Explainable AI Predicts Stroke Patients' Discharge Destination: Open-Source Implementation of a Clinical Decision Support System

Seif AI Lab has open-sourced the code for its stroke discharge destination prediction study published in the PLOS ONE journal. Based on an explainable AI framework, it aims to provide clinicians with transparent decision support to help optimize discharge planning for acute stroke patients. This open-source project helps enhance clinical trust in AI tools, promotes the implementation of AI-assisted decision-making in medical scenarios, and is a noteworthy resource in the fields of medical AI and explainable machine learning.

Acute stroke is one of the leading causes of death and long-term disability globally. A patient's discharge destination (home, rehabilitation facility, long-term care facility) directly affects their recovery outcomes and quality of life. Traditional decision-making relies on doctors' experience and lacks systematic data support; while AI technology has potential, its "black box" nature poses challenges in interpretability and credibility. This study addresses this issue by building an explainable AI framework to assist in discharge decision-making.

• Data source: Public dataset from Harvard Dataverse, Complete Stroke Data Repository V 3.0 (author: Voura E., DOI:10.7910/DVN/OYNYB4, 2026), which contains comprehensive clinical information.
• Research publication: The results were published in PLOS ONE with the paper title An Explainable Artificial Intelligence Framework for Clinical Decision Support in Stroke Discharge Planning.

Core goal of the framework: Balance predictive accuracy and interpretability, assist rather than replace doctors' decisions. Importance of interpretability:

1. Clinical trust: Doctors need to understand the basis of AI recommendations to adopt them;
2. Patient communication: Help medical staff explain discharge plans to family members;
3. Quality audit: Facilitate hospitals' review of decision-making processes;
4. Bias detection: Identify potential biases in the model and correct them.

Core content of the open-source repository:

• Main analysis notebook: Plos_One__Discharge_Disposition__code.ipynb (includes data processing, model training and evaluation);
• Supplementary document: S1_Code_README.txt;
• Serialized models: final_macro_f1_model_pipeline.joblib, etc.;
• Dependencies and citations: requirements.txt, CITATION.cff. Model evaluation uses macro-averaged F1 score, which balances precision and recall, and values each category in medical data with class imbalance.

Clinical significance:

1. Early planning: Start discharge planning early after admission to improve resource allocation efficiency;
2. Resource optimization: Identify patients needing rehabilitation/long-term care in advance to arrange beds reasonably;
3. Family communication: Provide data support to facilitate discussions on discharge plans;
4. Quality improvement: Analyze predictive factors to identify opportunities for clinical process improvement. Value of research reproducibility: Verify results, test with local data, explore model architecture, and extend to other disease areas.

Limitations:

1. Research use: Not clinically validated, not suitable for direct patient care;
2. Data dependency: Performance is affected by the quality and representativeness of input data;
3. Generalization ability: Performance may decline under different population distributions. Ethical considerations:
4. Privacy protection: Strictly protect patients' sensitive health information;
5. Fairness: Ensure the model is fair to different populations (age, gender, race); 3.Human-AI collaboration: AI assists rather than replaces doctors, with final decision-making power resting with doctors.

This open-source project demonstrates the application potential of explainable AI in the medical field. It enhances clinical trust through transparent decision-making bases and promotes the implementation of AI-assisted decision-making. For researchers and developers in medical AI, explainable machine learning, or clinical decision support systems, it is a high-quality open-source resource worth in-depth study.