P711 Stool microbiome communities predict remission in pediatric Crohn’s disease patients even after start of treatment
VerburgtMD, C.(1);Dunn, K.A.(2);Bielawski, J.P.(3);Otley, A.R.(4);Heyman, M.B.(5);Sunseri, W.(6);Shouval, D.(7);Levine, A.(8);de Meij, T.(1);Hyams, J.S.(9);Denson, L.A.(10);Kugathasan, S.(11);Benninga, M.A.(1);de Jonge, W.J.(12);Van Limbergen, J.E.(1);
(1)Emma Children’s Hospital- Amsterdam University Medical Centres, Department of Paediatric Gastroenterology and Nutrition, Amsterdam, The Netherlands;(2)Dalhousie university, Department of Biology, Halifax, Canada;(3)Dalhousie university, Department of Mathematics and Statistics, Halifax, Canada;(4)Dalhousie university, Department of Paediatrics, Halifax, Canada;(5)Department of Paediatrics, UCSF Benioff Children’s Hospital- University of California, San Francisco, United States;(6)UPMC Children’s Hospital of Pittsburgh, Department of Paediatrics, Pittsburgh, United States;(7)Schneider Children’s Medical Center of Israel, Institute of Gastroenterology- Nutrition and Liver Diseases, Petah Tikva, Israel;(8)Wolfson Medical Centre- Holon- Israel- Sackler School of Medicine- Tel Aviv University, Paediatric Gastroenterology and Nutrition Unit, Tel Aviv, Israel;(9)Connecticut Children’s Medical Center, Division of Digestive Diseases- Hepatology and Nutrition, Hartford, United States;(10)Cincinatti Children's Hospital Medical Center and the University of Cincinnati College of Medicine, Department of Paediatrics, Cincinnati, United States;(11)Emory University, Department of Paediatrics, Atlanta, United States;(12)Amsterdam University Medical Centers- University of Amsterdam, Tytgat Institute for Liver and Intestinal Research- Amsterdam Gastroenterology Endocrinology and Metabolism, Amsterdam, The Netherlands;
Background
Early relapse in children with Crohn’s Disease (CD) after reaching remission is associated with a more severe disease course that significantly impairs quality of life. We have previously shown that a Bayesian approach predicted clinical course in children with CD following the first year after diagnosis with high accuracy when ensuring samples were truly treatment-naïve. Here, we aimed to assess the impact on the accuracy when taking a broader timeframe of stool collection to baseline, in order to facilitate use in clinical trials and eventually daily practice.
Methods
We selected de novo paediatric CD patients with PCDAI>10 from the RISK cohort with baseline stool samples within 14 days after start of induction treatment. We assessed if they sustained remission at 12 months (PCDAI≤10). Using QIIME2 sequences were demultiplexed, joined, and denoised (deblur) to obtain amplicon sequence variants (ASVs). The ASVs were classified (classify-sklearn) using a pre-trained SILVA database. We used hierarchical Bayesian model for microbial community structure (BioMiCo), previously trained on treatment-naïve stool samples to predict treatment outcomes at 6 months according to baseline gut microbiome differences.
Results
Patient metadata and 16S rRNA amplicon data were available from 197 stool samples of newly diagnosed paediatric CD patients as part of the RISK cohort. Previous analysis of 42 truly treatment-naïve samples lead to prediction of samples in patients maintaining remission without early treatment escalation and those that did not in 81% and 75% (AUC=0.79). In this analysis, we selected 13 samples of children with PCDAI that were taken within 14 days after start of induction therapy. PCDAI varied from 15-47.5 at baseline. Therapy regimens started within the first 14 days were EEN, 5ASA, corticosteroids, immunomodulators and antibiotics. The Bayesian model predicted 12-month outcome of patients that maintained remission with a positive predictive value of 75% and negative predictive value of 60% (AUC 0.76).
Conclusion
Using treatment-naïve faecal samples only, a Bayesian approach predicted clinical course in treatment-naïve children with CD over the first year after diagnosis with high accuracy. When taking a broader timeframe of stool collection after start of treatment, the accuracy of the model decreased only slightly. Further exploration of microbiome signatures and potential use in practice should therefore emphasize the importance of treatment naivety of samples, but not necessarily rule them out for prediction of treatment outcome.