Explainable machine learning model for predicting the outcome of acute ischemic stroke after intravenous thrombolysis

You're that blitheringly stupid you don't know that predictions NEVER GET SURVIVORS RECOVERED?

And your mentors and senior researchers are no better? You're all fired!

 Explainable machine learning model for predicting the outcome of acute ischemic stroke after intravenous thrombolysis

Fanhai Bu^1,2†Runlu Cai^3†Wei Zhang²Xiaohong Tang⁴Guiyun Cui^2*‡Xinxin Yang^2*‡

• ¹Department of Neurology, The First Clinical College, Xuzhou Medical University, Xuzhou, China
• ²Department of Neurology, Affiliated Hospital of Xuzhou Medical University, Xuzhou, China
• ³Department of Anesthesiology, The First Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China
• ⁴Department of Neurology, Hongze District People's Hospital, Huaian, China

Introduction: Acute ischemic stroke (AIS) patients often experience poor functional outcomes post-intravenous thrombolysis (IVT). Novel computational methods leveraging machine learning (ML) architectures increasingly support medical decision-making. We aimed to develop and validate a machine learning model to predict 3-month unfavorable functional outcome after IVT in AIS patients.

Methods: This retrospective study developed ML prognostic models for 3-month functional outcome (modified Rankin scale scores of 3–6) in IVT-treated AIS patients. A derivation cohort (n = 938) was split 7:3 for training/testing, with an independent external validation cohort (n = 324). The least absolute shrinkage and selection operator (LASSO) regression selected predictors from clinical/neuroimaging/laboratory variables. Eight ML algorithms (including Logistic Regression, Random Forest, Extreme Gradient Boosting, Multilayer Perceptron, Support Vector Machine, Light Gradient Boosting Machine, Decision Tree, and K-Nearest Neighbors) were trained using 10-fold cross-validation and evaluated on test/external sets via the area under the curve (AUC), accuracy, precision, recall and F1-score. Additionally, the SHapley Additive exPlanations (SHAP) interpreted the optimal model.

Results: 938 patients constituted the derivation cohort (training: n = 656, test: n = 282) and 324 patients the external validation cohort. Unfavorable 3-month outcomes (mRS 3–6) occurred in 25.7% and 22.8%, respectively. LASSO regression selected five predictors: the neutrophil-to-lymphocyte ratio (NLR), admission National Institutes of Health Stroke Scale (NIHSS) score, the Alberta Stroke Program Early CT Score (ASPECTS), atrial fibrillation, and blood glucose. While tree-based methods like XGBoost and LightGBM showed elevated training performance (e.g., XGBoost training AUC = 0.878) but significant drops in validation (AUC = 0.791), LR demonstrated optimal performance: robust training AUC (0.792), minimal validation degradation (AUC = 0.787). LR model was subsequently employed as classification method demonstrating optimal performance with (AUC = 0.777) in the test dataset. External validation confirmed LR’s stability (AUC = 0.797). SHAP analysis ranked NLR as the strongest predictor (followed by NIHSS/ASPECTS), with higher values increasing risk. Learning curves indicated no overfitting. A nomogram enabled individualized risk quantification.

Conclusion: A parsimonious 5-variable LR model robustly predicts 3-month post-IVT outcomes, combining clinical utility, interpretability, and generalizability. NLR-driven inflammation is critical to prognosis. This tool facilitates early high-risk patient identification for personalized intervention.

1 Introduction

Stroke remains as a global health crisis, ranking as the second leading contributor to mortality worldwide and the third leading cause of long-term disability (1). It imposes a substantial global health burden at both individual and societal levels, with the rate of disability burden increasing more rapidly in low-income and middle-income countries than in high-income countries (2–4). Acute ischemic stroke (AIS) is defined as sudden neurological dysfunction caused by focal brain ischemia lasting more than 24 h or accompanied by evidence of acute infarction on brain imaging, regardless of symptom duration, accounts for approximately 70% of incident stroke events (5, 6). Intravenous thrombolysis (IVT), administered within the 4.5-h time window, constitutes the gold-standard therapy for AIS, as universally endorsed by international guidelines (7). Despite advancements in endovascular thrombectomy, IVT remains the most accessible and efficacious reperfusion treatment for patients with AIS in clinical practice, owing to its widespread availability and relative simplicity of administration (8, 9). Despite its established efficacy in enhancing functional recovery, nearly half of IVT-treated patients experience unfavorable functional outcomes at 3 months. The modified Rankin Scale (mRS; range 0–6, where 6 indicates death), which integrates both motor and cognitive components and encompasses the constructs of impairment, disability, and handicap, is considered to be the most accepted outcome for assessing the efficacy of interventions of AIS (10, 11). Given the substantial neurological disability burden associated with AIS (12), developing validated predictive tools remains imperative for the early identification of patients susceptible to adverse functional outcomes. Such prognostic stratification would facilitate targeted interventions and optimized resource allocation, ultimately improving long-term neurological prognosis. However, many existing prediction models are limited by their suboptimal predictive accuracy and the lack of robust external validation, resulting in uncertain generalizability to broader, more diverse populations (13, 14). Furthermore, numerous tools rely on high-dimensional data—incorporating extensive imaging, genomic, or biomarker variables—which complicates clinical interpretation and practical implementation, thereby hindering widespread adoption (15, 16). The development of novel, concise, yet robust prediction tools is therefore essential to enhance clinical relevance and facilitate translation into routine care.

Inflammation and immune responses critically mediate all phases of cerebral ischemia pathogenesis. Following ischemic insult, the inflammatory response initiated promptly. Focal brain ischemia stimulates what is called sterile inflammation (17), trigger inflammatory signaling through the activation of microglia, which subsequently release pro-inflammatory cytokines and chemokines, thereby promoting robust pro-inflammatory cascades, propelling the pathophysiological progression (18, 19). Critically, ischemic microenvironments trigger local immune responses, characterized by inflammatory cytokine production, which exacerbate blood–brain barrier (BBB) permeability (20, 21). Notably, neutrophils are the earliest leukocytes recruited from peripheral blood into the brain (22, 23). Neutrophils induce neurotoxicity through multiple mechanisms such as the participation in thrombus formation and expansion, upregulation of matrix metalloproteinases, excessive generation of reactive oxygen species, and the release of neutrophil extracellular traps (NETs) (24–26). The subsequent increase in capillary permeability, disruption of the BBB, and cellular edema can collectively impair post-stroke revascularization and vascular remodeling, thereby adversely affecting stroke outcomes (27). Clinical studies demonstrated the early increase of peripheral neutrophils as an independent predictor of neurological deterioration and poor outcome (28, 29). In addition, acute central nervous system injury can induce a state of immunodepression by activating the sympathetic nervous system and hypothalamic–pituitary–adrenal axis, leading to elevated catecholamines and steroids that cause apoptosis and functional deactivation of peripheral lymphocytes (30). Lymphocytes serve as pivotal regulators of host defense, and their depletion markedly elevates susceptibility to infections. Clinical research data indicates that low lymphocyte counts constitute an independent predictor of infection risk in stroke patients (31, 32). Emerging evidence underscores the prognostic significance of these mechanisms of leukocyte-derived inflammation in post-stroke outcomes (27), with the neutrophil-to-lymphocyte ratio (NLR) validated as a predictive biomarker for clinical outcome in AIS patients receiving IVT (33). While baseline NLR has been established as an independent risk factor for outcomes including early neurological improvement (ENI), hemorrhagic transformation (HT), and mortality in AIS patients (34), the predominant focus of current NLR research on univariate assessments fails to capture synergistic interactions with clinical covariates (35). This methodological constraint impedes clinical translation, given that isolated biomarkers inherently lack the discriminative power for complex multifactorial outcomes.

Machine learning (ML), a rapidly advancing branch of artificial intelligence (AI), leveraging computational advances to uncover predictive insights from high-dimensional data, demonstrates growing utility in clinical stroke research (36, 37). ML offers substantial advantages in predictive accuracy and in identifying previously overlooked patient subgroups defined by unique physiological characteristics and prognostic trajectories. Various methodologies exist for feature selection within the domain of ML. Notably, the least absolute shrinkage and selection operator (LASSO) regression distinguishes itself from conventional stepwise regression techniques, which utilize forward or backward variable selection, by facilitating the effective screening of a greater number of variables even when the sample size is limited (38). Moreover, LASSO regression provides superior feature selection from high-dimensional biomedical datasets while addressing multicollinearity limitations inherent in conventional methods (39). As a result, LASSO-based ML methods demonstrate enhanced prognostic discrimination across diverse medical applications (40–42). Furthermore, to compensate for the scarcity of interpretable evidence supporting predictive models, we deployed the SHapley Additive exPlanations (SHAP) analysis. This technique offers intuitive, feature-level explanations, which are critical for validating model efficacy and building trust (43). Consequently, integrating complementary clinical variables using ML models and SHAP interpretation may optimize the prediction of unfavorable outcomes for post-IVT AIS patients.

Therefore, we aimed to develop and validate a machine learning model for predicting 3-month functional outcomes in IVT-treated AIS patients, incorporating interpretability analysis to elucidate predictor contributions to the model predictions.

More useless crapola at the link.