This study was conducted at Beijing Ditan Hospital, Capital Medical University, between September 2019 and November 2021, and included both a case-control study and a prospective cohort component (Figure 1). The protocol was approved by the institutional ethics committee (DT-IRB-2018-040-01), and written informed consent was obtained. The prospective cohort was registered in the Chinese Clinical Trial Registry (trial registration number: ChiCTR-ROC-17013343).

In the case-control study, four groups were enrolled: healthy controls (HC, n=40), compensated HBV-related cirrhosis (CC, n=30), cirrhosis with ascites but without SBP (NSBP, n=40), and cirrhosis with SBP (SBP, n=25). Clinical and microbial characteristics were analyzed, and the SBP-MI was constructed from differential taxa and evaluated for discrimination. We conducted an unmatched, stratified case-control study with consecutive enrollment under prespecified criteria. The resulting group sizes and age distributions reflect real-world accrual patterns and underlying disease severity.

In the prospective cohort, 140 patients with HBV-related cirrhosis and ascites were followed for 6 months, with SBP occurrence as the primary endpoint. Stool samples were collected from 140 patients at baseline. The prospective cohort enrolled ascitic patients without SBP at baseline. Accordingly, SBP cases included in the case-control study were not part of the baseline prospective cohort. The SBP risk prediction model (SBP-RP) was constructed using baseline SBP-MI and baseline clinical variables. A subset of 40 patients provided paired baseline and follow-up fecal samples, enabling assessment of clinical progression or improvement on the basis of microbial dynamics and changes in SBP-MI.

Exclusion criteria: The case-control and prospective studies applied identical exclusion criteria. Exclusion criteria included non-HBV etiologies of liver disease, hepatocellular carcinoma, other malignancies, severe comorbidities, active tuberculosis, concomitant extrahepatic infections, non-hepatic ascites, and the presence of a transjugular intrahepatic portosystemic shunt (TIPS) at baseline and follow-up.

Inclusion criteria: Core inclusion criteria were harmonized in the case-control and prospective studies. Core inclusion criteria were >18 years of age and had not received antibiotics, probiotics, proton pump inhibitors, bowel-preparation cathartics, or motility-altering agents including lactulose within 1 month before enrollment. This 1-month window was used to reduce recent medication-related perturbations of the gut microbiota at the time of sampling. Participants had not consumed yogurt and had not used other microbiota-altering agents. All participants consumed a mixed diet of plant- and animal-based foods, without vegetarian or other restrictive dietary patterns. Because the enrolled clinical states differed, disease-specific inclusion criteria were as follows. In the case-control study: HC required normal clinical, laboratory, and ultrasound findings without major systemic disease. CC was defined by chronic HBV infection with histologic, imaging, or endoscopic evidence of cirrhosis, liver stiffness >12.0 kPa, and Child–Pugh A (<7). NSBP required chronic HBV infection, cirrhosis with ascites, and absence of infection, encephalopathy, or gastrointestinal bleeding. SBP required fulfillment of NSBP criteria plus acute peritonitis or systemic inflammatory response, and an ascitic polymorphonuclear cell count ≥0.25 × 10⁹/L without secondary causes. In the prospective cohort study: HBV-related cirrhosis with ascites but without SBP at baseline.

Patients with HBV-related cirrhosis and ascites were prospectively followed for 6 months.

Progression group: Progression was defined within 6 months if any of the following applied:

Improved group: Improved status required that all of the following criteria were met:

Patients with stable disease who had ALT ≤ 40 U/L and AST ≤ 50 U/L and albumin 40 to 55 g/L at baseline, remained within these ranges at follow-up, and had no increase in ascites grade and no SBP, death, or liver transplantation were classified as Improved.

Demographic and clinical data were recorded at enrollment. Assessments of oesophago-gastric varices and measurement of portal pressure by the hepatic venous pressure gradient (HVPG) are invasive and are not routinely indicated outside the prophylaxis or treatment of variceal bleeding. Transient elastography may be unreliable in the presence of tense ascites or acute infection. In SBP, clinical instability at enrollment therefore precluded routine assessment of oesophago-gastric varices, portal pressure, and transient elastography. Fasting peripheral blood samples were collected for biochemical and virological assays. Fresh stool samples were obtained using sterile containers, transported on ice, aliquoted within 30 minutes, and stored at –80 °C until analysis. Microbial DNA extracted from fecal samples was used to amplify the V3–V4 region of the 16S rRNA gene, and the resulting libraries were sequenced on the Illumina NovaSeq platform. Further details are provided in the Supplementary Methods.

Taxa for the SBP-MI were selected in the case-control discovery set (HC, CC, NSBP, SBP) using two complementary genus level criteria. First, LEfSe was used to determine enrichment direction and effect size, with LDA scores defining the candidate pool. Second, four group mean relative abundance heatmaps were inspected to confirm trends consistent with disease staging. Detailed genus selection was described in the Results section. The SBP-MI was then calculated as follows: x_g denotes the genus-level relative abundance in each sample. A pseudocount of 10^–6 was added to each x_g to handle zero values and enable log-ratio calculation.

HBCDI was previously developed by our group to quantify gut microbial imbalance in patients with HBV-related cirrhosis. The HBCDI was then calculated as follows: HBCDI = (Escherichia − Shigella + Streptococcus + Lactobacillus)/(Ruminococcus + Prevotella + Bacteroides) (7, 9).

Raw reads were processed in QIIME 2 (release 2025.7) for 97% OTU clustering, taxonomic assignment, and diversity analyses. Differential taxa were identified using LEfSe after exporting the feature table and metadata. Baseline laboratory data were complete, with no missing values. Data visualization was performed in R (v4.5.1). Group comparisons were conducted with appropriate parametric or non-parametric tests, and categorical variables with χ² or Fisher’s exact test. Given the limited number of SBP events in the prospective cohort, predictive modeling was performed using Firth’s penalized logistic regression to reduce small-sample bias and address potential separation. Model performance was assessed by ROC analysis, bootstrap validation, and calibration. P<0.05 (two-sided) was considered significant. Details are provided in the Supplementary Methods.

A total of 30 patients with CC, 40 with NSBP, 25 with SBP, and 40 HC were included in the analysis. Baseline characteristics of the four groups are summarized in Table 1. Compared with HC and CC, NSBP and SBP were characterized by significantly higher TBil and lower RBC, ALB, HGB, and PLT (P < 0.05). The median NE% was numerically higher in the SBP group; however, the difference was not statistically significant. Smoking status was comparable across groups. Diuretic therapy was universal in the NSBP and SBP cohorts, consistent with standard ascites management. Model for End-Stage Liver Disease (MELD) increased progressively with disease stage, and Child–Pugh class shifted toward more advanced categories. In the SBP group, 5 of 25 patients (20%) had a prior history of SBP.

Analysis of α-diversity showed that the HC group had the highest Shannon and phylogenetic diversity (PD) whole-tree indices, indicating the greatest microbial and phylogenetic diversity. Compared with HC and CC, both NSBP and SBP exhibited significantly decreased diversity (P < 0.05), with SBP showing the lowest PD whole-tree index (Figure 2A). Based on unweighted UniFrac β-diversity (Figure 2B), distance values in NSBP and SBP were higher than those in HC and CC. Principal coordinates analysis (PCoA) further revealed a clear separation between diseased groups and healthy controls, suggesting marked alterations in community composition during disease progression.

At the phylum level (Figure 2C), Firmicutes, Bacteroidota, Proteobacteria, Actinobacteria, and Verrucomicrobiota together accounted for more than 90% of the total abundance. HC was dominated by Bacteroidota. With disease progression, Bacteroidota progressively decreased, whereas Proteobacteria markedly increased. In SBP, Firmicutes were reduced, while Proteobacteria and Verrucomicrobiota were enriched. At the genus level (Figure 2D), HC was enriched in Bacteroides, Prevotella_9, and Faecalibacterium, whereas pathogenic genera such as Escherichia–Shigella, Klebsiella, and Streptococcus peaked in SBP. LEfSe analysis identified stage-specific enriched taxa (Figure 2E): Roseburia in HC; Faecalibacterium and Prevotella in CC; Enterococcus and Haemophilus in NSBP; and Escherichia–Shigella, Klebsiella, and Veillonella in SBP. Collectively, these findings indicate that the progression of HBV-related cirrhosis is accompanied by a gradual loss of commensals and an expansion of potential pathogens, with the shift being most pronounced at the SBP stage.

To quantify gut microbial imbalance associated with disease progression, we constructed the SBP-MI by integrating the results of LEfSe analysis and genus-level abundance heatmaps. Accordingly, the numerator comprised Escherichia–Shigella, Klebsiella, and Veillonella, which were enriched in SBP with LDA scores of 4.71, 4.49, and 3.80, respectively. Streptococcus showed a stepwise increase from HC/CC to NSBP/SBP on the heatmap (Figure 2D) and was included to capture the progression toward SBP. The denominator comprised Roseburia, Bacteroides, Prevotella, and Faecalibacterium. Roseburia was enriched in HC (LDA 3.74). Bacteroides and Prevotella showed the highest mean relative abundance in HC and were clearly separated from the other groups. Faecalibacterium was higher in HC and CC and lowest in SBP. The SBP-MI differed significantly across the four groups (P < 0.001). Pairwise comparisons revealed significant differences between HC and NSBP, HC and SBP, as well as between CC and SBP (Figure 2F). Trend analysis further demonstrated that the SBP-MI increased progressively with disease progression (P < 0.001). In cirrhosis (CC, NSBP, SBP), SBP-MI correlated positively with MELD (Spearman ρ=0.36; 95% CI, 0.17–0.54; P = 0.0003).

We prospectively enrolled 140 patients with HBV related cirrhosis and ascites and followed them for 6 months. Forty patients provided paired fecal samples at baseline and follow-up. Of these, 32 were classified as improved and 8 as progressed. The paired-sample analyses were designed to determine whether SBP-MI and overall community composition track clinical improvement versus progression.

In the improved group (Supplementary Table S1), ALB, HGB, and RBC increased, whereas TBil, AFP, HBV DNA, and MELD decreased at follow-up (P < 0.05). α-diversity based on observed species was significantly higher at follow-up (Figure 3A), and PCoA revealed separation between baseline and follow-up samples (Figure 3B). At the genus level, Klebsiella, Streptococcus, and Enterococcus decreased (Figure 3C). LEfSe confirmed reduced Streptococcus and enrichment of Bacteroides, Prevotella, and Faecalibacterium (Figure 3D).

In the progression group (Supplementary Table S2), TBil and MELD increased at follow-up (P < 0.05). At the genus level, Klebsiella and Haemophilus increased, whereas Akkermansia, Bifidobacterium, and Lactobacillus decreased (Figure 3E). LEfSe revealed higher abundance of Escherichia–Shigella at baseline and Haemophilus parainfluenzae at follow-up (Figure 3F). Overall, the improved group was associated with increased diversity and enrichment of commensals, whereas the progression group was characterized by persistence or further enrichment of potential pathogens.

To further assess longitudinal changes in microbial indices, we calculated ΔSBP-MI and ΔHBCDI as the differences between follow-up and baseline values in the 40 patients with paired fecal samples. (ΔSBP-MI=SBP-MI_follow-up-SBP-MI_baseline; ΔHBCDI= HBCDI_follow-up- HBCDI_baseline) ΔSBP-MI was significantly higher in the progression group than in the improved group (P = 0.02) (Figure 3G). ROC analysis showed that ΔSBP-MI had an AUC of 0.76 (95% CI: 0.60–0.93). At the optimal cut-off of −0.62, ΔSBP-MI achieved a sensitivity of 1.00 and a specificity of 0.50. By comparison, ΔHBCDI showed lower discrimination, with an AUC of 0.55 (95% CI: 0.33–0.78) (Figure 3H), supporting the better performance of ΔSBP-MI in distinguishing improved from progressed outcomes.

In a 6-month prospective cohort comprising 140 patients with HBV-related cirrhosis and ascites, the primary endpoint was incident SBP, which occurred in 15 patients (10.7%). No patient was lost to follow-up, died, or underwent liver transplantation. Baseline clinical characteristics of patients with HBV-related cirrhosis and ascites were summarized in Supplementary Table S3. The association between a previous history of SBP and incident SBP during follow-up was summarized in Supplementary Table S4. The AUC for SBP-MI was 0.83 (95% CI: 0.74–0.92), compared with 0.73 for CDR (95% CI: 0.58–0.89) and 0.71 for HBCDI (95% CI: 0.57–0.85), indicating superior discriminatory performance of SBP-MI (Supplementary Figure S1). As shown in Table 2, 10 variables with P < 0.10 in univariable analyses were entered into the multivariable Firth logistic regression. SBP-MI, INR, and previous history of SBP remained independent predictors (P < 0.05). The final predictive equation was: SBP-RP = –6.8107 + 2.9873 × SBP-MI + 2.4316 × INR + 2.0950 × previous history of SBP. Discrimination was good, with an AUC of 0.91 (95% CI: 0.82–0.99). At the optimal cutoff of 0.262, the sensitivity was 0.80 and the specificity was 0.944 (Figure 4A). Bootstrap optimism correction (500 resamples; seed 2025) gave an optimism-corrected AUC of 0.90 (95% CI: 0.82–0.99). Calibration was acceptable overall. The calibration slope was 1.11 (95% CI: 1.08–1.30) and the Brier score was 0.051, with a non-significant Hosmer–Lemeshow test (P = 0.77). However, the calibration curve deviated from the ideal line at higher predicted probabilities (Figure 4B). These findings should be interpreted cautiously given the low number of events. A nomogram was generated to enhance clinical applicability (Figure 4C). These results suggest that the SBP-RP model can stratify short-term SBP risk in HBV related cirrhosis with ascites. Its longer-term performance will be assessed with extended follow-up and validated in larger external cohorts.