Plataformas não-invasivas metabolômica e biofotônica para detecção da sepse neonatal baseadas em biomarcadores urinários de recém-nascidos pré-termo de muito baixo peso

Abstract

Introduction: Neonatal sepsis is a severe disease that rapidly progresses to multiorgan failure, primarily affecting preterm newborns with very low birth weights (VLBWs). Diagnosis is challenging, as clinical signs in this population are nonspecific, laboratory tests are limited, and blood cultures have reduced accuracy, requiring invasive sample collection and extended time for results, thus hindering early diagnosis. In this context, there is a need to develop new diagnostic strategies for the noninvasive detection of neonatal sepsis, with timely result delivery. Objective: platforms for the detection of neonatal sepsis in very low birth weight preterm newborns, based on the identification of novel urinary biomarkers applied to artificial intelligence algorithms, through a biophotonic platform and metabolomics. Materials and methods: A longitudinal study was conducted in a neonatal intensive care unit from January 2015 to September 2019. Preterm newborns with a gestational age < 30 weeks and birth weight < 1,500 g were included. Neonates with and without neonatal sepsis were evaluated. Urine samples collected on the second day after birth were analyzed via Fourier transform infrared spectroscopy with attenuated total reflectance (ATR-FTIR) for portable identification in spectral vibrational mode and via high-resolution mass spectrometry (MS) coupled with high-performance liquid chromatography (HPLC) for metabolite identification. The spectra from both platforms were subsequently supported by artificial intelligence algorithms to assess diagnostic performance (accuracy, sensitivity, and specificity). Results: ATR-FTIR analysis of urine samples from 71 VLBW preterm newborns, including 35 with sepsis and 36 without sepsis, revealed that urinary vibrational modes related to carbohydrates, proteins, and lipids could discriminate between septic and nonseptic VLBW preterm newborns. The highest-performing classification used the random forest algorithm associated with preprocessing (second-order derivative with vector normalization – spectra at 3050--2800/1800--900 cm-1), yielding a sensitivity of 74%, specificity of 77%, and accuracy of 76%. Metabolomic analysis via HPLC-MS, which was based on urine samples from 77 VLBW preterm newborns, including 40 with sepsis and 37 without sepsis, revealed that eight urinary metabolites presented significant concentration differences between septic and nonseptic neonates, indicating alterations in amino acid and lipid metabolism during the early phase of sepsis. The best-performing classification used the neural network algorithm, which achieved a sensitivity of 80.43%, specificity of 97.06%, and accuracy of 87.50%. Conclusions: The potential of the portable ATR-FTIR platform combined with artificial intelligence for the use of urine for neonatal sepsis diagnostic screening was demonstrated. The noninvasive detection platform based on urinary metabolomics and artificial intelligence identified eight novel urinary metabolites and shows promise as an alternative in neonatal sepsis detection. Although further studies with larger sample sizes and multicenter designs are needed, we believe that these platforms can facilitate advances in the screening, diagnosis, and monitoring of neonatal sepsis, enabling timely and precise interventions that reduce antimicrobial resistance, costs, and complications associated with delayed sepsis treatment.