Machine learning-based prediction of clinical outcomes after first-ever ischemic stroke

Abstract

BackgroundAccurate prediction of clinical outcomes in individual patients following acute stroke is vital for healthcare providers to optimize treatment strategies and plan further patient care. Here, we use advanced machine learning (ML) techniques to systematically compare the prediction of functional recovery, cognitive function, depression, and mortality of first-ever ischemic stroke patients and to identify the leading prognostic factors.MethodsWe predicted clinical outcomes for 307 patients (151 females, 156 males; 68 ± 14 years) from the PROSpective Cohort with Incident Stroke Berlin study using 43 baseline features. Outcomes included modified Rankin Scale (mRS), Barthel Index (BI), Mini-Mental State Examination (MMSE), Modified Telephone Interview for Cognitive Status (TICS-M), Center for Epidemiologic Studies Depression Scale (CES-D) and survival. The ML models included a Support Vector Machine with a linear kernel and a radial basis function kernel as well as a Gradient Boosting Classifier based on repeated 5-fold nested cross-validation. The leading prognostic features were identified using Shapley additive explanations.ResultsThe ML models achieved significant prediction performance for mRS at patient discharge and after 1 year, BI and MMSE at patient discharge, TICS-M after 1 and 3 years and CES-D after 1 year. Additionally, we showed that National Institutes of Health Stroke Scale (NIHSS) was the top predictor for most functional recovery outcomes as well as education for cognitive function and depression.ConclusionOur machine learning analysis successfully demonstrated the ability to predict clinical outcomes after first-ever ischemic stroke and identified the leading prognostic factors that contribute to this prediction.

Explainable Risk Prediction of Post-Stroke Adverse Mental Outcomes Using Machine Learning Techniques in a Population of 1780 Patients

“…As such, we tested several ML algorithms, including the following: logistic regression, decision tree, gradient-boosted tree, random forest, XGBoost, CatBoost, AdaBoost, and LightGBM. Compared with recently published ML models that have either studied only PSD or PSA [ 24 , 25 , 26 ], our model attained higher accuracy for the combined PSAMO diagnosis ( Table 1 ).…”

“…There are limitations to conduct a one-to-one comparison against previous studies as our study is the first to study PSAMO, while previous studies have examined PSD or PSA separately [ 24 , 25 , 26 ]. Other studies have also utilized different evaluation metrics and ML algorithms that were not used in this study [ 24 , 26 ].…”

“…Other studies have also utilized different evaluation metrics and ML algorithms that were not used in this study [ 24 , 26 ]. The closest study that could be referenced is Fast et al’s [ 25 ] study. Fast et al’s best performing model is also GBT.…”

Reliability and Validity of the Brief Attention and Mood Scale of 7 Items (BAMS-7): A Self-Administered, Online, and Real-World Measure

“…Cognition and mood are impacted by numerous medical conditions (Armstrong & Okun, 2022; Bar, 2009; Eyre et al, 2015; Fast et al, 2023), lifestyle choices (Santos et al, 2014; Sarris et al, 2020; van Gool et al, 2007), healthy development and aging (Fernandes & Wang, 2018; Mather & Carstensen, 2005; Tomaszewski Farias et al, 2024; Yurgelun-Todd, 2007), and medications or other interventions (Keshavan et al, 2014; Koster et al, 2017; Reynolds et al, 2021; Skirrow et al, 2009). Conditions principally defined by impaired cognition – such as ADHD or mild cognitive impairment – are often associated with concomitant changes in mood status, either directly or indirectly (Chen et al, 2018; D’Agati et al, 2019; Ismail et al, 2017; Retz et al, 2012; Yates & Woods, 2013).…”

Developing machine learning models to predict multi-class functional outcomes and death three months after stroke in Sweden

“…Previous studies on stroke outcome prediction have focused on ‘unfavorable’ or ‘favorable’ functional outcomes, using 30-day mortality [ 5 , 6 ] or different cut-offs [ 4 – 6 , 10 , 11 ]. For instance, some studies have used a binary version of modified Rankin Scale (mRS) with two classes: good (mRS ≤2) and poor (mRS >2) outcomes [ 12 – 14 ]. Ordinal analysis of the three-month mRS is more useful, since it better captures the spectrum of outcomes after stroke [ 10 , 15 ].…”