# Hospital Readmission in Stroke Survivors in Social Vulnerability: Predictive Modeling with Machine Learning from the Perspective of the Chronic Conditions Care Model

## TL;DR

This study uses machine learning to predict hospital readmissions in stroke survivors facing social vulnerability, identifying key risk factors to improve care.

## Contribution

A novel predictive model for stroke readmissions is developed using the Chronic Conditions Care Model and machine learning techniques.

## Key findings

- The decision tree model achieved 92.45% accuracy in predicting readmissions.
- Falls, stroke history, caregiver presence, and sleep issues were key predictors of readmission.
- Logistic regression showed falls increased readmission risk by 235%.

## Abstract

Hospital readmission among stroke survivors is frequent, especially in contexts of social vulnerability, compromising recovery and overburdening health services. This study aimed to develop a predictive model of hospital readmission among socially vulnerable stroke survivors, based on the Chronic Conditions Care Model (CCCM). Machine learning algorithms were applied, specifically decision tree and logistic regression, with data split into training (70% and 80%) and testing (30% and 20%) sets. Analyses were conducted using Python, with accuracy evaluated through ROC curves, AUC, and the confusion matrix in Analyse-it®, adopting a 5% significance level. The decision tree with an 80/20 partition achieved an accuracy of 92.45%. The variables most associated with readmission were falls, time since the first stroke, presence of a caregiver, and difficulty sleeping. In logistic regression, falls increased the risk by 235%, ischemic stroke by 155%, complications by 153.53%, COVID-19 by 132%, and time since stroke by 11.5% per year. The model proved to be feasible and robust, with the decision tree standing out, highlighting its potential to support preventive strategies and enhance care management.