Interpretable Machine Learning to Predict Anxiety or Depression in Acute Ischemic Stroke after IV Thrombolysis: A Retrospective Cohort Study

BACKGROUND: Acute Ischemic Stroke (AIS) represents a prevalent cerebrovascular condition characterized by significant levels of disability and mortality. This study aimed to develop and validate a Machine Learning (ML) model designed to predict the occurrence of anxiety or depression in AIS patients following Intravenous Thrombolysis (IVT). METHODS: Participants from two centers included in CNSR between March 2019 and January 2024 were randomly allocated into a training set (70%) and a test set (30%). 5 ML models (DT, LR, RF, XGBoost, KNN) were constructed utilizing selected variables. Model performance was assessed through calibration curves, AUC, accuracy, and precision metrics. RESULTS: A total of 1,711 participants were analyzed, with 467 diagnosed with anxiety or depression. The cohort was divided into a training set comprising 1,197 patients and a testing set of 514 patients. Among 56 clinical variables, 7 key features were identified based on SHAP values: lesional site, AST, BMI, ALB, age, Apo A1, and hemorrhagic complications. The RF model demonstrated superior overall performance, achieving an AUC of 0.841, accuracy of 0.808, and precision of 0.841 in the training set. DISCUSSION: The RF prediction model developed in this study has the potential to estimate the risk of mood disorders post-IVT in AIS patients prior to treatment, thereby facilitating clearer communication and expedited clinical decision-making. Additionally, it underscores modifiable clinical factors that may enhance patient quality of life. However, the generalizability and efficacy of the model may be constrained by the characteristics of the sample population and the variable selection process. CONCLUSION: The RF model presented in this study may represent a valuable instrument for evaluating the risk of anxiety or depression in AIS patients following IVT within clinical practice.