Nomogram to predict ventilator-associated pneumonia in large vessel occlusion stroke after endovascular treatment: a retrospective study

Abstract

Background: Ventilator-Associated Pneumonia (VAP) severely impacts stroke patients' prognosis after endovascular treatment. Hence, this study created a nomogram to predict the occurrence of VAP after endovascular treatment.

Methods: The individuals with acute ischemic stroke and large vessel occlusion (AIS-LVO) who received mechanical ventilation and endovascular therapy between July 2020 and August 2023 were included in this retrospective study. The predictive model and nomogram were generated by performing feature selection optimization using the LASSO regression model and multifactor logistic regression analysis and assessed the evaluation, verification and clinical application.

Results: A total of 184 individuals (average age 61.85 ± 13.25 years, 73.37% male) were enrolled, and the rate of VAP occurrence was found to be 57.07%. Factors such as the Glasgow Coma Scale (GCS) score, duration of stay in the Intensive Care Unit (ICU), dysphagia, Fazekas scale 2 and admission diastolic blood pressure were found to be associated with the occurrence of VAP in the nomogram that demonstrating a strong discriminatory power with AUC of 0.862 (95% CI, 0.810-0.914), and a favorable clinical net benefit.

Conclusion: This nomogram, comprising GCS score, ICU duration, dysphagia, Fazekas scale 2 and admission diastolic blood pressure, can aid clinicians in predicting the identification of high-risk patients for VAP following endovascular treatment in large vessel occlusion stroke.

Keywords: acute ischemic stroke with large vessel occlusion; endovascular treatment; mechanical ventilation; nomogram; ventilator-associated pneumonia.

Figure 1

Flowchart.

Figure 2

The distribution of the modified Rankin scale.

Figure 3

The LASSO regression was used to obtain optimal lambda parameter values for thirty-six features related to VAP-related risk factors. (A) Log(Lambda) value for the thirty-six characteristics in the LASSO regression. A plot of the coefficient profile was generated by plotting against the logarithmic sequence of lambda. The lambda value corresponding to the minimum mean square error is 0.023, while the lambda value for the standard error of the minimum distance is 0.049. (B) Curves for partial likelihood deviation (binomial deviation) and logarithmic (lambda) curves. Draw a vertical dashed line at the optimal value with the minimum standard and twelve variables had a regression coefficient that was not zero. LASSO, the least absolute shrinkage and selection operator.

Figure 4

Nomogram illustrating the prediction of ventilator-associated pneumonia in postoperative mechanically ventilated patients after EVT. GCS score, Glasgow Coma Scale score.

Figure 5

(A) ROC curve of the VAP prediction model. (B) The 10-fold cross-ROC curves for internal validation of the training set. (C) The 10-fold cross-ROC curves of the validation set. (D) The development of the AUC for the training set and validation set throughout the training of the logistic regression model.

Figure 6

Calibration curve with bootstrap resampling 1,000 times of nomogram. The horizontal axis depicts the forecasted VAP vulnerability, while the vertical axis portrays the real diagnosis of VAP. The dashed line that runs diagonally represents a flawless forecast made by an optimal model. The performance of the column chart is represented by the solid line, and the prediction effect improves as it gets closer to the diagonal dashed line.

Figure 7

Analysis of decision curves for VAP-risk nomogram. The x-axis depicts the probability threshold, where the patient’s treatment should be initiated upon reaching a specific threshold. The y-axis represents the net clinical benefit for the patient receiving the treatment. The VAP clinical diagnostic model is represented by the red line in the figure, along with the two lines representing the two extreme cases. A sturdy, dense line depicts the assumption that no patients have VAP. The assumption that all patients have postoperative VAP is represented by a thin solid line.