Renal interstitial inflammation predicts IgA nephropathy progression via multiple machine learning models

Renal interstitial inflammation (RII) is a frequent pathological feature in IgA nephropathy (IgAN), but its prognostic value remains uncertain. This study investigated the effect of RII on renal outcomes and developed a machine learning-based model incorporating RII for individualized prognosis.

We retrospectively analyzed 540 IgAN patients diagnosed by renal biopsy at Zhongda Hospital and the First People's Hospital of Huai'an (2012 - 2023). The endpoint was a ≥ 50% decline in eGFR or end-stage renal disease, with follow-up to June 2024. Predictors included demographics, clinical/laboratory parameters (blood tests, serum biochemistry, 24-hour urine protein), and histopathology (Oxford MEST-C and RII scores). Variable selection used random forest, extreme gradient boosting, artificial neural networks, and LASSO regression. A logistic regression model and nomogram were developed and validated internally and externally.

Of 540 patients (mean age 40.8 years; 50.6% male), 273 were in the derivation, 117 in the internal validation, and 150 in the external validation cohort. Patients with progression had lower baseline serum albumin (p = 0.023), lower estimated glomerular filtration rate (eGFR) (p < 0.001), and higher systolic blood pressure (SBP) and proteinuria (all p < 0.001). In multivariate analysis, RIIS1 (odds ratio (OR) 4.16, 95% CI 0.91 - 24.51, p = 0.048) and RIIS2 (OR 6.80, 95% CI 0.98 - 54.49, p = 0.039) independently predicted adverse outcomes. Use of renin-angiotensin-aldosterone system inhibitors was protective (OR 0.34, p = 0.026), while higher SBP increased risk (OR 1.04, p < 0.001). The nomogram achieved C-indices of 0.91, 0.90, and 0.92 in the derivation, internal, and external validation cohorts, respectively.

RII is an independent predictor of renal progression in IgAN. The developed model and nomogram may assist in individualized risk stratification.