Predicting 1-Hour Thrombolysis Effect of r-tPA in Patients With Acute Ischemic Stroke Using Machine Learning Algorithm

Abstract

Background: Thrombolysis with r-tPA is recommended for patients after acute ischemic stroke (AIS) within 4.5 h of symptom onset. However, only a few patients benefit from this therapeutic regimen. Thus, we aimed to develop an interpretable machine learning (ML)–based model to predict the thrombolysis effect of r-tPA at the super-early stage.Methods: A total of 353 patients with AIS were divided into training and test data sets. We then used six ML algorithms and a recursive feature elimination (RFE) method to explore the relationship among the clinical variables along with the NIH stroke scale score 1 h after thrombolysis treatment. Shapley additive explanations and local interpretable model–agnostic explanation algorithms were applied to interpret the ML models and determine the importance of the selected features.Results: Altogether, 353 patients with an average age of 63.0 (56.0–71.0) years were enrolled in the study. Of these patients, 156 showed a favorable thrombolysis effect and 197 showed an unfavorable effect. A total of 14 variables were enrolled in the modeling, and 6 ML algorithms were used to predict the thrombolysis effect. After RFE screening, seven variables under the gradient boosting decision tree (GBDT) model (area under the curve = 0.81, specificity = 0.61, sensitivity = 0.9, and F1 score = 0.79) demonstrated the best performance. Of the seven variables, activated partial thromboplastin clotting time (time), B-type natriuretic peptide, and fibrin degradation products were the three most important clinical characteristics that might influence r-tPA efficiency.Conclusion: This study demonstrated that the GBDT model with the seven variables could better predict the early thrombolysis effect of r-tPA.

The feasibility and accuracy of machine learning in improving safety and efficiency of thrombolysis for patients with stroke: Literature review and proposed improvements

“…A recent model in 2021 ( 42 ) predicted the final infarct volumes for patients after thrombolysis therapy. Zhu et al ( 43 ) only predicted 1-h NIHSS after thrombolysis. The early outcome advantage of thrombolysis does not necessarily persist during long-term follow-up.…”

“…Feature selection serves to decrease the number of input variables to both reduce the computational cost of modeling and avoid overfitting. Previous studies performed feature selection with a combination of clinical and statistical judgment: initially selected clinical features were identified by neurologists with clinical expertise or based on related studies, feature engineering was then adopted by some studies to transform raw data (we will explore feature engineering in details in the Section 6); stepwise model building ( 19 , 25 , 27 , 29 , 34 , 39 ), univariate analysis ( 17 , 20 , 28 , 30 , 33 , 38 , 43 , 48 ), multivariable analysis using logistic regression ( 16 , 21 , 24 , 26 , 31 , 32 ), plots displaying the pattern of predictors, and outcome ( 21 ), and Least Absolute Shrinkage and Selection Operator (LASSO) ( 25 , 40 ), was performed to further select statistically significant features among initially selected features and new features generated in feature engineering.…”

“…When we reviewed past related studies, we found that most of the previous models did not pay much attention to feature engineering. There is either no feature engineering ( 16 , 22 , 23 , 26 , 29 , 30 , 32 , 37 – 39 , 41 – 43 , 48 ), or simple feature engineering by calculating clinical assessment scores based on past studies ( 18 , 24 , 31 , 34 ), creating interaction terms ( 21 , 25 , 34 , 36 ), creating dummy variables using different cutoff points ( 17 , 19 , 20 , 33 , 35 ), visual detection of radiological features ( 17 – 20 , 24 , 27 , 28 , 35 ). Tang et al ( 40 ) performed an advanced radiological feature engineering by first dividing the brain into six gray matter regions plus a white matter area and then calculating the mismatch ratio between diffusion lesion and perfusion lesion in each of these seven brain areas.…”

“…The more estimated parameters the model algorithm depends on, the more flexible the model is considered to be. The algorithms used by previous studies, from the most restrictive to the most flexible, were risk score ( 16 , 17 , 19 – 23 , 33 , 35 ), nomogram ( 27 , 31 , 32 , 37 – 39 ), logistic regression ( 25 , 26 , 28 , 29 , 34 , 36 , 40 , 43 , 48 ), tree-based machine learning models ( 18 , 29 , 43 , 48 ), support-vector machine (SVM) ( 18 , 24 , 29 ), and deep learning neural network ( 29 , 30 , 41 , 42 , 48 ).…”

A new machine learning algorithm with high interpretability for improving the safety and efficiency of thrombolysis for stroke patients: A hospital-based pilot study

“…Previous studies leveraged machine learning models to assist neurologists in deciding the safety and efficiency for each patient more accurately 8 – 36 : they all simply reused the existent machine learning algorithm and trained the algorithm based on their patient cohort to predict thrombolysis outcome. However, when reusing the current machine learning algorithms, there is always a trade-off between flexibility and interpretability 37 : Inflexible algorithms have a restrictive ability to estimate the boundaries between different outcome classes, therefore presenting lower predictive power.…”