Analyzing and predicting the risk of death in stroke patients using machine learning

Abstract

BackgroundStroke is an acute disorder and dysfunction of the focal neurological system that has long been recognized as one of the leading causes of death and severe disability in most regions globally. This study aimed to supplement and exploit multiple comorbidities, laboratory tests and demographic factors to more accurately predict death related to stroke, and furthermore, to make inferences about the heterogeneity of treatment in stroke patients to guide better treatment planning.MethodsWe extracted data from the Medical Information Mart from the Intensive Care (MIMIC)-IV database. We compared the distribution of the demographic factors between the control and death groups. Subsequently, we also developed machine learning (ML) models to predict mortality among stroke patients. Furthermore, we used meta-learner to recognize the heterogeneity effects of warfarin and human albumin. We comprehensively evaluated and interpreted these models using Shapley Additive Explanation (SHAP) analysis.ResultsWe included 7,483 patients with MIMIC-IV in this study. Of these, 1,414 (18.9%) patients died during hospitalization or 30 days after discharge. We found that the distributions of age, marital status, insurance type, and BMI differed between the two groups. Our machine learning model achieved the highest level of accuracy to date in predicting mortality in stroke patients. We also observed that patients who were consistent with the model determination had significantly better survival outcomes than the inconsistent population and were better than the overall treatment group.ConclusionWe used several highly interpretive machine learning models to predict stroke prognosis with the highest accuracy to date and to identify heterogeneous treatment effects of warfarin and human albumin in stroke patients. Our interpretation of the model yielded a number of findings that are consistent with clinical knowledge and warrant further study and verification.

Machine learning-based prognostication of mortality in stroke patients

“…As seen in prior literature, there is a well-established correlation between higher age and increased stroke mortality [ 26 , 58 , 59 ]. In this study, the mean age was 54 ± 13.5 years, which could be considered comparatively lower when assessed on regional and global scales [ 7 ].…”

“…SHAP analysis corroborated that BMI significantly contributes to the model's predictions. In general, obesity stands as a notable risk factor for stroke development and poor prognosis [ 21 , 26 ]. However, a considerable body of prior research has shown a counterintuitive correlation between BMI and stroke outcomes, reflecting a negative association [ 21 , [78] , [79] , [80] ].…”

“…Similarly, post-thrombectomy, the National Institutes of Health Stroke Scale (NIHSS) scores and the necessity for decompressive hemicraniectomy have emerged as crucial predictors of mortality [ 22 , 25 ]. Furthermore, certain studies have identified the body mass index's (BMI) relevance in predicting mortality among stroke patients [ 26 ]. Hospital-acquired infections, such as urinary tract infections and pneumonia, have been noted as significant factors in forecasting stroke prognosis, and mortality [ 12 , 27 ].…”

Utilizing machine learning to facilitate the early diagnosis of posterior circulation stroke

“…SHAP analysis reaffirmed the significant influence of BMI on the model’s classification performance. Typically, obesity is well-recognized as a leading risk factor for stroke and is linked with unfavorable prognoses [ 41 , 42 ]. A study conducted in Qatar in 2009 similarly identified obesity as one of the major risks associated with PCS, finding that patients with a BMI > 30 faced a heightened risk of PCS compared to those with lower BMIs [ 43 ].…”

Optimizing adjuvant treatment options for patients with glioblastoma

“…This is primarily due to the fact that a patient cannot simultaneously receive both treatments, and confounding variables are prevalent in observational studies ( 15 ). Benefitting from advances in machine learning (ML) and statistical theories, we can use balanced representation-based ( 16 ), tree-based ( 17 ), and conditional average treatment effect (CATE)-based ( 18 , 19 ) methods to counterfactually infer patients’ individual treatment effect (ITE) directly from observational data and thus attempt identify the relatively optimal treatment choice for specific individuals. With the development of deep learning (DL) and representation learning, novel techniques enable combining DL with survival models and learning balanced representations directly from the data to reason about unbiased counterfactual survival outcomes ( 20 ).…”