Scholarlink

Purpose <p>Sepsis is a severe health condition caused by an exaggerated response of the body to an infection that leads to organ failure and the death of individuals, accounting for approximately one-fifth of global mortality. The early detection of sepsis remains a medical challenge due to the heterogeneity in the source of infection. The lack of a specific diagnosis and early predictions hinder proper clinical treatment, thereby resulting in an increased mortality rate. The diagnosis of sepsis is particularly challenging due to a broad and complex set of factors such as tachycardia, shortness of breath, fever, pre-existing comorbidities, among others. Sepsis prediction has garnered special attention owing to the abundance of available data and the use of machine learning models. Such conditions eventually generate a large number of attributes, and attempting to find patterns among them often poses a significant challenge.</p> Methods <p>In this context, the present research proposes to identify potential evolutionary algorithms to determine statistical relevance among these attributes and, based on this, reduce the dimensionality of the Database. Furthermore, the study proposes a hybrid model, composed of a committee of classifiers for sepsis detection.</p> Results <p>As a result, the optimization model showed that the BRKGA algorithm demonstrated better results. Then, a hybrid model composed of Random Forest, MLP and Logistic Regression was developed, both acting in a classification committee on an optimized basis that achieved an accuracy of 66.8%; Kappa: 0.336; Sensitivity: 0.527; Specificity: 0.810; AUC: 0.719.</p> Conclusion <p>The results indicate that the use of the BRKGA algorithm combined with Random Forest demonstrates significant potential for optimizing predictive modeling in high-dimensional ICU data — offering a scalable framework for feature selection that, with adaptation to lower-data-density environments, could inform earlier detection systems beyond the ICU. The results indicate that the use of the BRKGA algorithm combined with Random Forest demonstrates significant potential in distinguishing sepsis risk within the ICU context, with moderate sensitivity and strong specificity — suggesting its potential as a ‘rule-in’ tool to augment, not replace, clinical vigilance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Early prediction of sepsis in intensive care units: A comparative analysis based on optimization techniques and committees

  • 期刊论文

  • 发表日期：

    2025年10月10日

查看全文
  • André Luiz Vale de Araújo,
  • Flavio Secco Fonseca,
  • Arianne Sarmento Torcate,
  • Ana Clara Gomes da Silva,
  • Flavio Monteiro de Oliveira Júnior,
  • José Carlos da Silva Júnior,
  • Dayane Aparecida Gomes,
  • Juliana Carneiro Gomes,
  • Maíra Araújo de Santana,
  • Clarisse Lins de Lima,
  • Wellington Pinheiros dos Santos

摘要

Purpose

Sepsis is a severe health condition caused by an exaggerated response of the body to an infection that leads to organ failure and the death of individuals, accounting for approximately one-fifth of global mortality. The early detection of sepsis remains a medical challenge due to the heterogeneity in the source of infection. The lack of a specific diagnosis and early predictions hinder proper clinical treatment, thereby resulting in an increased mortality rate. The diagnosis of sepsis is particularly challenging due to a broad and complex set of factors such as tachycardia, shortness of breath, fever, pre-existing comorbidities, among others. Sepsis prediction has garnered special attention owing to the abundance of available data and the use of machine learning models. Such conditions eventually generate a large number of attributes, and attempting to find patterns among them often poses a significant challenge.

Methods

In this context, the present research proposes to identify potential evolutionary algorithms to determine statistical relevance among these attributes and, based on this, reduce the dimensionality of the Database. Furthermore, the study proposes a hybrid model, composed of a committee of classifiers for sepsis detection.

Results

As a result, the optimization model showed that the BRKGA algorithm demonstrated better results. Then, a hybrid model composed of Random Forest, MLP and Logistic Regression was developed, both acting in a classification committee on an optimized basis that achieved an accuracy of 66.8%; Kappa: 0.336; Sensitivity: 0.527; Specificity: 0.810; AUC: 0.719.

Conclusion

The results indicate that the use of the BRKGA algorithm combined with Random Forest demonstrates significant potential for optimizing predictive modeling in high-dimensional ICU data — offering a scalable framework for feature selection that, with adaptation to lower-data-density environments, could inform earlier detection systems beyond the ICU. The results indicate that the use of the BRKGA algorithm combined with Random Forest demonstrates significant potential in distinguishing sepsis risk within the ICU context, with moderate sensitivity and strong specificity — suggesting its potential as a ‘rule-in’ tool to augment, not replace, clinical vigilance.