Retrospective assessment of the predictors of neonatal and infantile cholestasis with and without liver failure: an experience from Southeast China

TL;DR

This study developed and validated age-specific prediction models to identify neonatal and infantile cholestasis with liver failure in Southeast China, aiming to improve early diagnosis and outcomes.

Contribution

The study introduces novel nomograms for predicting liver failure in neonatal and infantile cholestasis patients specific to Southeast China.

Findings

Neonatal nomogram includes six significant predictors with an AUC of 0.743 in development and 0.736 in external validation.

Infant nomogram includes three significant predictors with an AUC of 0.784 in development and 0.711 in external validation.

Calibration and decision curve analyses confirmed the models' clinical utility and consistency with real outcomes.

Abstract

Neonatal and infantile cholestasis with liver failure (LF) is a life-threatening condition. To identify predictive factors, it is essential to develop and validate novel nomograms for predicting neonatal and infantile cholestasis with LF separately in Southeast China.

The medical records of neonates and infants with cholestasis at Fujian Maternity and Child Health Hospital from April 27, 2012, to July 11, 2023, were retrospectively analyzed as the development cohort. An external validation cohort was assembled from Fujian Children’s Hospital during the same period. Univariate analysis was initially conducted on the relevant indices, then the least absolute shrinkage and selection operator was performed to assess independent predictive factors. Further multivariate logistic regression analysis was conducted to identify independent predictors and develop predictive nomograms. Area under the curve (AUC) of receiver operating characteristic, calibration curves and decision curve analysis (DCA) were used to evaluate and validate the model and subsequently confirmed with the external validation group.

A total of 1,793 neonates and 583 infants were included in the development cohort, and 374 neonates and 232 infants in the external validation cohort. The neonatal nomogram included six variables that were significant independent predictors of LF: gestational age (p = 0.00), high-density lipoprotein (p = 0.008), red cell distribution width-standard deviation (p = 0.00), C-reactive protein (p = 0.00), albumin/fibrinogen (p = 0.00) and aspartate aminotransferase/platelets (p = 0.00). In the infant group, three variables, including vomiting (p = 0.005), lactate dehydrogenase (p = 0.00) and the albumin/fibrinogen (p = 0.00), were significant independent predictors of LF and were included in the infant nomogram. In the development cohort, the nomograms predicted LF with AUC values of 0.743 and 0.784 in the neonatal and infant groups, respectively. In the external validation cohort, the nomograms had AUC values of 0.736 and 0.711 in the neonatal and infant groups, respectively. The Hosmer–Lemeshow test results indicated that there was no significant difference between the predicted and true values. Calibration curves confirmed the consistency of the predicted outcomes with the real outcomes, and DCA curves demonstrated potential benefits for all patients.

This study developed and externally validated age-specific models for predicting LF in cholestasis patients. These nomograms show good clinical utility and can help pediatricians identify LF cases early, potentially improving outcomes in Southeast China.

Figures