Machine learning–based predictive model for post-stroke cardiorespiratory fitness

Q: What are the key findings of this study?

Random Forest machine learning model classified low versus high cardiorespiratory fitness post-subacute stroke with an AUC of 0.7905, outperforming other models in clinical utility.

Q: What is the clinical evidence from this study?

Population: Adult patients with subacute stroke (1-6 months) and hemiplegia with Brunnstrom stage ≥3, functional ambulation category ≥2, adequate cognition and communication, without unstable cardiovascular or severe comorbidities (n=114). Intervention: Random Forest-based machine learning model vs. Other machine learning models including SVM, elastic net, logistic regression, multilayer perceptron, LightGBM, decision tree, k-nearest neighbors, extreme gradient boosting. Primary outcome: Classification accuracy of concurrent cardiorespiratory fitness status defined by VO2peak threshold <15 mL/(kg·min) vs ≥15 mL/(kg·min) (AUC 0.7905).

Zeyan Zhang; Zhixuan Duan; HuiQun Wu; Guodong Wang; Hua Ling; Han Gong; Xinao Mao; Xiaoxia Du

doi:10.1186/s12883-026-04737-4

February 18, 2026BMC NeurologyOpen Access

Machine learning–based predictive model for post-stroke cardiorespiratory fitness

View Full Paper

Ask AI

Bookmark

Share

Key result

Random Forest model outperforms others in classifying post-stroke cardiorespiratory fitness with an AUC of ~0.79.

AUC 0.7905
n=114

Figures

Why the study?

Reduced cardiopulmonary fitness is common in subacute stroke and may limit functional recovery, making timely classification of fitness status valuable to guide rehabilitation when CPET is unavailable.

Can machine learning models accurately classify concurrent cardiorespiratory fitness status in patients with subacute stroke using routinely collected clinical variables?

Structured PICO

Can machine learning models accurately classify concurrent cardiorespiratory fitness status in patients with subacute stroke using routinely collected clinical variables?

P

Population

114 patients with subacute stroke (duration 1-6 months, first-ever onset, single lesion, hemiplegia with Brunnstrom stage ≥3 in affected lower limb, FAC ≥ grade 2), hospitalized from July 2023 to January 2025 at a single center in China.

I

Intervention

Machine learning models (including Random Forest, Support Vector Machine, and Elastic Net) using 6 routinely collected clinical features (age, FMA score, 6MWT distance, sex, lower extremity FMA score, and FAC) to classify cardiorespiratory fitness.

O

Outcome

Classification of concurrent cardiorespiratory fitness status (low CRF defined as VO2peak <15 mL/(kg·min) vs high CRF defined as VO2peak ≥15 mL/(kg·min)) measured by Cardiopulmonary Exercise Testing (CPET).surrogate

Main Result

Effect estimate: AUC 0.7905

Limitations

Study sample limited to single center in China, which may limit generalizability
Relatively small sample size (114 patients)
CRF classification based on dichotomized VO2peak threshold which may reduce information and increase label instability near cutoff
No external validation cohort reported
Model performance measures limited primarily to AUC and decision curve analysis, lacking other calibration statistics
Single-centre study with a relatively small sample size
Outcome defined by dichotomizing VO2peak at a pragmatic threshold, which can reduce information
Models assessed using internal validation only
Binary definition may not capture more nuanced CRF stratification

Implication

Machine learning models, particularly Random Forest, can accurately classify cardiorespiratory fitness in subacute stroke patients using routine clinical variables, supporting clinical triage when CPET is unavailable.

Abstract

Reduced cardiopulmonary fitness (CRF) is common in subacute stroke and may limit functional recovery. Timely classification (risk stratification) of CRF status during the subacute phase may help guide rehabilitation when CPET is not readily available. 6 relevant features were selected using the Boruta algorithm. Subsequently, nine machine learning (ML) models were developed and evaluated, including logistic regression (LR), elastic net (EN), k-nearest neighbors (KNN), decision tree (DT), extreme gradient boosting (XGB), support vector machine (SVM), random forest (RF), multilayer perceptron (MLP), and Light Gradient Boosting Machine (LightGBM). A total of 114 patients with subacute stroke were included in this study. Among the nine models developed for classifying CRF, RF, SVM, and EN demonstrated the highest discriminative performances, with areas under the receiver operating characteristic (ROC) curve (AUC) of 0.7900, 0.7952, and 0.7905, respectively. Of these models, RF exhibited the greatest clinical applicability. The most important features contributing to classification included age, FMA score, 6MWT distance, sex, lower extremity FMA score, and FAC. ML models—particularly RF—most accurately classify concurrent CRF status in subacute stroke using routinely collected variables, supporting clinical triage when CPET is unavailable.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Zhang et al. (2026) studied Adult patients with subacute stroke (1-6 months) and hemiplegia with Brunnstrom stage ≥3, functional ambulation category ≥2, adequate cognition and communication, without unstable cardiovascular or severe comorbidities (n=114). Random Forest-based machine learning model vs. Other machine learning models including SVM, elastic net, logistic regression, multilayer perceptron, LightGBM, decision tree, k-nearest neighbors, extreme gradient boosting was evaluated on Classification accuracy of concurrent cardiorespiratory fitness status defined by VO2peak threshold <15 mL/(kg·min) vs ≥15 mL/(kg·min) (AUC 0.7905). Random Forest machine learning model classified low versus high cardiorespiratory fitness post-subacute stroke with an AUC of 0.7905, outperforming other models in clinical utility.

synapsesocial.com/papers/69994bdd873532290d01fed7 https://doi.org/10.1186/s12883-026-04737-4

View Full Paper

Ask AI

Bookmark

Share

Machine learning–based predictive model for post-stroke cardiorespiratory fitness

Figures

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Machine learning–based predictive model for post-stroke cardiorespiratory fitness

Figures

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading