Personalized federated learning for predicting disability progression in multiple sclerosis using real-world routine clinical data

Early prediction of disability progression in multiple sclerosis (MS) remains challenging despite its critical importance for therapeutic decision-making. We present the first systematic evaluation of personalized federated learning (PFL) for 2-year MS disability progression prediction, leveraging multi-center real-world data from over 26,000 patients. While conventional federated learning (FL) enables privacy-aware collaborative modeling, it remains vulnerable to institutional data heterogeneity. PFL overcomes this challenge by adapting shared models to local data distributions without compromising privacy. We evaluated two personalization strategies: a novel AdaptiveDualBranchNet architecture with selective parameter sharing, and personalized fine-tuning of global models, benchmarked against centralized and client-specific approaches. Baseline FL underperformed relative to personalized methods, whereas personalization significantly improved performance, with personalized FedProx and FedAVG achieving ROC-AUC scores of 0.8398 ± 0.0019 and 0.8384 ± 0.0014, respectively. These findings establish personalization as critical for scalable, privacy-aware clinical prediction models and highlight its potential to inform earlier intervention strategies in MS and beyond.

Authors

Institutions

Publication Details

Journal
npj Digital Medicine
Published
2025-07-24
DOI
https://doi.org/10.1038/s41746-025-01788-8
Citations
4
Primary Topic
AI in cancer detection
Type
article
Field-Weighted Citation Impact
9.92

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Personalized federated learning for predicting disability progression in multiple sclerosis using real-world routine clinical data

Koen de Gans, Neil Shuey, Martijn Oldenhof, José Luis Sánchez-Menoyo et al.
4 citations
npj Digital Medicine
AI in cancer detection
9.92
article

Personalized federated learning for predicting disability progression in multiple sclerosis using real-world routine clinical data

Koen de Gans, Neil Shuey, Martijn Oldenhof, José Luis Sánchez-Menoyo, Attila Sas, Edward De Brouwer, Jeannette Lechner‐Scott, Guy Laureys, Nevin John, Allan G. Kermode, Radek Ampapa, Antoine Passemiers, Ádám Arany, Tomáš Kalinčík, Mihaela Simu, Talal Al‐Harbi, Patrice H. Lalive, Alessandra Lugaresi, Tamara Castillo‐Triviño, William W. Carroll, Yves Moreau, Anneke van der Walt, Ángel Pérez Sempere, Jens Kühle, Maria Pia Amato, Julie Prévost, D. Decoo, Raed Alroughani, Valentina Tomassini, Cavit Boz, Izanne Roos, Csilla Rózsa, Pamela McCombe, Carmen Adella Sîrbu, Marzena J. Fabis‐Pedrini, Samia Joseph Khoury, Justin Garber, Liesbet M. Peeters, Ashkan Pirmani, Pierre Grammond, Bruce Taylor, Aysun Soysal, Daniele Spitaleri, Barbara Willekens, Francesco Patti, Alexandre Prat, Abdullah Rashid Al-Asmi, Todd A. Hardy, Bianca Weinstock‐Guttman, Riadh Gouider, Eva Havrdová, Ayşe Altıntaş, Sudarshini Ramanathan, Cecília Rajda, Tünde Csépány, Axel Faes, Vincent Van Pesch, Serkan Özakbaş, Claudio Marcello Solaro, Deborah Mason, Yolanda Blanco, Orla Gray, Joana Guimarães, Marija Cauchi, Bart Van Wijmeersch, Dana Horáková, Bhim Singhal, Oliver Gerlach, Elisabetta Cartechini, Katherine Buzzard, Davide Maimone, Helmut Butzkueven, Stella Hughes
article en
4 citations

Abstract

Early prediction of disability progression in multiple sclerosis (MS) remains challenging despite its critical importance for therapeutic decision-making. We present the first systematic evaluation of personalized federated learning (PFL) for 2-year MS disability progression prediction, leveraging multi-center real-world data from over 26,000 patients. While conventional federated learning (FL) enables privacy-aware collaborative modeling, it remains vulnerable to institutional data heterogeneity. PFL overcomes this challenge by adapting shared models to local data distributions without compromising privacy. We evaluated two personalization strategies: a novel AdaptiveDualBranchNet architecture with selective parameter sharing, and personalized fine-tuning of global models, benchmarked against centralized and client-specific approaches. Baseline FL underperformed relative to personalized methods, whereas personalization significantly improved performance, with personalized FedProx and FedAVG achieving ROC-AUC scores of 0.8398 ± 0.0019 and 0.8384 ± 0.0014, respectively. These findings establish personalization as critical for scalable, privacy-aware clinical prediction models and highlight its potential to inform earlier intervention strategies in MS and beyond.

npj Digital MedicineVol. 8(1)
Cliniques Universitaires Saint-Luc (BE), The Royal Melbourne Hospital (AU), University of Debrecen (HU), Box Hill Hospital (AU), Koç University (TR), University of Ulster (GB), University of Antwerp (BE), İzmir University of Economics (TR), Karadeniz Technical University (TR), The University of Western Australia (AU), Universidade do Porto (PT), University of Basel (CH), University of Miskolc (HU), Charles University (CZ), University of Szeged (HU), Royal Brisbane and Women's Hospital (AU), Harry Perkins Institute of Medical Research (AU), Concord Repatriation General Hospital (AU), Mater Dei Hospital (MT), Ghent University Hospital (BE), The Alfred Hospital (AU), Sir Charles Gairdner Hospital (AU), King Fahad Specialist Hospital (SA), University Hospital of Basel (CH), Christchurch Hospital (NZ), Royal Hobart Hospital (AU), Westmead Hospital (AU), Amiri Hospital (KW), Ente Ospedaliero Ospedali Galliera (IT), Cegep de Saint Jerome (CA), Bakırköy Psychiatric Hospital (TR), St Vincent's Hospital Melbourne (AU), Hunter Medical Research Institute (AU), Victor Babeș University of Medicine and Pharmacy Timișoara (RO), Queen Elizabeth II Medical Centre (AU), Hospital Jihlava (CZ), Biogipuzkoa Health Research Institute (ES), Antwerp University Hospital (BE), American University of Beirut Medical Center (LB), University Hospital of Geneva (CH), Azienda Ospedaliera S.Giuseppe Moscati (IT), Hospital Clínic de Barcelona (ES), Hospital de Galdakao (ES), Don Carlo Gnocchi Foundation (IT), Osakidetza (ES), Royal Victoria Hospital (GB), County Hospital (GB), Zuyderland Medisch Centrum (NL), Bombay Hospital (IN), Hospital General Universitario de Alicante Doctor Balmis (ES), Groene Hart Ziekenhuis (NL), Perron Institute for Neurological and Translational Science (AU), Alma (BE), Centre intégré de santé et de services sociaux de Chaudière-Appalaches (CA), Jacobs Institute (US), Jahn Ferenc Dél-Pesti Kórház és Rendelőintézet (HU), General University Hospital in Prague (CZ), South Eastern Health and Social Care Trust (GB), Hôpital Razi de La Manouba (TN), University of Florence (IT), Sultan Qaboos University (OM), University of Macerata (IT), Monash University (AU), Geneva College (US), Université de Montréal (CA), Neuroscience Research Australia (AU), Hasselt University (BE), University of Bologna (IT), KU Leuven (BE)
Amgen, Bristol-Myers Squibb, Sanofi, Biogen, Teva Pharmaceutical Industries, Alexion Pharmaceuticals, Celgene, Univerzita Karlova v Praze, American University of Beirut, Universität Basel, Agentschap Innoveren en Ondernemen, European Commission, Hunter Medical Research Institute, Monash University, Fonds Wetenschappelijk Onderzoek, Eisai, KU Leuven, Karadeniz Teknik Üniversitesi, Razi University, Sultan Qaboos University, Università degli Studi di Firenze, Università di Catania, Università di Bologna, Koning Boudewijnstichting, Universidade do Porto, Multiple Sclerosis International Federation, Universiteit Antwerpen, Debreceni Egyetem, Universiteit Hasselt, Vlaamse regering, H. Lundbeck A/S, Szegedi Tudományegyetem, Všeobecná Fakultní Nemocnice v Praze, Koç Üniversitesi Translasyonel Tıp Araştırma Merkezi, Koç Üniversitesi, EMD Serono, Allergan, Sanofi Genzyme, lékařská fakulta Univerzity Karlovy, Faculty of Medicine and Health, University of Sydney, Cliniques Universitaires Saint-Luc
Quality Education
Openalex Percentile: Top 2%
AI in cancer detection
9.92
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.