Kidney Week

Abstract: PUB141

Machine Learning-Based Approaches to Characterizing Patients at High Risk of Progression in IgAN: A Scoping Review

Session Information

• Publication Only

Category: Glomerular Diseases

• 1402 Glomerular Diseases: Clinical, Outcomes, and Therapeutics

Authors

• Radhakrishnan, Jai, Columbia University, New York, New York, United States

Jai Radhakrishnan, MD, MBBS, MS, FASN

• Bagaria, Anurag Kumar, Novartis Pharma, Hyderabad, India

Anurag Kumar Bagaria, MSc, PhD

• Barbour, Sean, The University of British Columbia, Vancouver, British Columbia, Canada

Sean Barbour, MD, MSc

• Caravaca-Fontan, Fernando, Hospital Universitario 12 de Octubre, Madrid, Community of Madrid, Spain

Fernando Caravaca-Fontan, MD, PhD

• Crook, Jonathan, Bedrock Healthcare Communications Limited, Camberley, England, United Kingdom

Jonathan Crook, PhD

• Fakhouri, Fadi, Centre Hospitalier Universitaire Vaudois, Lausanne, VD, Switzerland

Fadi Fakhouri, MD

• Hirst, Ceri, Novartis Pharmaceuticals, London, United Kingdom

Ceri Hirst, PhD

• Huber, Tobias B., Universitatsklinikum Hamburg-Eppendorf, Hamburg, HH, Germany

Tobias B. Huber, MD, FASN

• Joshi, Akshay Suresh, Novartis Pharma, Hyderabad, India

Akshay Suresh Joshi

• Przybysz, Raymond, Novartis Pharmaceuticals, East Hanover, New Jersey, United States

Raymond Przybysz, PhD

• Roberts, Ian, Oxford University Hospitals NHS Foundation Trust, Oxford, England, United Kingdom

Ian Roberts, MBChB

• Tang, Sydney, Division of Nephrology, Department of Medicine, The University of Hong Kong, Hong Kong, Hong Kong

Sydney Tang, MD, MBBS, PhD

• Trimarchi, Hernan, Hospital Britanico de Buenos Aires, Buenos Aires, Argentina

Hernan Trimarchi, MD, PhD, FASN

• Vivarelli, Marina, Bambino Gesu’ Children’s Hospital IRCCS, Rome, Lazio, Italy

Marina Vivarelli, MD

• Wada, Jun, Okayama University, Okayama, Japan

Jun Wada, MD, PhD

• Wong, Edwin Kwan Soon, National Renal Complement Therapeutics Centre, Newcastle upon Tyne Hospitals NHS Foundation Trust, Newcastle Upon Tyne, United Kingdom

Edwin Kwan Soon Wong, MBChB, PhD

• Zhang, Hong, Peking University First Hospital, Beijing, China

Hong Zhang, MD, PhD

Group or Team Name

• The Renal Council

Background

Patients with immunoglobulin A nephropathy (IgAN) carry risk of progression to kidney failure, however, the disease course is heterogeneous. Identifying patients at high risk of progression is essential to support clinical decision making. Machine Learning (ML)–based prognostic models have been proposed to improve individual risk stratification, however, their practical value, limitations and evidence gaps require evaluation. This review aims to summarize recent literature published on ML-based prognostic models in IgAN.

Methods

We conducted this review in accordance with PRISMA extension for scoping reviews (PRISMA-ScR). We searched Embase for publications between 2009 and 2025. Inclusion criteria focused on studies reporting ML-based prognostic models in IgAN. Reviews, case reports/series, and studies with a sample size <100 patients were excluded. STROBE and PROBAST-AI appraisal frameworks were applied. Data were extracted on variables used to develop ML models, variable feature importance, and overall model predictive performance.

Results

Of 2,505 records screened, 14 publications focusing on ML-based prognostic models met the inclusion criteria, after exclusion of disease management and individual biomarker-focused studies. Studies used diverse ML approaches, including decision trees, random forests, gradient boosting, support vector machines, and artificial neural networks. Two studies applied unsupervised clustering methods, including k-means. 8/14 publications received a ‘strong’ STROBE grading, with the remainder graded as ‘good’. Variables frequently identified as having high feature importance included estimated glomerular filtration rate and/or serum creatinine, age, proteinuria, hypertension, serum uric acid, serum albumin, and Oxford classification S and T scores.

Conclusion

The top predictor variables in terms of feature importance are well-established prognostic factors. Generalizability of these studies may be limited due to their retrospective design, specific patient populations, small sample sizes, potential sources of bias (e.g. non-randomized treatment), and limited external validation. Due to their objective nature, there is potential for integration of ML models into current prognostic approaches, however there are limitations to the current state of evidence supporting these.

Acknowledgment

The authors acknowledge Amal Haque (Bedrock Healthcare Communication), for providing medical writing assistance with this abstract. They also sincerely thank Serge Smeets, Ph.D. (Novartis Pharma AG, Basel, Switzerland) and Manasi Desai (Novartis Pharmaceuticals UK Ltd, London, UK) for providing valuable input and advice during the project. Novartis is the funding source of the scoping review project.

Funding

• Commercial Support – Novartis Pharmaceuticals