A kidney–brain-axis machine-learning framework for predicting 28-day mortality in chronic kidney disease with acute brain injury: derivation and external validation
Year: 2026
Type: article
Abstract: Chronic kidney disease (CKD) and acute brain injury (ABI) frequently co-exist in the ICU, yet no dedicated prognostic instrument has been validated for this population. Using MIMIC-IV ( n = 1,887) for derivation and eICU-CRD ( n = 791) for external validation, we derived three decision-oriented tools from a common 43-feature set: a calculator-free 10-feature bedside score (CKD-BI), a 10-variable LASSO-logistic nomogram, and seven machine-learning models benchmarked against SOFA, SAPS-II, APS-III and OASIS, with SHAP interpretability. The CKD-BI score stratified 28-day mortality (Low 3.0%, Intermediate 13.9%, High 36.9%; OR 18.92) and outperformed all four legacy scores (internal CV AUC 0.7937; external 0.7310). The nomogram enabled hand-computable probabilities (external AUC 0.7445), close to the ensemble model (0.7768). Random forest achieved the highest internal test AUC (0.8011; ΔAUC + 0.117, P < 0.001). Glasgow Coma Scale (GCS) total minimum was the dominant predictor (mean |SHAP| 1.04). The kidney–brain axis operated through parallel additive pathways; magnesium lacked independent prognostic value after adjustment (OR 0.87/SD, P = 0.226) whereas sodium retained a directionally reversed association (OR 1.21/SD, P = 0.002). These complementary tools offer transparent risk stratification and individual probability estimation with performance comparable to ensemble models; prospective validation with local recalibration is required before clinical use.
Source: Scientific Reports
Language: English
Topic: Acute Kidney Injury Research
Subfield: Nephrology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236730070
