Development and validation of a nomogram for predicting treatment failure in culture-negative peritoneal dialysis-associated peritonitis.

BACKGROUND

Culture-negative peritoneal dialysis-associated peritonitis (CNPDP) carries a high risk of treatment failure but lacks validated prediction tools. This study aimed to develop and validate a clinical nomogram for individualized risk assessment of treatment failure in CNPDP patients.

METHODS

In this multicenter retrospective study, 288 CNPDP patients treated at Jining Medical University Affiliated Hospital (2013-23) were randomly allocated to training (n = 173) and internal validation (n = 115) cohorts. An independent external cohort (n = 103) from Zaozhuang Municipal Hospital and Heze Municipal Hospital assessed generalizability. First, we used Random Forest to estimate missing data for variables with <30% missing values. Then, we used LASSO regression to analyze 32 candidate predictors. These predictors covered areas like patient demographics, clinical scores and lab test results. The final multivariate logistic regression model was visualized as a clinical nomogram. Performance was rigorously evaluated through area under receiver operating characteristic curve (AUC), calibration plots and decision curve analysis. The primary endpoint was composite treatment failure (catheter removal or peritonitis-related mortality ≤30 days).

RESULTS

LASSO identified five independent predictors: effluent white blood cell count on Day 3 (Eff_WBC_D3), serum albumin (ALB), total cholesterol (TC), magnesium (Mg) and phosphorus (P). The nomogram achieved excellent discrimination: training cohort AUC = 0.897 (95% confidence interval 0.817-0.978), internal validation AUC = 0.861 (0.770-0.952) and external validation AUC = 0.849 (0.750-0.948) with minimal optimism (ΔAUC = 0.036). Eff_WBC_D3 demonstrated the strongest univariate predictive power (AUC = 0.830). Calibration curves showed optimal fit (Hosmer-Lemeshow P = .32), while decision curve analysis confirmed clinical utility across probability thresholds of 5%-50%. For bedside implementation, an interactive web tool was developed (https://liuliangmianhua.shinyapps.io/dynnomapp/).

CONCLUSION

This externally validated five-variable nomogram, deployed as a freely accessible online tool, offers a robust, practical tool for predicting treatment failure in CNPDP. Its integration of dynamic dialysate markers with routine laboratory data enables personalized early intervention and supports timely clinical decision-making.