NEONATAL SOCIETY ABSTRACTS
Multivariate Network Analysis of Cerebral and Systemic Variables for Assessment of Injury Following Hypoxic Ischaemic Encephalopathy
Presented at the Neonatal Society 2017 Summer Meeting (programme).
Dablander M1, Mitra S2, Bale G1, Dinan M2, Uria-Avellanal C2, Price D3, Sokolska M3, Bainbridge A3, Kendall G2, Meek J1, Tachtsidis I1, Robertson NJ2
1 Medical Physics and bioengineering, University College London (UCL)
2 Department of Neonatology, Institute for Women’s Health, University College London (UCL)
3 Medical Physics and bioengineering, University College London Hospital (UCLH)
Background: Hypoxic ischaemic brain injury remains as a significant cause of neonatal mortality and morbidity, even after therapeutic hypothermia (TH). A better understanding of pathophysiological changes in cerebral metabolism and haemodynamics is critical for early assessment. We hypothesised that a novel mSultivariate network analysis of near infrared spectroscopy (NIRS) and systemic signals will identify infants with severe injury.
Methods: Ethical approval obtained from local REC. Relationships between arterial blood pressure (ABP) and NIRS signals (cytochrome-c-oxidase (CCO) and haemoglobin difference (HbD)) were assessed in 25 infants. For each neonate, the multivariate signal was transformed into a multilayer network. The topological properties of each multilayer network were investigated by computing structural descriptors from mathematical graph theory. 1H MRS derived Lac/NAA threshold of 0.3 (2) was used to identify infants with good (n=13) and poor (n=12) outcome.
Results: Significant differences were noted between the two outcome groups for the relationship between CCO and ABP in both average edge overlap (p=0.048) (fig. 1A) and mutual information analysis (p=0.039) (fig. 1B). Average edge overlap between HbD and ABP did not reveal a significant difference (p=0.146) although difference in mutual information was significant (p=009) between the groups. Individual network maps revealed the difference between good and poor outcome group (examples of CCO-ABP map in a good (fig. 2A) and a poor outcome infant (fig. 2B) are presented here).
Conclusion: Relationship between CCO and ABP revealed an increased correlation as well as an increased dependency between these signals for infants with severe brain injury, resulting from a loss metabolic reactivity probably indicating mitochondrial injury. HbD-ABP had increased dependency between themselves in severe group, but did not have strong correlation. This novel multivariate network analysis provided an appropriate mathematical approach to identify these physiological changes, which can be a useful assessment tool at cot side during TH.
Corresponding author: email@example.com
1. Lacasa et al., Scientific reports, 2014
2. Thayyil et al., Pediatrics, 2010