Objectives
Avian influenza such as H7N9 is currently a major global public health risk, and at present, there is a lack of relevant diagnostic and treatment markers.
Methods
We collected plasma samples from 104 confirmed H7N9 patients, 31 of whom died. Plasma metabolites were detected by UHPLC-HRMS, and a survival prediction model based on metabolites was constructed by machine learning models.
Results
A total of 1536 metabolites were identified in the plasma samples of H7N9 patients, of which 64 metabolites were up-regulated and 35 metabolites were down-regulated in the death group. The enrichment analysis of Tryptophan metabolism, Porphyrin metabolism and Riboflavin metabolism were significantly up-regulated in the death group. We found that most Lipids and lipid?like molecules were down-regulated in the death group, and Organoheterocyclic compounds were significantly up-regulated in the death group. A machine learning model was constructed for predicting mortality based on Porphobilinogen, 5-Hydroxyindole-3-acetic acid, L-Kynurenine, Biliverdin, and D-Dimer. The AUC on the test set was 0.929.
Conclusions
We first revealed the plasma metabolomic characteristics of H7N9 patients and found that a machine learning model based on plasma metabolites could predict the risk of death for H7N9 in the early stage of admission.