We conducted our meta-analysis following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines (Supplementary Table S1) (19). The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO # CRD42024567537; July 10, 2024; revised October 2024; available at http://www.crd.york.ac.uk/PROSPERO/).

Publicly accessible databases, including Google Scholar, PubMed, the China National Knowledge Infrastructure, and the China Biology Medicine database (CMBdisc), were searched from database inception to 20 February 2025 to identify RCTs conducted in China for the treatment of PAGE comparing S. boulardii CNCM I-745 with controls in children who also received smectite. The search strategies are provided in Supplementary Table S2. No language restrictions were imposed, and non-English publications were translated. Recursive searches of the gray literature were performed by screening reference lists, authors, and reviews.

The inclusion criteria were as follows: RCTs with prospective, parallel-group designs, participants randomized to either S. boulardii CNCM I-745 in combination with smectite (diosmectite or montmorillonite) or smectite alone (controls), children aged ≤18 years with acute diarrhea (defined as ≥3 loose or watery stools per day lasting <14 days), and living in China. All participants may receive standard therapies (oral or IV rehydration therapy, antiviral medications or antibiotics as needed, or dietary changes). Randomization was required to be clearly stated (not just “divided into two groups” or not specified). The strain had to be clearly identified as S. boulardii CNCM I-745, either by brand name (“Yihuo” or “Bioflor”), Biocodex import license number, or by listing Biocodex as the manufacturer. The probiotic had to be given orally at a designated daily dose for at least 3 days.

The exclusion criteria were as follows: adult patients (>18 years old); studies not conducted in China; non-human studies; case reports or case series; early phase 1 (safety) or phase 2 (mechanism of action, dose ranging, formulation, or pharmacokinetic) studies; retrospective case–control studies; studies without a control group; interventions that were not well-described; and reviews, meta-analyses, duplicate reports, or studies lacking original quantitative data. Trials that administered additional treatments, including specific antibiotics not directed at bacterial diarrhea, zinc, racedotril, fructose, aluminum diphosphate, other probiotics, or Chinese medicines, were also excluded. Studies involving other types of diarrhea (antibiotic-associated diarrhea, C. difficile infection, irritable bowel syndrome, diarrhea secondary to pneumonia, or allergic diarrhea) were also excluded.

Two reviewers (LM and TL) independently screened titles and abstracts and extracted data for 24 recommended items using a pre-designed data extraction form (18) following the standard methods for systematic reviews and meta-analyses (19–21). Any disagreements were resolved through discussion until a consensus was reached.

The data extracted included population, intervention, control, and outcome data (PICO): (1) population characteristics (pediatric, age range, and country), (2) intervention details (type of S. boulardii, daily dose, formulation, treatment duration, and follow-up period), (3) comparison groups (type of control group, including placebo-controlled or open-label, unblinded designs), and (4) outcomes, including improvement in PAGE symptoms (cure rate, duration of diarrhea, effectiveness rating, stool frequency/day by end of study), time to resolution of vomiting or fever, length of hospitalization, safety measures, and changes in immune markers. For data not reported in the published article, we attempted to contact the author or co-authors to obtain the missing information.

Each trial was reviewed for quality and risk of bias (RoB) and scored independently by both co-authors using standard methods (22). Study quality was assessed for the 24 recommended items for clinical trials and graded as high quality [≥18 items (75%) present], moderate quality (12–17 items present, at least 50%), and low quality (<12/24 items present). Risk of bias was assessed using the RoB 2.0 tool and was graded as low, some concerns, or high risk across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of the reported result (22). The randomization process was rated as “low” if the randomization method was reported or if baseline characteristics were not significantly different. Deviations from intended interventions were rated as “low” when double- or single-blinding was used, “some concerns” when non-blinded controls were used but no deviations from group assignments were found, and “high” when open-label trials showed significant deviations from group assignments. Missing outcome data were scored based on attrition rates and rated as “low” for 0%–10% attrition, “some concerns” for 11%–50% attrition, and “high” for >50% attrition. Measurement of outcomes was rated as “low” when outcome assessments were single- or double-blinded or when outcomes in open-label trials were documented in inpatient medical records and “high” when unblinded trials were in outpatient settings or when patient status was not reported. Selection of the reported results was rated as “low” if outcomes were defined a priori and no post hoc outcomes were reported; otherwise, it was rated as “high” bias. A summary risk-of-bias figure was generated, and the impact of study quality was assessed (23). The strength and certainty of the evidence were assessed using the Grading of Recommendations, Assessment, Development and Evaluations tool (24).

The cure rate was defined as cessation of diarrhea (less than three loose or watery stools per day) by the end of the study intervention. The duration of diarrhea was defined as the number of days from enrollment to the first day of resolution of diarrhea symptoms. The total effective rate (TER), which measures the improvement of diarrheal symptoms, was categorized into three categories: “markedly effective or cured” (resolution of diarrhea symptoms by the end of the intervention), “effective or improved” (moderate improvement in symptoms), or “not effective” (persistent diarrhea symptoms at the end of the intervention). The total effectiveness rate was defined as the proportion of patients classified as either markedly effective or effective. Safety outcomes were defined as the occurrence of any adverse events reported during the study period.

Other outcomes were also assessed and included time to resolution of vomiting or fever (defined as the number of days from study enrollment to the last day of symptoms), stool frequency (number of bowel movements per day at the end of treatment), length of hospitalization (mean duration of hospital stay among hospitalized patients), changes in cytokine levels from enrollment to the end of the study (CD4/CD8 ratio, TNF, CD3, CRP, interferon levels, IL-3, and IL-10), and changes in the intestinal microbiome from enrollment to the end of the study.

Factors that might impact the efficacy of S. boulardii were also assessed, including the etiology of diarrhea (rotavirus, bacterial, or other causes), the daily dose of S. boulardii, the timing of intervention initiation (within 48 h of diarrhea onset or longer), risk of bias, and patient age group.

Inclusion of studies in the meta-analysis required at least two RCTs using a common outcome measure that compared S. boulardii with a non-probiotic control. Statistical analyses and forest plot generation for pooled summary estimates were performed using Stata software version 16 (StataCorp, College Station, Texas) with the meta-analysis modules (25). Dichotomous outcomes were assessed using relative risks (RRs) with 95% confidence intervals (CIs), while continuous outcomes were evaluated using standardized mean differences (SMDs) with 95% CIs (26). The significance level was set at p-value <0.05. Heterogeneity across studies was evaluated using the I² statistic (with values >50% indicating a high degree of heterogeneity). Bayesian random-effects models were used when heterogeneity was high (I² > 50%); otherwise, fixed-effects models were used (26). Publication bias was assessed using funnel plots and Egger's test (25). Subgroup analyses were performed to explore sources of heterogeneity, which were assessed using the Cochrane Q-test (25). Sequential sensitivity analyses were performed to explore the extent to which outcomes depended on a particular trial. Trials with missing data for specific outcomes were excluded from the analyses of those respective outcomes.

Our literature search screened 877 abstracts from database inception to 20 February 2025, of which 810 were excluded (Figure 1). Initial screening excluded non-RCT study designs or reviews/meta-analyses (n = 403). Full-text articles were then screened for eligibility, and 407 were excluded based on predefined exclusion criteria (most commonly due to inclusion of other types of diarrhea) (n = 127) or the use of S. boulardii in both intervention and control groups (n = 73). A total of 67 trials reported smectite use; however, after careful translation from Chinese into English and review by two researchers, 10 trials were excluded for failure to fulfil the inclusion criteria (Supplementary Table S3) (27–36). Ultimately, 57 randomized controlled trials (N = 5,767 participants) were included in the meta-analysis (37–93). Evidence of publication bias was found (Supplementary Figure S1), P < 0.001, suggesting the potential omission of small-scale trials with negative or non-significant results.

The characteristics of the included trials and study participants are summarized in Supplementary Table S4. The RCTs enrolled children, either neonates (<1 month old) (43) or children aged 2 months to 10 years old; most trials enrolled patients shortly after the onset of diarrhea: 1–3 days (n = 34, 59.6%), 4–7 days (n = 11, 19.3%), and 8–14 days (n = 4,7%); however, 8 (14.0%) did not report when the timing of diarrhea onset. The most commonly identified etiology of PAGE was rotavirus (n = 25, 43.8%). One trial reported mixed viral etiologies (63), one reported bacterial etiologies (Enterococcus, Salmonella, or enteropathogenic E. coli) (74), and one reported a mix of bacterial, viral, and parasitic etiologies without specifying the types (86). Notably, 29 (50.9%) trials did not report etiological data. Most of the children were inpatients (n = 42, 73.7%). Only three trials enrolled outpatients (5.3%) (46, 58, 63), four enrolled a mix of inpatients and outpatients (7%) (37, 40, 57, 60), and eight trials (14%) did not report the status of enrolled children.

The study size ranged from 50 to 246 participants per trial (mean 101 ± 34 participants per trial). Most children (93%) received rehydration therapy. Some trials (n = 29, 50.9%) also allowed other treatments to be used in both groups as needed (antivirals, antibiotics, anti-inflammatory agents, or dietary changes). Two trials were single-blinded (patients) (62, 72), and none used a placebo control. Of the 57 trials, 12 (21%) had an overall “high” risk of bias, 43 (75.5%) were rated as having “some concerns” mainly due to the lack of blinding, and only two trials (3.5%) had a “low” risk of bias (Supplementary Figure S2). A review of 24 recommended factors for clinical trials (94) found that 55 studies (96.5%) reported 50%–70% of the factors, while only two studies (3.5%) reported >70% of the factors. The most common unreported factors included the method of randomization (36.8%) and adverse event or safety data (50.9%). The most common randomization method was the use of a random number table (n = 26); other methods (n = 10) included computer programs, lottery methods, card drawing, odd/even medical record numbers, or colored balls. Most trials reported no attrition or loss to follow-up (n = 54, 94.7%), while three trials reported low attrition rates (3%–8%) (38, 47, 48). Most trials did not follow children after discontinuation of the intervention (n = 53, 93%); however, four trials reported follow-up periods ranging from 2 days to 3 months post-intervention (45, 46, 68, 72). None of the trials reported sample size calculations.

The S. boulardii strain was confirmed as CNCM I-745 when identified by brand name (Yihuo or Bioflor) in nine trials (15.8%), by import/registration number in 39 trials (68.4%), or by listing Biocodex as the manufacturer in nine trials (15.8%). Trials in which the strain could not be verified were excluded from our review. The duration of treatment ranged from 3 to 15 days in 45 RCTs (78.9%), extended to 12 weeks in one trial (64), and was not reported in 11 trials (19.3%) (Supplementary Table S4). The formulation of S. boulardii was as a powder in a sachet (n = 54, 94.7%), but it was not reported in three trials (5.3%) (60, 73, 76). The dose of S. boulardii varied across studies: 250 mg/day in seven trials (12.3%), 500 mg/day in one trial (1.8%), a range of 250–1,000 mg/day in one trial (84), and age-adjusted dosing in 48 trials (84.2%), typically ranging from 250 to 500 mg/day. Intervention initiation occurred within 48 h of diarrhea onset in 21 trials (36.8%), within 14 days in 28 trials (49.1%), and was not reported in eight trials (14%). The type of smectite was predominantly diosmectite (n = 53, 93%), while four trials used montmorillonite (7%) (40, 47, 68, 75).

Cure rates were reported in 54 trials (94.7%) but were not available in three trials (5.3%) (69, 70, 78). In the S. boulardii group, the cure rate ranged from 26.5% to 97.3%, while cure rates in controls ranged from 11.1% to 82.7% (Supplementary Table S5). The meta-analysis of pooled data from 54 trials showed a significantly higher cure rate with S. boulardii compared with control treatment (RR = 1.45, 95% CI 1.38, 1.53, I² = 6%, P < 0.0001; Figure 2). Four factors had sufficient data to permit subgroup analyses (Supplementary Table S6), but none had a significant impact on efficacy. These factors included daily dose of S. boulardii (250 mg/day vs. age-adjusted dosing), etiology (rotaviral vs. other causes), or timing of treatment initiation (within 48 h vs. longer), and type of patient (inpatient, outpatient, or mixed).

The duration of diarrhea was reported in 39 trials (68.4%) but was not available in 18 trials (31.6%). The mean duration of diarrhea ranged from 1.2 to 5.7 days in children receiving S. boulardii and from 2.2 to 6.8 days in controls. Analysis of pooled data from 39 trials demonstrated that S. boulardii significantly reduced the duration of PAGE by 1.54 days (SMD: −1.54 days, 95% CI −1.79, −1.29, I² = 92.2%, P < 0.0001; Figure 3). Sensitivity analysis excluding two trials with strong effects (56, 70) did not significantly change the outcome measurement (SMD = −1.34 days, 95% CI −1.53, −1.14, I² = 77.3%, P < 0.0001). Factors that might reduce heterogeneity were explored using subgroup analyses, but they did not have a significant impact on this outcome (Supplementary Table S6). A daily dose of S. boulardii at 250 mg/day showed the greatest reduction in diarrhea duration (−2.20 days) compared to age-adjusted dosing (−1.47 days), but dose groups did not result in lower heterogeneity (97.1% and 91.3%, respectively). Among the 20 trials that reported diarrhea etiology, reduced heterogeneity was observed for rotaviral (86.7%) and other etiologies (57.5%), but the overall reduction in diarrhea duration was not significantly greater (SMD = −1.18 and −1.21 days, respectively), as shown in Supplementary Table S6. Timing of treatment initiation did not significantly reduce heterogeneity in the duration of diarrhea (started within 48 h, I² = 92.9% or started after 48 h, I² = 89%). The type of patient was assessed for inpatients; the duration was not significantly different in this group (SMD = −1.59 days, 95% CI −1.88, −1.29) nor was heterogeneity significantly reduced (I² = 93.2%).

The TER was a common outcome in trials conducted in China and was reported in 53 trials (93.0%) (Supplementary Table S5). The TER ranged from 83.7% to 98.4% in the S. boulardii groups and was lower in the controls (63.3%–86.7%). Analysis of pooled data from the 53 trials demonstrated that S. boulardii had a significantly higher TER compared with controls (RR = 1.21, 95% CI 1.18, 1.24, I² = 0%, P < 0.0001; Table 1). Subgroup analyses based on daily dose, diarrhea etiology, timing of treatment initiation, and type of patient did not have a significant impact on this outcome (Supplementary Table S6).

Of the 57 included trials, 28 (49.1%) reported safety data. Among these, 15 trials (26.3%) explicitly stated that no adverse events occurred, while 13 trials (22.8%) reported at least one adverse event. The remaining 29 trials (50.9%) did not report any safety data (Supplementary Table S4). Of the 13 RCTs reporting adverse events, 11 provided data stratified by study group, while two trials did not (52, 77). Children receiving S. boulardii had a significantly reduced rate of adverse events (RR = 0.64, 95% CI 0.43, 0.97, I² = 0%, P = 0.03; Figure 4). No significant differences were observed in the types of adverse events reported between the two groups.

Daily stool frequency, duration of vomiting, and length of hospitalization were also analyzed. Stool frequency (bowel movements per day) was reported in 26 trials (45.6%) . In the S. boulardii groups, stool frequency ranged from 1.1 to 3.2 bowel movements per day, whereas controls reported higher frequencies (ranging from 1.5 to 5.5 bowel movements per day) (Supplementary Table S7). Pooled data analysis found a significant reduction in stool frequency when S. boulardii was given (SMD = −1.76/day, 95% CI −2.19, −1.34, I² = 95.8%, P < 0.001; Table 1). Vomiting, a marker of severe dehydration, was reported in only 16 trials (28.1%). The duration of vomiting ranged from 1 to 2 days in the S. boulardii groups and from 1 to 4 days in controls. A meta-analysis found that S. boulardii significantly reduced the duration of vomiting (SMD = −1.75 days, 95% CI −2.27, −1.23, I² = 95.3%, P < 0.0001; Table 1). Length of stay was reported in 11 trials (19.3%) (Supplementary Table S7). Children were hospitalized for 4–6 days in the S. boulardii group and 6–7 days in the controls. A meta-analysis found that S. boulardii significantly reduced the length of hospitalization by nearly 2 days (SMD = −1.68 days, 95% CI −2.19, −1.16, I² = 92.9%, P < 0.0001; Table 1). Subgroup analyses did not find any factors that significantly reduced heterogeneity; heterogeneity persisted across all examined subgroups, including rotavirus etiology (I² = 88.4%), age-adjusted dosing (I² = 93.8%) or fixed dosing of 250 mg/day (I² = 86.4%), initiation of treatment within 48 h (I² = 96.2%) or after 48 h (I² = 93.4%), and inpatients (I² = 92.9%).

Changes in immune markers were also analyzed (Supplementary Table S8). Changes in TNF-α were reported in 10 trials (17.5%). A meta-analysis revealed a significant reduction in TNF-α levels (SMD = −2.42, 95% CI −3.23, −1.60, I² = 96.5%, P < 0.0001; Figure 5). Given the high heterogeneity, an analysis was conducted to identify factors that might reduce it, but none were identified. Change in TNF-α levels were similar for rotaviral diarrhea (SMD = −2.12, 95% CI −2.91, −1.33, I² = 93.9%), age-adjusted dosing (SMD = −2.42, 95% CI −3.23, −1.6, I² = 96.5%), and initiation of treatment (within 48 h: SMD = −2.32, 95% CI −3.85, −0.79, I² = 97.4% or after 48 h: SMD = −2.48, 95% CI −3.54, −1.43, I² = 96.5%). All these trials were in inpatients. Changes in the CD4/CD8 ratio during the study were reported in 17 trials (29.8%), while 40 trials (70.2%) did not report this outcome. No significant difference in CD4/CD8 ratios was observed between the study groups (Table 1). Thirteen trials reported CD3 data, and pooled analysis showed that S. boulardii significantly increased the levels of CD3 compared with controls (SMD = 1.90, 95% CI 1.4, 2.4, I² = 92.9%, P < 0.0001). Changes in other immune markers (CRP, IL-10, IL-3) were not reported by sufficient numbers of trials to allow meaningful analysis.

Other outcomes were infrequently reported and thus could not be robustly analyzed, including reductions in abdominal pain (reported in eight trials), changes in the intestinal microbiome (reported in five trials), gastrointestinal (GI) hormone levels (reported in two trials) (82, 90), and intestinal barrier markers (reported in one trial) (93).