Systemic immune-inflammation index predicts post-thrombectomy outcomes and reveals a mediating role in the association between neurocardiac stress and prognosis: a multicenter study

Figures

• 

  Figure 1. Risk gradient associated with combined AF-RVR and elevated SII for poor functional outcome after EVT. (A) Forest plot displaying adjusted odds ratios (aORs) with 95% confidence intervals for poor functional outcome across predefined patient strata. (B) Bar chart illustrating the observed proportion of patients with poor functional outcome (modified Rankin Scale [mRS] score 3–6) within each stratum.

• 

  Figure 2. Causal mediation path diagram illustrating the role of high SII in the association between AF-RVR and poor 3-month functional outcome. Values on paths represent adjusted odds ratios (OR) derived from logistic regression models: Path a (AF-RVR → high SII): OR = 2.21 (95% CI, 1.26–3.86), p = 0.006; Path b (high SII → poor outcome, adjusted for AF-RVR and covariates): OR = 3.53 (95% CI, 2.36–5.29), p < 0.001; Path c’ (AF-RVR → poor outcome, adjusted for high SII and covariates): OR = 1.47 (95% CI, 0.78–2.76), p = 0.234. Mediation effects estimated on the absolute risk difference scale (bootstrap, 1,000 iterations): Total effect: 0.123 (95% CI, 0.002–0.258), p = 0.046; Indirect effect (ACME): 0.046 (95% CI, 0.011–0.084), p = 0.018; Proportion mediated: 37.2% (95% CI, −14.8 to 241.0%). All models were adjusted for age, thrombectomy attempts >1, onset-to-puncture time, procedure duration, diabetes, and pre-EVT NIHSS score. Abbreviations: AF-RVR, atrial fibrillation with rapid ventricular response; SII, systemic immune-inflammation index; ACME, average causal mediation effect.

• 

  Figure 3. Nomogram for predicting poor 3-month functional outcome after EVT. The nomogram visualizes the final prediction model derived from the interaction analysis. To estimate an individual patient’s risk, locate each predictor value on its corresponding axis and draw a vertical line to the “Points” scale to obtain the assigned score. The points for all predictors are then summed to yield the “Total Points,” which correspond to the predicted probability of poor functional outcome on the lower risk axis.

• 

  Figure 4. Discrimination and calibration of the prediction model in the derivation and external validation cohorts. (A) Receiver operating characteristic (ROC) curves for the LASSO model and the interaction model in the derivation cohort, with areas under the curve (AUCs) of 0.789 (95% CI, 0.749–0.828) and 0.793 (95% CI, 0.754–0.832), respectively. (B) ROC curves for the two models in the external validation cohort, with corresponding AUCs of 0.710 (95% CI, 0.633–0.787) and 0.712 (95% CI, 0.635–0.789). (C,D) Calibration plots for the interaction model in the derivation (C) and external validation (D) cohorts, demonstrating agreement between predicted probabilities and observed event rates of poor functional outcome. The Hosmer–Lemeshow goodness-of-fit test indicated adequate calibration (p = 0.870 and p = 0.078, respectively). The dashed diagonal line represents perfect calibration.