A Machine Learning Approach to Differentiate Congenital and Transient Neonatal Hyperammonemia: A 10-Year Case Series

TL;DR

This paper explores using machine learning to distinguish between congenital and acquired neonatal hyperammonemia, improving early diagnosis and treatment.

Contribution

The novel use of machine learning to identify key predictors for early differential diagnosis of neonatal hyperammonemia is presented.

Findings

Machine learning can help differentiate congenital and acquired neonatal hyperammonemia.

Key predictors identified can guide clinical decision-making for timely treatment.

Early diagnosis improves patient survival and neurological outcomes.

Abstract

Elevated blood ammonia concentration, resulting from various hereditary and acquired conditions, can cause severe damage to the central nervous system, leading to increased rates of disability and infant mortality. In newborns, hyperammonemia is etiologically classified into two main categories: congenital, associated with inborn errors of the urea cycle or organic acidemias, and acquired, which arises secondary to other pathological conditions such as severe perinatal hypoxia, renal or hepatic failure, or intrauterine infections. Despite the differing etiologies, the clinical presentation is often non-specific and may include lethargy, hypotonia, feeding difficulties, respiratory distress, and seizures. This non-specificity frequently leads to initial misdiagnosis. Consequently, a thorough understanding of the pathogenesis, clinical features, and differential diagnosis of congenital versus acquired hyperammonemia is critical for pediatricians, neonatologists, and intensive care specialists. Timely initiation of treatment is paramount, as it directly impacts patient survival and long-term neurological outcomes. Our findings underscore the utility of machine learning in the early differential diagnosis of neonatal hyperammonemia, identifying key predictors that can guide clinical decision-making.

Figures