Patient characteristics

A total of 333 patients, comprising 143 women (median age = 73.0 years, IQR = 55.5 to 82.0) and 190 men (median age = 71.0 years, IQR = 58.3 to 76.0) were analyzed (Table 1). There were no sex differences in stroke severity (NIHSS). Successful recanalization was achieved in 73 women (51.0%) and 88 men (46.0%), with no significant difference between women and men (χ2Pearson = 0.73, p = 0.39). AIS etiology was also similar between women and men (χ2Pearson = 1.25, p = 0.74).

Table 1 Baseline characteristics of AIS patients who underwent endovascular thrombectomySex differences in blood cell characteristics

In total, 71 BCCs were available (Supplemental Table S1). In the blood sample prior to EVT, 21 out of 71 available erythrocyte-, leucocyte-, and thrombocyte-related BCCs were significantly different between women and men (pFDR-corrected < 0.05) (Fig. 1, Step 1; Supplemental Table S2). For example, the platelet count was higher in women (median = 269.0 × 109/L, IQR = 228.0 to 325.3 × 109/L), than in men (median = 217 × 109L, IQR = 178 to 267 × 109/L, pFDR-corrected =  < 0.001) and the hemoglobin level was lower in women (median = 8.5 mmol/L, IQR = 7.9 to 9.1 mmol/L) than in men (median = 9.0 mmol/L, IQR = 8.3 to 9.6 mmol/L) (Supplemental Table S2).

Clustering of blood cell characteristics

Multicollinearity was present in multiple BCCs (Fig. 1, Step 2; Supplemental Figure S1). Highly collinear BCCs were clustered, based on which synthetic variables representing BCCs clusters were created. The optimal number of clusters was 58 (similar for men and women) (Rand-index 0.98). For example, red blood cell characteristics like hemoglobin and hematocrit were clustered, as well as platelet characteristics like the absolute platelet count by impedance and optics (Supplemental Table S1). The following analyses were based on the 58 BCC clusters (Fig. 1, Step 2). A mapping between the cluster numbers and the corresponding blood cell characteristics can be found in Supplemental Table S1.

Sex stratification leads to better explanation of recanalization success in women

First, a pooled sPLS-DA analysis was performed to distinguish between successful and unsuccessful recanalization based on the BCC clusters, with an average classification error rate of 0.43 (SD = 0.02) (Fig. 1, Step 3 and 4). Sex-stratified analyses showed that the error rate was significantly lower for the female-only models (0.41; SD = 0.02) compared to the pooled (t(192.4) = 5.9, p < 0.001), and male-only models (0.47; SD = 0.03) (t(182.6) = -15.6, p < 0.001) (Fig. 1, Step 4, Fig 2). These results were visually confirmed in the sPLS-DA plot, where successful and unsuccessful recanalization groups were further apart for women than men (Fig. 3).

Fig. 2

Error rates of the pooled, female, and male sPLS-DA models to distinguish successful from unsuccessful recanalization with blood cell characteristics, based on fivefold cross-validation repeated 100 times. Differences in error rate distributions, defined as the average number of misclassified samples divided by the total number of samples, tested using Welch two sample t-tests

Fig. 3

Sparse PLS Discriminant Analysis (sPLS-DA) to distinguish patients with successful (TICI 3) and unsuccessful (TICI 0-2c) recanalization after enodvascular thrombectomy (EVT) based on blood cell characteristics. A–C Plots of sPLS-DA performed on baseline BCC clusters reflect two components trained to distinguish successful recanalization (green, circles) from unsuccessful recanalization (red, triangles) after endovascular thrombectomy. A Results of the sPLS-DA procedure trained on pooled data of women and men; B Results of the sPLS-DA procedure trained on data of women; C Results of the sPLS-DA procedure trained on data of men

Generic and sex specific blood cell characteristics related to recanalization success and acute ischemic stroke etiology

In women, 25 of the 58 BCC clusters were selected in relation to recanalization success in more than 90% across repeated cross-validation folds and repeats (Fig. 1, Step 5). In total, 13 of these clusters overlapped with men, 12 were female-specific (Fig. 4A, B). In men, 33 of the 58 BCC clusters fulfilled the prespecified criteria, of which 13 overlapped with women, 20 were male-specific (Fig. 4C, D).

Fig. 4

Selection frequency and loadings of BCC clusters in relation to recanalization success of endovascular thrombectomy in the female and male models across repeated cross-validation folds and repeats. A, C shows the frequency of selection across cross-validation folds and repeats for each BCC cluster for component 1. The stability threshold of 0.9 is indicated by the red line. Only BCCs with stability > 0.9 are shown. The loading plots B, D show the loadings of BCC clusters on the first sPLS-DA component, with the colors indicating in which EVT outcome group the median value was highest (green = successful, red = unsuccessful). For example, the reticulocyte percentage/absolute count (cluster 26) was selected in relation to recanalization success in 100% of the cross-validation folds and repeats in women (A), median reticulocyte percentage/absolute count (cluster 26) was higher in the successful recanalization group (B). Cluster 26 was not selected in relation to recanalization success in men

In women, of the 25 stable BCC clusters related to recanalization success, 15 clusters were also being robust and informative for AIS etiology (Fig. 1, Step 6). In total, 3 of these overlapped with men, including the coefficient of variance of lymphocyte complexity of intracellular structure (cluster 48), and the absolute white blood cell count (cluster 4), with elevated levels in women and men with LAA as the cause of stroke, and the immature reticulocyte fraction (cluster 27), with elevated levels in women and men with a cardioembolism as the cause of stroke. The other 12 were female-specific (Fig. 5A, B), including primarily erythrocyte-related BCC clusters with elevated levels in patients with a cardioembolism as the cause of stroke, i.e. the red blood cell distribution width (cluster 19), reticulocyte count and reticulocyte percentage (cluster 26), and the percent of erythrocytes with hemoglobin concentration < 28 g/dL (cluster 28), among others, and leucocyte-related BCC clusters with elevated levels in LAA as the cause of stroke, i.e. the white blood cell viability fraction (cluster 5), the neutrophilic granulocyte absolute count (cluster 6), and neutrophil granularity (cluster 39), among others.

Fig. 5

Selection frequency and loadings of BCC clusters previously related to recanalization success of endovascular thrombectomy (EVT), analyzed with  AIS etiology as outcome to provide biological meaning to the BCC clusters related to EVT success. A, C shows the frequency of selection across cross-validation folds and repeats for each BCC cluster for component 1. The stability threshold of 0.9 is indicated by the red line. Only BCCs with stability > 0.9 are shown. The loading plots B, D show the loadings of BCC clusters on the first sPLS-DA component, with the colors indicating in which AIS etiology group the median value was highest (red = cardioembolism, yellow = large-artery atherosclerosis). For example, hemoglobin/hematocrit (cluster 1) was selected in relation to recanalization success in 100% of the cross-validation folds and repeats in women (A), median hemoglobin/hematocrit (cluster 1) was higher in the unsuccessful recanalization group

In men, of the 33 BCC clusters associated with recanalization success, 14 fulfilled the prespecified criteria for being robust and informative for AIS etiology. In total, 3 overlapped with women as mentioned above, 11 were male-specific (Fig. 5C, D), including primarily leucocyte-related BCCs with elevated levels in LAA as the cause of stroke, i.e. the monocyte absolute count (cluster 8), % basophilic granulocytes (cluster 16), and coefficient of variance of DNA/RNA staining in neutrophils (cluster 44), and erythrocyte-related BCCs with elevated levels in cardioembolism as the cause of strokey, i.e. hemoglobin distribution width (cluster 30), and red blood cell complexity of intracellular structure (cluster 53).

Blood cell characteristics related to recanalization success and acute ischemic stroke etiology after controlling for cardiovascular risk factors

In women, after controlling for known cardiovascular risk factors, IVT prior to EVT (β = 0.37, 95% CI = 0.15–5.50, p = 0.026), and higher levels of the reticulocyte count and reticulocyte percentage (cluster 26) (β = 1.02, 95% CI = 0.03–1.51, p = 0.045), were associated with successful recanalization (Supplemental Table S3) (Fig. 1, Step 7). Higher levels of the mean corpuscular hemoglobin concentration (cluster 3) were associated with unsuccessful recanalization (β = − 0.55, 95% CI = -0.97 to − 0.02, p = 0.043). Thus, in combination with previous results, higher reticulocyte levels were associated with increased recanalization success and cardioembolism as the cause of stroke, and higher levels of the mean corpuscular hemoglobin concentration were associated with unsuccessful recanalization and LAA as the cause of stroke in women only.

In men, after controlling for cardiovascular risk factors, the coefficient of variance of lymphocyte complexity of intracellular structure (cluster 48) was associated with unsuccessful recanalization (β = − 0.60, 95% CI = − 0.97 to − 31, p = 0.001) (Supplemental Table S4). In combination with previous results, higher levels of the coefficient of variance of lymphocyte complexity of the intracellular structure was associated with unsuccessful recanalization and LAA as the cause of stroke in men only.