RESUMO
BACKGROUND: Infections associated with nephrotic relapses (NR) are often managed according to physician preferences. A validated prediction tool will aid clinical decision-making and help in rationalizing antibiotic prescriptions. Our objective was to develop a biomarker-based prediction model and a regression nomogram for the prediction of the probability of infection in children with NR. We also aimed to perform a decision curve analysis (DCA). METHODS: This cross-sectional study included children (1-18 years) with NR. The outcome of interest was the presence of bacterial infection as diagnosed using standard clinical definitions. Total leucocyte count (TLC), absolute neutrophil count (ANC), quantitative C-reactive protein (qCRP), and procalcitonin (PCT) were the biomarker predictors. Logistic regression was used to identify the best biomarker model, followed by discrimination and calibration testing. Subsequently, a probability nomogram was constructed and DCA was done to determine the clinical utility and net benefits. RESULTS: We included 150 relapse episodes. A bacterial infection was diagnosed in 35%. Multivariate analysis showed the ANC + qCRP model to be the best predictive model. This model displayed excellent discrimination (AUC: 0.83), and calibration (optimism-adjusted intercept: 0.015, slope: 0.926). A prediction nomogram and web-application was developed. The superiority of the model was also confirmed by DCA in the probability threshold range of 15-60%. CONCLUSIONS: An ANC-based and qCRP-based internally validated nomogram can be used for the prediction of probability of infection in non-critically ill children with NR. Decision curves from this study will aid in the decision-making of empirical antibiotic therapy, incorporating threshold probabilities as a surrogate of physician preference. A higher resolution version of the Graphical abstract is available as Supplementary information.
