What question did this study set out to answer?

To develop and validate a model predicting mortality in pre-dialysis patients with chronic kidney disease.

March 28, 2026Open Access

WCN26-3202 Plasma Exchange for Anti-glomerular Basement Membrane Antibody Disease with Dialysis Dependency: A Case Series on Clinical Outcomes and Safety

Key Points

To develop and validate a model predicting mortality in pre-dialysis patients with chronic kidney disease.
Extracted data from 3,545 pre-dialysis patients in the MIMIC-IV database.
Employed LASSO regression for variable selection and Cox regression for survival analysis.
Utilized SHAP method to identify key predictors and construct a nomogram for mortality risk estimation.
Average patient age was 63 years, with 60.8% male.
Mortality rates increased over time, reaching 21.47% by Year 3.
The Cox regression model showed a C-index of 0.734 and strong predictive accuracy for mortality.

Abstract

simple model based on a small set of readily available predictors could enhance its practical applicability in clinical settings.Methods: Data from 3,545 pre-dialysis patients(defined as those with an eGFR <15 mL/min/1.73m 2 not yet on dialysis) were extracted from the MIMIC-IV database.The collected data includes demographic information, vital signs, laboratory indicators, comorbidities, survival status, and survival duration.Variables with a missing rate exceeding 30% were excluded, while those with a missing rate below 30% underwent multiple imputation.Of these patients, 1064 served as the validation group and 2081 served as the training cohort.LASSO regression was used for variable selection, and a Cox regression model was established.The SHapley Additive exPlanations (SHAP) method was then applied to identify the five most important predictors.Based on these predictors, a nomogram was constructed to estimate overall survival (OS).Model performance was assessed via calibration curves, decision curve analysis, time-dependent AUCs, and the concordance index (C-index).Results: The average age of the patients was 63.0 (52.0, 74.0) years, with a higher proportion of male patients (60.8%).All-cause mortality rates were 12.55%, 17.49%, and 21.47% at 1, 2, and 3 years, respectively.No significant differences were observed between the training and validation groups in clinical baseline characteristics, except for blood phosphorus levels (P < 0.05).LASSO regression was used to perform variable selection on 23 clinical features in the training set.When was set to .1se, the model exhibited optimal fit.A total of 11 potential prognostic-related variables were preliminarily selected: gender, age, Charlson comorbidity index, potassium, albumin, aniongap, chloride, white blood cell count(WBC), red blood cell count(RBC), platelet count, and blood phosphorus.Cox regression analysis of the 11 independent risk factors revealed all variables to be statistically significant (P < 0.05).SHAP analysis identified the most important variables: age, albumin, red blood cell count, blood phosphorus, and blood chloride.A Cox regression model was constructed based on the five identified risk factors, and a nomogram was developed to visualize predialysis mortality risk.In the validation cohort, the model achieved a C-index of 0.734 (95% CI: 0.709-0.760).ROC analysis showed that the areas under the curve (AUCs) for predicting 1, 2, and 3-year mortality in the validation cohort were 0.832, 0.799, and 0.792, respectively.Calibration curves indicated only minor deviations between predicted and observed probabilities, suggesting good agreement between predicted risks and actual outcomes.Conclusion: We developed and validated a simple five-variable Cox regression model to predict mortality in pre-dialysis CKD patients.The nomogram demonstrated good discriminative ability and calibration, suggesting that it could serve as a practical tool for individualized risk assessment and clinical decision-making in this population.I have no potential conflict of interest to disclose.I did not use generative AI and AI-assisted technologies in the writing process.

Bookmark

View Full Paper