<?xml version="1.0" encoding="UTF-8"?>



<records>

  <record>
    <language>eng</language>
          <publisher>Oriental Scientific Publishing Company</publisher>
        <journalTitle>Biomedical and Pharmacology Journal</journalTitle>
          <issn>0974-6242</issn>
            <publicationDate>2026-11-16</publicationDate>
    
        <volume>19</volume>
        <issue>4</issue>

 
    <startPage></startPage>
    <endPage></endPage>

	    <publisherRecordId>74260</publisherRecordId>
    <documentType>article</documentType>
    <title language="eng">Personalized Clinical Decision Support in Multi-Hospital Acute and Chronic Care Using Privacy-Preserving Federated Transfer Learning</title>

    <authors>
	 


      <author>
       <name>Ruchi Gupta</name>

 
		
	<affiliationId>1</affiliationId>
      </author>
    

	 


      <author>
       <name>Sandeep Gupta</name>


		
	<affiliationId>2</affiliationId>

      </author>
    

	 


      <author>
       <name>Tadiwa Elisha Nyamasvisva</name>

		
	<affiliationId>3</affiliationId>
      </author>
    

	 


      <author>
       <name>Anupama Sharma</name>

		
	<affiliationId>4</affiliationId>
      </author>
    


	 


      <author>
       <name>Nitin Tyagi</name>

		
	<affiliationId>5</affiliationId>
      </author>
    


	 


      <author>
       <name>Ashif Ali</name>

		
	<affiliationId>6</affiliationId>
      </author>
    
    </authors>
    
	    <affiliationsList>
	    
		
		<affiliationName affiliationId="1">Postdoctoral Researcher, Kuala Lumpur University of Science and Technology, Malaysia and Faculty of Department of Information Technology, Ajay Kumar Garg Engineering College, Ghaziabad, U.P., India</affiliationName>
    

		
		<affiliationName affiliationId="2">Department of CSE, Sharda University, Greater Noida U.P., India and International Visiting Professor, Faculty of Engineering , Science and Technology, Kuala Lumpur University of Science and Technology, Malaysia</affiliationName>
    
		
		<affiliationName affiliationId="3">Faculty of Engineering, Science and Technology, Kuala Lumpur University of Science and Technology, Malaysia</affiliationName>
    
		
		<affiliationName affiliationId="4">Faculty of Department of Information Technology, Ajay Kumar Garg Engineering College, Ghaziabad, U.P., India</affiliationName>
    
		
		<affiliationName affiliationId="5">Postdoctoral Researcher, Kuala Lumpur University of Science and Technology, Malaysia & Department of CSE-AIML, Accurate Institute of Management and Technology, Greater Noida, U.P., India</affiliationName>
    
		
		<affiliationName affiliationId="6">Department of CSE, Aravali College of Engineering and Management, Faridabad, Haryana , India</affiliationName>
    
	  </affiliationsList>






    <abstract language="eng">Machine learning-based clinical decision support systems (CDS) are becoming more common in healthcare, but their implementation in a multi-hospital setting is limited by data privacy laws, institutional heterogeneity, and personalization requirements. Centralized learning approaches cannot align with healthcare data governance stipulations, whereas current federated learning (FL) frameworks exhibit a lack of robust convergence and deteriorate in performance when applied to non-independent and non-homogeneously distributed (non-IID) clinical data. We address these issues with a personalized federated transfer learning (FTL) model for clinical decision support in which population-level representation learning is kept apart from hospital-specific adaptation. All sites share one global backbone, while each hospital keeps its own adapter layers and prediction head on premises. Training combines proximal-regularized federated optimization with secure aggregation and differential privacy using privacy budgets ε = 1, 5, and 10 and δ = 1 × 10⁻⁵.

The proposed framework is tested on simulated multi-hospital partitions derived from two publicly available clinical datasets for two clinically distinct tasks: acute sepsis survival prediction and chronic kidney disease (CKD) classification. In the sepsis task, the proposed approach achieves an AUROC of 0.69 ± 0.01 (95% CI: 0.68–0.70; p &lt; 0.001 vs. FedAvg), which is better than standard federated baselines of FedAvg (0.43 ± 0.09) and FedProx (0.48 ± 0.12), and the removal of hospital-level personalization decreases performance (AUROC = 0.53 ± 0.14). The framework achieves an average AUROC of 0.86 ± 0.04 on the CKD task, which is more accurate and less variable across inter-hospital settings than local training and global federated strategies. All findings suggest that explicit personalized federated transfer learning can support potentially deployable clinical inference systems in heterogeneous medical organizations without disclosing raw patient information. Key limitations include the use of virtual hospital partitions rather than real multi-institutional data, the absence of real-world latency and workflow evaluation, and the need for external validation before clinical deployment.</abstract>

    <fullTextUrl format="html">https://biomedpharmajournal.org/vol19no4/personalized-clinical-decision-support-in-multi-hospital-acute-and-chronic-care-using-privacy-preserving-federated-transfer-learning/</fullTextUrl>

<keywords language="eng">

      
        <keyword>Clinical decision support systems</keyword>
      

      
        <keyword> Federated transfer learning</keyword>
      

      
        <keyword> Privacy-preserving machine learning</keyword>
      

      
        <keyword> Personalized healthcare</keyword>
      

      
        <keyword> Sepsis dataset</keyword>
      
</keywords>
  </record>
</records>