Abstract: TH-OR004
Multimodal Prediction of Steroid Resistance in IgAN Using CureGN and Taiwan Kidney Biopsy Pathology and Electronic Health Record Data
Session Information
- Advances in Renal Pathology and Laboratory Medicine
October 22, 2026 | Location: Room 605, Convention Center
Abstract Time: 05:00 PM - 05:10 PM
Category: Pathology and Lab Medicine
- 1700 Pathology and Lab Medicine
Authors
- Chang, David R., China Medical University Hospital Department of Internal Medicine, Taichung City, Taiwan
- Lee, Tsung-Hua, Harvard Medical School Department of Biomedical Informatics, Boston, Massachusetts, United States
- Vremenko, Dmytro, Harvard Medical School Department of Biomedical Informatics, Boston, Massachusetts, United States
- Kuo, Chin-Chi, China Medical University Hospital Department of Internal Medicine, Taichung City, Taiwan
- Yu, Kun-Hsing, Harvard Medical School Department of Biomedical Informatics, Boston, Massachusetts, United States
Background
Glucocorticoids are a first-line treatment option for patients with progressive IgA nephropathy (IgAN), but steroid resistance exposes patients to substantial toxicity without clear renal benefit. At the time of biopsy, treatment decisions rely on clinical and histologic markers with limited ability to predict steroid response. A model that identifies patients at high risk for steroid resistance before treatment could support individualized treatment decisions, enrich clinical trial enrichment, and reduce avoidable treatment-related harm. We developed pathology-only, electronic health record (EHR)-only, and multimodal models to predict steroid resistance in IgAN patients.
Methods
We included adults with IgAN from CMUH and CureGN who received steroid monotherapy and had available baseline and follow-up data. We defined steroid resistance as persistent uPCR/uACR ≥0.5 g/g or dialysis/kidney failure within the response window of 16 weeks after steroid exposure. Pathology-only models used kidney biopsy whole slide image-derived pathology embeddings. EHR-only models used baseline demographic, laboratory, comorbidity, and clinical variables. Multimodal models integrated pathology embeddings with EHR features in a joint representation. Because the CMUH cohort was small, models were evaluated in the pooled CMUH-CureGN cohort using internal cross-validation, with performance reported overall and stratified by cohort.
Results
The analytic cohort included 36 CMUH patients and 232 CureGN patients with IgAN. Steroid resistance was present in 24 CMUH patients (66.7%) and 161 CureGN patients (69.4%), for a combined total of 185 steroid-resistant patients (69.0%) and 83 steroid-responsive patients (31.0%). In pooled internal cross-validation, multimodal models achieved the highest discrimination, with an AUROC of 0.727, compared with 0.677 for pathology-only models and 0.670 for EHR-only models.
Conclusion
Pathology and EHR features may provide complementary information for predicting steroid resistance in IgAN. Multimodal modeling offers a framework for integrating biopsy-derived pathology signal with baseline clinical data, but the current sample size limited external validation across cohorts. Larger multi-cohort datasets are needed to assess model transportability and support diagnosis-specific prediction tools for treatment response in glomerular disease.
Acknowledgment
We thank the CureGN Pathology Committee and Digital Pathology Working Group for supporting access to pathology annotations and whole slide image resources.
Funding
- Other NIH Support