outcome. These findings indicated that EMR may be a potential prognostic biomarker for ischemic stroke.
Previous studies have shown that stroke triggers an acute decrease in circulating eosinophil counts while monocytes are significantly increased after AIS [6]. Furthermore, post-stroke low circulating eosinophil count was inversely associated with stroke severity and risk of mortality, and high peripheral blood monocyte level was associated with high risk of poor outcome after stroke [7, 8]. These studies, however, did not control for stroke severity and have other issues such as its retrospective nature or the small sample size that limit interpretation of the findings. In addition, all of these studies are mainly focused on a single subpopulation of leukocytes, which may not provide a comprehensive study for the eosinophils and monocytes [7, 8]. Thus, eosinophil-to-monocyte ratio (EMR), a novel biomarker reflecting the integrated application value of eosinophils and monocytes, is needed to identify patients at high risk of poor prognosis. A recent study revealed that a lower EMR on admission was associated with higher 1-month and long-term mortality in patients with ST-elevation myocardial infarction undergoing primary percutaneous coronary intervention [14]. To the best of our knowledge, this is the first study to study EMR in stroke. We found the lower EMR was associated with poor outcome, and also proved EMR was of certain value in predicting poor outcome in patients with AIS.
In the present study, we demonstrated that there was a saturating effect on the linear relationship between poor outcome and EMR. The inflection point we calculated by the recursive algorithm was 0.28. The result means the negatively linear association between poor outcome and EMR is only present in participants with relatively low EMR level. For those with relatively high EMR level, this linear relationship cannot be found. There is no current study to explore the non-linear relationships between poor outcome and EMR, but the previous investigations might give us some clues. A recent study revealed that the adjusted odds ratio reflecting the effect sizes of eosinophils on cerebral infarct volume in the Q4 (the reference group), Q3, Q2 and Q1 group were 1.00, 2.416, 4.988 and 50.791, respectively. Similarly, on the effect sizes of eosinophils on poor outcome, the odds ratio in 4 quartiles of eosinophils (from high to low) were 1.00, 2.747, 4.804 and 30.2991, respectively [7]. Moreover, on the effect sizes of monocytes on novel plaque formation, the odds ratio in 4 quartiles of monocytes (from low to high) were 1.00, 1.17, 1.12 and 1.85, respectively [15]. This kind of non-equidistant changes in effect size suggested that there may be a nonlinear relationship between EMR and poor outcome. Given that GAM can handle non-parametric smoothing and has obvious advantages in dealing with non-linear relations, the use of GAM will help us to better discover the real relationships between EMR and poor outcome [11]. Therefore, we use the GAM to clarify their nonlinear relationship in the present study and a nonlinear relationship between EMR and poor outcome was detected after adjusting for all potential covariates.
Given that the percentage of eosinophils was negatively correlated with infarct volume and eosinopenia had the potential to predict the severity of AIS [7, 16, 17], we adjusted baseline NIHSS in the multivariable models and performed further subgroup analyses stratified by NIHSS score to eliminate the confounding effects of the severity of AIS. We found that the significant association of EMR with poor outcome was independent of the baseline NIHSS in stroke patients, especially in participants with minor stroke (NIHSS score < 4; P for interaction=0.0015 in the fully adjusted model). These results suggested that EMR had additional prognostic value when baseline NIHSS was considered, and the application conditions of EMR for the prediction of functional outcome among patients with AIS needed to arouse our attention; in other words, we should apply EMR for the prediction of poor outcome in participants with minor stroke. The reason why the effect sizes of EMR on poor outcome showed significant differences in different NIHSS score remains unclear. We hypothesized that the complicated complication of patients with severe stroke and higher NIHSS itself might cause some disruptions to the relationship between EMR and poor outcome. Further studies are needed to investigate this hypothesis.
In the present study, we also explored whether EMR had greater and additional prognostic value for poor outcome by using of various statistical methods. First, the AUC of EMR for poor outcome is greater than that of eosinophil or monocyte count, suggesting that EMR is superior to only the eosinophil count or monocyte count for distinguishing the occurrence of poor outcome. The reasons for such a superiority of EMR may relate to what the EMR represents. The EMR reflects the balance between eosinophil and monocyte levels, which may be comprehensively summarize the overall systematic inflammation conditions [14]. Second, the NRI value is positive, though statistically it makes no difference. Moreover, the IDI value is not only positive, but also statistically significant. That, along with the guide advises that the visual representation of the relationship between predicted and observed (not the specific P value) is the best way to evaluate calibration [17], suggest EMR could significantly improve the predictive power for primary outcome beyond established traditional risk factors (NRI: 3.54%; IDI: 2.11%).Therefore, we hypothesized that serum EMR might be useful in risk stratification of poor outcome among patients with ischemic stroke and could assist the selection of high-risk patients in future clinical practice. If patients have low EMR levels at admission, they may be at high risk of poor outcome and should receive aggressive monitoring and therapeutic interventions.
The mechanisms underlying these observations are not well established, but they seem to be related to the roles of eosinophils and monocytes in ischemic insult. Eosinophils are able to secrete over 35 cytokines, growth factors and chemokines, including IL-4, IL-13 and vascular endothelial growth factor (VEGF). IL-4 and IL-13 are capable of inducing the activation of the M2 phenotype microglia, which possess neuroprotective properties by facilitating the resolution of inflammation. And VEGF might be neuroprotective by the modulation of angiogenesis [5, 19, 20]. Monocytes are able to infiltrate into the central nervous system as early as 4 h following acute brain ischemia, which can contribute to inflammation and brain injury on one hand, but on the other hand, certain subpopulations of monocytes are beneficial with a phenotype that could promote the resolution of inflammation, angiogenesis and tissue repair [21]. The reasons for such a difference may relate to monocytes being a heterogeneous population with proinflammatory or anti-inflammatory phenotypes. The phenotypes of monocytes include CD14highCD16- (classical monocytes), CD14dimCD16 + and CD14highCD16 + [21]. It is worth noting that the classical monocytes account for nearly 90%, which have deleterious effects after stroke. Nevertheless, the other two phenotypes contribute about 10% of monocytes, which play a beneficial role in patients with stroke [21]. Therefore, the comprehensive effects of peripheral blood monocytes reflect the roles of classical monocytes during the study of monocytes as a whole. Given the deleterious effects of classical monocytes and the possible neuroprotective effect of eosinophils, our study found the lower EMR on admission was associated with higher risk of three-month poor functional outcome in patients with AIS.
The main strength of our study is that the clinical information and blood samples of all patients were collected in a prospective fashion with a relatively large sample size.
The great stroke severity is associated with increase in leukocyte levels and the risk of poor outcome [5, 19-22]. We limited these potential confounders by adjusting NIHSS score in multivariate logistic regression model and conducting subgroup analyses to assess the robustness of association between low EMR and poor outcome of AIS. In addition, we provided a comprehensive study for the eosinophils and monocytes by combining the two into a novel biomarker (EMR). Nonetheless, our current findings also have some limitations. First, the majority of acute critical patients were transferred to higher level hospitals due to the grass-roots hospitals nature of ours, which might result in the baseline NIHSS scores being relatively low and existing deviations of the enrolled patients. Thus, the findings might not generalize to other populations, particularly those with high NIHSS scores. Second, the predictive value of EMR for poor outcome is relatively low. The reasons for the low predictive value may relate to the complicated roles of eosinophils and monocytes in ischemic insult, which have proinflammatory or anti-inflammatory properties [5, 19, 21]. For example, we did not categorize the monocytes subpopulations which include CD14highCD16-, CD14dimCD16 + and CD14highCD16 + phenotypes [21]. Further studies are needed to investigate this hypothesis. Third, we neither explored the mechanisms by which eosinophils and monocytes affected the neurovascular unit damage nor investigated what factors regulated the changes of eosinophils and monocytes after ischemic strokes in animal studies. These are going to be the focus of our next work, especially exploring the role of eosinophils in AIS and its mechanism.