# Predictive factors of stigma in stroke patients based on logistic regression and decision tree mode

## TL;DR

This study identifies factors that predict stigma in stroke patients using logistic regression and decision tree models.

## Contribution

The study compares logistic regression and decision tree models for predicting stigma in stroke patients.

## Key findings

- Threat assessment is a risk factor for stigma, while social support and resilience are protective factors.
- The decision tree model identified psychological resilience as the most important factor affecting stigma.
- Logistic regression had slightly higher AUC and accuracy than the decision tree model.

## Abstract

Logistic regression and decision tree model were used to analyze the predictive factors of stigma in stroke patients, and to explore the application value of the two models.

This was a retrospective study. The data of 342 stroke patients were collected from Baoding No.1 Central Hospital from December 2023 to March 2024. Data were retrospectively retrieved from the hospital information and management system. The regression model and decision tree model of influencing factors of stroke patients’ sense of stigma were established, to analyze the influencing factors of the sense of stigma, and to compare the predictive effects, advantages and disadvantages of the two models.

Logistic regression analysis showed that threat assessment (OR=2.7761) was a risk factor for stigma, while irrelevant cognitive appraisal (OR=0.321), social support (OR=0.098) and resilience (OR=0.438) were protective factors. The results of the decision tree model showed that the patients’ psychological resilience was the most important factor affecting the sense of stigma, followed by social support and threat assessment. The AUC of the decision tree model and Logistic regression model were 0.854 and 0.880, respectively, and the accuracy were 78.7% and 79.6%, respectively.

Threat, irrelevant cognitive appraisal, social support and resilience might be the predictive factors of stigma in stroke patients. The AUC and accuracy of the decision tree model were slightly lower than that of the Logistic regression model.