Presentation Information
[SY7-4]Gut Microbiota Analysis for Predicting Clinical Relapse in Patients with Quiescent Ulcerative Colitis: Multivariate Analysis Using MaAsLin2 and Development of a Predictive Score
*Hiroaki Kitae1, Tomohisa Takagi1, Mariko Kajiwara1, Kazuhiko Uchiyama1, Yuji Naito1,2 (1. Molecular Gastroenterology and Hepatology, Graduate School of Medical Science, Kyoto Prefectural University of Medicine, Japan, 2. Department of Human Immunology and Nutrition Science, Kyoto Prefectural University of Medicine)
Background: Predicting relapse in quiescent ulcerative colitis (qUC) remains challenging. The utility of gut microbiota as a relapse predictor has not been fully evaluated.
Methods: A total of 59 qUC patients attending Kyoto Prefectural University of Medicine between November 2016 and September 2017 were prospectively enrolled and followed for up to 3.5 years. Clinical remission was defined as a Lichtiger index score of 4 or less. Fecal microbiota was analyzed by 16S rRNA sequencing. After centered log-ratio (CLR) transformation, MaAsLin2 identified relapse-associated genera. A Ridge regression-based predictive score was validated by leave-one-out cross-validation (LOO-CV). Cox regression and Youden index-based cutoff analysis were performed.
Results: Of 59 patients, 19 (32.2%) relapsed. MaAsLin2 identified five genera associated with relapse (all p<0.05): Prevotella (p=0.008), Faecalibacterium (p=0.023), Odoribacter (p=0.033), Christensenellaceae (p=0.035), and Mogibacteriaceae (p=0.042). Of these, four remained significant in multivariate analysis: Prevotella, Faecalibacterium, Odoribacter, and Christensenellaceae. A 4-genera Ridge regression score achieved LOO-CV AUC=0.804, demonstrating good discriminative ability for predicting relapse. At the optimal cutoff, sensitivity=84.2%, specificity=72.5%, and negative predictive value (NPV)=90.6%. Cox regression showed HR=51.0 (95%CI 7.1–368.0, p<0.0001). Kaplan-Meier analysis showed significantly different relapse-free rates between risk groups (log-rank p<0.0001).
Conclusions: A 4-genera fecal microbiota score using Ridge regression achieved LOO-CV AUC=0.804, enabling risk stratification and potentially guiding personalized treatment in qUC remission.
Methods: A total of 59 qUC patients attending Kyoto Prefectural University of Medicine between November 2016 and September 2017 were prospectively enrolled and followed for up to 3.5 years. Clinical remission was defined as a Lichtiger index score of 4 or less. Fecal microbiota was analyzed by 16S rRNA sequencing. After centered log-ratio (CLR) transformation, MaAsLin2 identified relapse-associated genera. A Ridge regression-based predictive score was validated by leave-one-out cross-validation (LOO-CV). Cox regression and Youden index-based cutoff analysis were performed.
Results: Of 59 patients, 19 (32.2%) relapsed. MaAsLin2 identified five genera associated with relapse (all p<0.05): Prevotella (p=0.008), Faecalibacterium (p=0.023), Odoribacter (p=0.033), Christensenellaceae (p=0.035), and Mogibacteriaceae (p=0.042). Of these, four remained significant in multivariate analysis: Prevotella, Faecalibacterium, Odoribacter, and Christensenellaceae. A 4-genera Ridge regression score achieved LOO-CV AUC=0.804, demonstrating good discriminative ability for predicting relapse. At the optimal cutoff, sensitivity=84.2%, specificity=72.5%, and negative predictive value (NPV)=90.6%. Cox regression showed HR=51.0 (95%CI 7.1–368.0, p<0.0001). Kaplan-Meier analysis showed significantly different relapse-free rates between risk groups (log-rank p<0.0001).
Conclusions: A 4-genera fecal microbiota score using Ridge regression achieved LOO-CV AUC=0.804, enabling risk stratification and potentially guiding personalized treatment in qUC remission.