Gut metagenomic profiles predict 90-day hospitalizations despite controlling for diet, demographics & cirrhosis details in a multi- country cohort of patients with cirrhosis

Abstract

Background and aims: Prediction of outcomes such as hospitalizations is critical in cirrhosis to ensure timely prevention. Microbial composition is altered in cirrhosis but can show wide variation between countries. Aim: define microbial features that associate with 90-day hospitalization in a multi-national cirrhosis cohort. Method: Cirrhosis outpatients were recruited from 7 countries (USA, Türkiye, Korea, Denmark, Lithuania, Brazil & Mexico). Demographics, cirrhosis severity (MELD, complications, labs, medications), diet history, were recorded, stool collected & pts followed for 90 days for hospitalizations. Shotgun metagenomics for NGS was performed on stool. Regression analysis was performed using Maaslin2 for 90-day hospitalizations prediction. Covariates were chosen after PERMANOVA analysis & microbial species were added. Variables with p < 0.05 & q < 0.05 were considered significant. Results: 679 cirrhosis pts (170 US, 90 Brazil, 53 Denmark, 93 each Lithuania & Mexico, 99 Korea & 84 Türkiye) were enrolled. 60% were men, MELD 12 ± 5, albumin3.5 ± 0.70, Na 138 ± 7. Most common etiology was alcohol (35%) then viral (31%) & MASLD (17%). 56% were decompensated (35% on lactulose, 19% on rifaximin, 44% prior ascites, 19% prior variceal bleed, 31% on NS βblockers). Diet varied esp with yogurt & coffee intake. 90-day Hospitalizations: Seen in 169 (25%), most of which 133 (79%) were liver-related. Country-wise rate differed with lowest in Brazil & Türkiye and highest in Lithuania & Korea). PERMANOVA showed that MELD, decomp status, yogurt intake, country of origin & HE-medications were significant covariates. Maaslin2: 15 species associated with hospitalizations after controlling for country, decomp status, MELD score, yogurt, HE medications & alcohol etiology. Linked with ↓ hospitalizations: All had p < 0.01, Ruthenibacterium lactatiformans (coeff = 2.9, q < 0.001), Bacteroides uniformis (−2.3, p & q < 0.001), Flavonifactor plautii (−2.2, q < 0.001), Faecalibacterium prausnitzii (−1.9, q = 0.02), Clostridium fessum (−1.7, p & q < 0.001), Eggerthella lenta (−1.7, q = 0.01), Clostridiales bacterium (−1.7, q = 0.01), Bifidobacterium pseudocatenulatum (−1.7, q = 0.05), Anaerobutyricum hallii (−1.6, q = 0.02), Eubacterium ramulus (−1.4, q = 0.02), Clostridium leptum (−1.4, q = 0.04), Intestinimonas butyriciproducens (−1.3, q = 0.04), GGB9614_SGB15049 (−1.2 q = 0.04) & GGB2653_SGB3574 (−1.2, q = 0.03). These are short-chain fatty acid or lactate producers, or those involved in beneficial immune function. Linked with ↑ hospitalizations: Enterococcus faecium (Coeff 2.0, p < 0.001, q = 0.003), which is a pathobiont. Conclusion: In a prospective cohort of outpatients with cirrhosis from 7 countries, microbial profiles could predict 90-day hospitalizations despite controlling for diet, cirrhosis details, country of origin, and demographics. Microbial profiling could improve risk prediction in cirrhosis beyond single center studies.