Risk assessment of acute heart failure after endovascular therapy in acute ischemic stroke: a nomogram-based study

Figures

• 

  Figure 1. The flowchart of the AHF after endovascular therapy in patients with acute large vessel occlusion stroke.

• 

  Figure 2. LASSO regression analysis for posT,stroke AHF feature selection. (a) Binomial deviance versus −log(λ). The displays two vertical dashed lines representing values determined through 10-fold cross-validation. The horizontal axis shows the negative logarithmic transformation of the optimal penalty parameter [–log(λ)], where λ. min (minimum deviation criterion) corresponds to –log(λ) = 4, whereas λ.1se (1-standard-error criterion) corresponds to –log(λ) = 3. The vertical axis depicts the binomial deviance obtained from cross-validation. (b) LASSO coefficient profiles. This illustrates the trajectories of the LASSO coefficients for all 24 variables across a sequence of log-transformed λ values [–log(λ)]. Based on the optimal λ = 0.05, five variables with nonzero coefficients were retained.

• 

  Figure 3. Nomogram of the posT,stroke EVT patients for AHF prediction. The dots represent the prediction results of patients. *** represents the highest level of significance (***p < 0.001) and * (*p < 0.05).

• 

  Figure 4. ROC curves for the AHF predictive nomogram in AIS, LVO patients after EVT. (a) In the training cohort. The area under the AUROC was 0.913 (95% CI: 0.873–0.953). At a risk probability cut-off of 0.331, the model showed a specificity of 0.910 and sensitivity of 0.829. (b) For the validation cohort. The AUROC was 0.909 (95% CI: 0.837–0.980). At a cutoff of 0.175, specificity reached 0.865 with a sensitivity of 0.923.

• 

  Figure 5. The calibration curves for the AHF nomogram in AIS,LVO patients after EVT. (a) The training cohort. (b) The validation cohort. The diagonal dashed line denotes the benchmark prediction of a perfectly calibrated model, and the solid line represents the probability forecasts derived from the nomogram. The level of congruence between predicted and real outcomes is evaluated by the closeness of the solid to the dashed line; minimal separation indicates superior calibration performance. The calibration intercept is 0 and the calibration slope is 1 for both the training set and the validation set.

• 

  Figure 6. The DCA curves illustrate the performance of the AHF prognostic nomogram in AIS, LVO patients following EVT. The red curve (Nomogram) represents the nomogram model constructed in this study, reflecting its standardized net clinical benefit at different high-risk thresholds. The black horizontal line (“No intervention” strategy) serves as a benchmark reference for clinical decision-making, while the gray diagonal line (“Full intervention” strategy) represents an extreme decision-making mode without risk stratification. (a) The training cohort. (b) The validation cohort. A risk threshold range of 5%–80% in the training cohort and 5%–70% in the validation cohort, the model demonstrates superior clinical net benefit compared with both reference curves.

• 

  Figure 7. The CIC analysis for the AHF predictive nomogram in AIS-LVO patients is illustrated. (a) The training cohort. (b) The validation cohort. The number of patients at high risk of AHF per 1000 patients is shown on the vertical axis. The horizontal axis represents the probability threshold for determining patients as being at high risk of AHF. “Cost: Benefit Ratio” refers to the cost-benefit ratio. High-risk AHF patients (solid red line) versus those with actual AHF events (dashed red line).