Scholarlink

<p>The COVID-19 pandemic has underscored the need for adaptive, data-driven frameworks to support clinical decision-making in dynamic and uncertain healthcare environments. We introduce a finite-horizon Markov decision process (MDP) to optimize treatment strategies for COVID-19 patients. The model integrates real-world data from 1335 hospitalized patients, accounting for disease severity, comorbidities, and gender-specific risk profiles to provide personalized recommendations. To improve state-transition modeling, we employ a gated recurrent unit (GRU) neural network trained on longitudinal electronic health records. The MDP is solved efficiently via a discounted hierarchical backward induction (DHBI) algorithm, enabling effective decision-making in large, complex state spaces. The framework shows high concordance with physician-prescribed treatments, achieving agreement rates of 82% for male and 77% for female patients, and it significantly delays the onset of severe complications, demonstrating clinical benefit. By combining interpretability, adaptability, and real-world validation, this approach offers a scalable decision-support tool for precision treatment in COVID-19 care and other high-risk medical conditions.</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

Optimizing personalized COVID-19 treatment strategies using finite-horizon MDP

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月4日

查看全文
  • Bouchra El Akraoui,
  • Fatima Es-sabery,
  • Marwan Albahar,
  • Cherki Daoui,
  • Abdelhadi Larach

摘要

The COVID-19 pandemic has underscored the need for adaptive, data-driven frameworks to support clinical decision-making in dynamic and uncertain healthcare environments. We introduce a finite-horizon Markov decision process (MDP) to optimize treatment strategies for COVID-19 patients. The model integrates real-world data from 1335 hospitalized patients, accounting for disease severity, comorbidities, and gender-specific risk profiles to provide personalized recommendations. To improve state-transition modeling, we employ a gated recurrent unit (GRU) neural network trained on longitudinal electronic health records. The MDP is solved efficiently via a discounted hierarchical backward induction (DHBI) algorithm, enabling effective decision-making in large, complex state spaces. The framework shows high concordance with physician-prescribed treatments, achieving agreement rates of 82% for male and 77% for female patients, and it significantly delays the onset of severe complications, demonstrating clinical benefit. By combining interpretability, adaptability, and real-world validation, this approach offers a scalable decision-support tool for precision treatment in COVID-19 care and other high-risk medical conditions.