Predicting The Efficacy Of Washed Microbiota Transplantation In Patients With Crohn’s Disease Based On Clinical Characteristics And Metabolomics

Background and Aims: Washed microbiota transplantation(WMT)can effectively induce remission in some patients with Crohn’s disease(CD),while some patients respond poorly to it.We aimed to established machine learning models using variables of clinical characteristics and serum metabolomics to predict the efficacy of WMT in patients with CD.Methods: Patients with active CD were enrolled.The baseline serum metabolic profiles of those subjects were evaluated using liquid chromatography-mass spectrometry(LC-MS).Patients’ symptoms(abdominal pain and diarrhea),inflammatory indexes were also recorded before WMT.The efficacy(remission vs.non-remission)was defined using Harvey–Bradshaw Index(HBI).Identified the differential clinical characteristics and metabolites between the groups before WMT,then constructed the machine learning models with variables selected by Boruta method to predict the efficacy.Results: Overall,eighty-six patients were included for analysis.The symptoms of patients gradually improved and the inflammatory indexes significantly decreased within one week after WMT.Forty-seven patients(54.65%)achieved clinical remission at three-month after WMT.Patients with penetrating CD,perianal disease,high HBI score,low body mass index(BMI)and family history had a lower possibility to achieve clinical remission after WMT.There were 148 serum metabolites that were significantly different between the two groups before WMT.We constructed random forest(RF)classifiers using the most discriminatory clinical(pre-albumin,BMI,behavior＿B1,B3 and family history)or metabolic(Eucommin a,Cefetamet,Monomenthyl succinate,N-phenylacetylglutamine(pos),4-(1-Methylethyl)-2,6-dinitro-N,N-dipropylbenzenamine,N-phenylacetylglutamine(neg),Diisodecyl phthalate,3-thiomorpholine-carboxylic acid)factors,respectively,with the corresponding area under the curve(AUC)of 0.85 and 0.88.The RF model combined2 clinical factors(behavior＿B1,B3)and 9 metabolites had better predictive performance(AUC = 0.95).Conclusion: In this study,the constructed machine learning models using serum metabolites,clinical characteristics have good predictive performance of WMT efficacy in CD.The combination of clinical factors and serum metabolomics will better serve the clinical decision-making.