7. Development of a web application for predicting Asia Impairment Scale at discharge in spinal cord injury patients: a machine learning approach

Abstract

BACKGROUND CONTEXT

Precise ASIA Impairment Scale (AIS) prediction at discharge for spinal cord injury (SCI) patients is crucial for guiding treatments, indicating regenerative medicine, and rehabilitation. Machine learning (ML) models are promising to improve such prognostic accuracy and aid clinical decisions.

PURPOSE

We aimed to create an ML model that predicts discharge AIS, to identify predictive factors, and to integrate this model into a web application.

STUDY DESIGN/SETTING

A retrospective cohort study.

PATIENT SAMPLE

This study used data from a nationwide database in Japan, the Japan Rehabilitation Database (JARD), consisting of records from 1991 to 2015. JARD contains both the SCI patients admitted to the SCI center right after the injury and the SCI patients referred to a rehabilitation hospital following acute phase treatment. In total, 3,703 cases formed the study cohort.

OUTCOME MEASURES

N/A

METHODS

Patient demographics, SCI-specific characteristics, and neurological evaluations at admission were used for ML model training. Utilizing the PyCaret library for preprocessing and validating the models, the best-performing algorithm was selected based on R², accuracy, and the weighted Kappa coefficient. Shapley additive explanations (SHAP) were used to determine the contribution of individual variables to the model's predictions. Using the optimal ML model and Streamlit, a web application to predict AIS at discharge was deployed.

RESULTS

The study divided the dataset into 2,592 training cases and 1,111 testing cases. The best-performing model exhibited an R² of 0.869, an accuracy of 0.814, and a weighted Kappa of 0.940. Eleven significant variables were identified with SHAP, including AIS at admission, days from injury to admission, and the motor score of L3. Using the Streamlit library, this best-performing model was deployed as an open-access web application. (http://3.138.174.54:8502/)

CONCLUSIONS

The developed ML model accurately predicts the AIS at discharge, using 11 essential variables. It has been integrated into a publicly accessible web application.

FDA Device/Drug Status

This abstract does not discuss or include any applicable devices or drugs.