Abstract: TH-PO0719
Glomerular C3 Staining Is an Independent Predictor of Renal Survival and eGFR Slope in IgAN: Findings from the UK RaDaR Cohort and Its Nested GPT-4 Data Extraction Study
Session Information
- Glomerular Innovations: Artificial Intelligence, Multiomics, and Biomarkers
November 06, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Glomerular Diseases
- 1402 Glomerular Diseases: Clinical, Outcomes, and Therapeutics
Authors
- Masoud, Sherry, UK National Registry of Rare Kidney (RaDaR), Bristol, Bristol, United Kingdom
- Pitcher, David, UK National Registry of Rare Kidney (RaDaR), Bristol, Bristol, United Kingdom
- Shavick, Alex Matthew, Fondazione Human Technopole, Milan, Lombardy, Italy
- Kim, Yunsoo, University College London, London, England, United Kingdom
- Wong, Katie, UK National Registry of Rare Kidney (RaDaR), Bristol, Bristol, United Kingdom
- Levine, Adam P., University College London, London, England, United Kingdom
- Barratt, Jonathan, University of Leicester, Leicester, England, United Kingdom
- Gale, Daniel P., University College London, London, England, United Kingdom
- Roberts, Ian, University of Oxford, Oxford, England, United Kingdom
Background
Complement plays a key role in IgAN pathogenesis, but the prognostic significance of glomerular C3 deposition remains uncertain. This study investigates whether C3 deposition, assessed by immunofluorescence, is associated with eGFR slope and renal survival, using manually extracted data from diagnostic reports. As a secondary aim, we investigated the performance of GPT-4o (via Azure) at extracting pathology data using the manually curated dataset as ground truth.
Methods
UK RaDaR participants with IgAN (2010–2024) and eGFR ≥ 30 ml/min/1.73m2 at biopsy were included. Renal survival was defined as absence of kidney failure (eGFR<15 for ≥4 weeks, or kidney replacement therapy) or death. Renal survival analyses were conducted using Cox Regression and manually extracted pathology data. Covariates were MEST-C scores (M, E, S: 0 vs 1; T, C: 0 vs ≥1), C3 staining (<2+ vs ≥2+), age, sex, eGFR and uPCR. eGFR slope (covariates: age, sex, eGFR, uPCR) was estimated using linear mixed models. For the secondary aim, multi-step and chain-of-thought prompting were applied (train n=100, test n=467).
Results
567 patients had adequate biopsies (≥8 glomeruli). Mean age was 42 years (SD 16). Median eGFR was 58ml/min/1.73m2 (IQR 42–82) and uPCR 159mg/mmol (64–302). Median follow-up was 8.2 years (3.3–9.4). 366/567 (65%) had C3 staining reported as ≥2+. C3 (HR 2.24; 95% CI 1.27–3.94), eGFR (HR 0.82 per 5ml, 0.76–0.89), and uPCR (HR 1.12 per 50mg/mmol; 1.07–1.17) were all independently associated with renal survival. Adjusted 5-year eGFR slope was –2.5 vs –4.7ml/min/1.73m2 in patients with C3 <2+ vs ≥2+ (p=0.04). For the GPT-4 analysis precision, recall, F1 scores were ≥0.95 when MEST scores were explicitly reported, ≥0.88 for C3, and ≥0.92 for C or E scores inferred from descriptive text. For text-inferred M, S, and T scores, values were ≥0.80.
Conclusion
In this IgAN cohort, we demonstrate that GPT-4o can effectively extract selected pathology data from free-text reports. Furthermore, we use real-world data to show that C3 staining ≥2+ is strongly associated with worse renal survival and steeper eGFR decline. This association is independent of clinical factors (eGFR, uPCR, age, sex) and histological severity (MEST-C scores).