Probiotics for the management of irritable bowel syndrome: a systematic review and three-level meta-analysis

Objective: Previous systematic reviews demonstrated a potentially beneficial effect of probiotics on irritable bowel syndrome (IBS). However, these studies are either affected by the inclusion of insufficient trials or by the problem of dependent data across multiple outcomes, and an overall effect size has not been provided. We aimed to determine the effect of probiotics on IBS through a three-level meta-analysis and clarify potential effect moderators. Methods: We searched MEDLINE, Embase, and Web of Science, screening for randomized controlled trials (RCTs) that examine the effect of probiotics on IBS. The primary outcome was the improvement in the severity of global IBS symptoms at the end of treatment. The secondary outcomes were the improvement in abdominal pain and the quality of life. The effect sizes of the probiotics were measured by using the standardized mean difference (SMD) and pooled by a three-level meta-analysis model. Results: We included 72 RCTs in the analysis. The meta-analysis showed significantly better overall effect of probiotics than placebo on the global IBS symptoms (SMD −0.55, 95% CI −0.76 to −0.34, P<0.001), abdominal pain (SMD −0.89, 95% CI −1.29 to −0.5, P<0.001) and quality of life (SMD 0.99, 95% CI 0.45 to 1.54, P<0.001), respectively. Moderator analysis found that a treatment duration shorter than 4 weeks was associated with a larger effect size in all the outcomes, and Bacillus probiotics had better improvement on the abdominal pain. Conclusions: Probiotics had a short-term effect and a medium effect size on the global IBS symptoms. Treatment duration and types of probiotics affected the effect size of probiotics, and shorter durations and Bacillus probiotics were associated with better treatment effects. Registration: Open Science Framework.


Introduction
Irritable bowel syndrome (IBS) is a disorder of the brain-gut axis characterized by frequent abdominal pain, bloating, flatulence, and change of bowel habitsconstipation or diarrhea.The global prevalence of IBS was 9.2% but varied across different regions; the prevalence was similar in Western countries, which is between 8.6 and 9.5% when the Rome III criteria were adopted and is between 4.5 and 4.7% with Rome IV [1] , and the prevalence was as high as 21.2% in Japan when the Rome III criteria were adopted [2] .IBS affects the quality of life substantially to the same degree as inflammatory bowel diseases [3] .The high prevalence and heavy disease burden urge the development of treatments for patients with IBS.
Owing to the complexity of IBS pathophysiology and the long disease duration, dietary supplements and alternative treatments are supposed to be more appropriate than pharmacological treatments since they are acknowledged to be harmless or with few adverse events.However, dietary supplements, like

HIGHLIGHTS
• Despite the previous reports of several systematic reviews and meta-analyses examining the effect of probiotics for irritable bowel syndrome (IBS), the general effect size of probiotics on IBS symptoms and the essential effect moderators are unknown.• In this meta-analysis incorporating 72 randomized controlled trials (RCTs), probiotics showed a medium effect size on the improvement of global IBS symptoms (standardized mean difference, − 0.55, 95% CI − 0.76 to − 0.34) compared with placebo.• A treatment duration shorter than 4 weeks and Bacillus probiotics were associated with larger effect sizes.
probiotics, are in the conundrum of 'no harm, might help' [4] .Numerous systematic reviews and meta-analyses were published to examine the effect of probiotics on IBS, and most of them suggested a beneficial effect [5][6][7][8] , but convincing evidence cannot be reached owing to the small sample size, single-center design, and high risk of bias of the included trials.Additionally, the diverse outcome assessments and differential assessment time points hinder a general evaluation of the effect size of the probiotics.The previously published meta-analyses normally selected a specific time point and one of the outcomes to pool, which caused a loss of informationmany of the outcomes are correlated and should be included for analysis [9] .In addition, numerous factors might affect the effect size of probiotics, and previous reviews concluded that specific strains of the probiotics had larger effects than the others [7] .Other effect moderators like treatment duration and the patient's characteristics are rarely evaluated.
Based on these grounds, we raised two clinically relevant questionswhat is the overall effect size of probiotics in the management of IBS, and what are the major effect moderators that significantly affect the size of the probiotic effect?One problem, not being fully settled in previously published meta-analyses on the topic, is that the effect sizes reported by the included trials might not be independent.For example, the studies conducted in the same region (i.e.European countries) might report similar results, which introduces dependence.A three-level meta-analysis is developed to solve this problem, which treats effect sizes nested within a study as dependent variables and examines the source of heterogeneity within a study and between studiesavoiding the inflation of type I error [10] .In addition, a three-level meta-analysis can include effect moderators in the model and assess the impact of the moderators on the effect sizes, which gives a better explanation for the effect of an intervention than the conventional meta-analysis.We aimed to assess the overall effect of probiotics on the improvement of IBS symptoms and find out the important effect moderators through the threelevel meta-analysis.

Study overview
We performed a systematic review and multi-level meta-analysis, and this work had been reported in line with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [11] (Supplemental Digital Content 1, http://links.lww.com/JS9/A874,Supplemental Digital Content 2, http://links.lww.com/JS9/A875) and AMSTAR (Assessing the methodological quality of systematic reviews) Guidelines [12] (Supplemental Digital Content 3, http://links.lww.com/JS9/A876).The review was registered in Open Science Framework prior to conduction.The meta-analysis used aggregate-level data from published randomized controlled trials (RCTs).Ethical approvals and patient informed consent were acquired in each participating center of the RCTs.The work has been reported.

Literature search and study selection
We searched MEDLINE, Embase, and Web of Science from inception to 12 November 2022, aiming to screen for RCTs that examined the efficacy of probiotics on IBS.Comprehensive search strategies with the combination of MeSH (Medical Subject Headings) terms and keywords were developed for the search in the databases.The search strategies were shown in eTables 1-3 (Supplemental Digital Content 4, http://links.lww.com/JS9/A877)We also searched previously published systematic reviews and read the reference lists of the reviews, trying to find out missing RCTs from our literature search.One author (C.-R.X.) performed the literature search, and two authors (L.Y. and X.-Y.X.) independently screened the retrieved articles.The inclusion and exclusion criteria are listed below.
RCTs were included if they (1) recruited participants who aged over 18 years and were diagnosed with IBS or any of its subtypes; (2) tested the effect of probiotics by comparing them with a placebo; (3) measured any of the following outcome: improvement of IBS symptoms, improvement of abdominal pain, or quality of life.
RCTs were excluded if they (1) also recruited other gastrointestinal diseases (e.g.functional dyspepsia, inflammatory bowel diseases); (2) published as letters that had insufficient information to judge the exact type of probiotics and their controls, or insufficient information on the types of outcomes.

Outcome assessments
The primary outcome of this study was the severity of global IBS symptoms, which normally include abdominal pain, discomfort in the abdominal region (i.e.bloating, urgency, indigestion), and changed bowel habits (constipation, diarrhea, or diarrhea alternating with constipation).These symptoms could be assessed by asking questions with yes-or-no answers like 'Are your global IBS symptoms relieved?' or by adopting scales such as the IBS-SSS (Irritable Bowel Syndrome Severity Scoring System) scale.
Secondary outcomes included the severity of abdominal pain and the improvement in quality of life.Abdominal pain is one of the most essential symptoms of IBS, and it is usually separately reported.The severity of abdominal pain could be assessed by using binary outcomes indicating whether relief of pain was achieved or by rating scales such VAS (Visual Analog Scale) score.The quality of life in IBS patients is assessed by scales like IBS-QoL (Irritable Bowel Syndrome Quality of Life).Our study included all these scales.

Data extraction
Two reviewers (X.-Y.W. and S.-J.F.) independently extracted study data from the included studies.Characteristics of the study design, participants, intervention, controls, and outcomes were separately extracted, and the characteristics were also coded and prepared for moderator analysis, which was described in detail in the statistical analysis section.

Risk of bias assessment
We assessed the risk of bias (RoB) using the revised Cochrane RoB tool (RoB 2.0), in which five domainsbias arising from the randomization process, bias due to deviation from intended interventions, bias due to missing outcome data, bias in the measurement of an outcome, and bias in the selection of the reported resultwere assessed and an overall RoB (low/high RoB or some concerns) was provided for each study.The certainty of the evidence was assessed by using the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) approach.

Statistical analysis
We estimated the effect size of probiotics versus placebo by using the standardized mean difference (SMD, also known as Cohen's d).
Based on the assumption that the underlying continuous measurements in each intervention group follow a logistic distribution and that the variability of the outcomes is the same in both the intervention and control groups, the odds ratios (ORs) can be reexpressed as an SMD according to the following simple formula [13,14] . The effect size was interpreted as small, medium, and large with the cut-off points of 0.2, 0.5, and 0.8, respectively [15] .To ensure the consistency of the direction of outcome measurements, the category outcomes were measured with the ORs of participants with failures of improvement, while continuous outcomes were measured with the change in the severity of the global IBS symptoms or abdominal pain.
We used a three-level random-effects meta-analysis model to pool the effect sizes, estimated the heterogeneity within-study (level 2) and between-study (level 3), and used Cochran's Q test to evaluate whether the heterogeneity was statistically significant (defined as P < 0.05).We compared the traditional two-level meta-analysis with the three-level model by evaluating the AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion) of the models and estimated the significance of the difference by the likelihood ratio testwhich also used a cut-off point of 0.05.To assess potential publication bias, we generated funnel plots for the three outcomes to perform a visual assessment for asymmetry [16] .
We further performed moderator analyses to find out which factors substantially affected the effect size by using the metaregression model.Four groups of factors were analyzed.The first group was study-design-related: countries hosting the studies, types of RCT (single vs. multicenter design), number of study sites, and the total study duration (measured by weeks).The second group was participant-related: age, proportion of females, the proportion of participants who dropped out from the study, duration of disease, and diagnostic criteria.The third group was intervention-related: duration of the intervention (measured by weeks) and the types of probiotics (classified as Bacillus, Bifidobacterium, Enterococcus, Escherichia coli, Lactobacillus, Saccharomyces, and a combination of differential probiotic strains).The fourth group was outcome-related: the type of data (continuous vs. categorical data) and types of outcome definition (i.e.global IBS symptoms could be further classified as adequate relief of symptoms, any relief of global symptoms, bloating, bowel habit, abdominal discomfort, and defecation urgency).The analysis was performed in the R environment (version 4.2.2) and the metafor package (version 4.2-0).

Trial characteristics
We identified 72 RCTs  after screening for 2725 pieces of articles and included a total of 8581 participants. The proess of screening is shown in Figure 1.The included population had a mean age of 41.7 years and a mean proportion of 65.8% of females.Sixty-seven (93.1%) of the included studies adopted the Rome criteria as the diagnostic standard, and 53 (73.6%) of the studies recruited at least two subtypes of IBS.Of the 18 studies that focused on a single subtype, 13 studied diarrhea-predominated IBS.
Seventy studies compared probiotics with placebo controls.More detailed information is shown in Table 1.Of the 72 included RCTs, nine were classified with a low risk of bias, two were with a high risk of bias, and 61 with some concerns; the assessment of each domain was presented in eFigure 1 (Supplemental Digital Content 4, http://links.lww.com/JS9/A877)The GRADE assessment for all three outcomes showed low certainty of the evidence.The summary of the GRADE table is shown in eTable 4 (Supplemental Digital Content 4, http://links.lww.com/JS9/A877).Figure 2 shows the concept of the three-level meta-analysis and the results of the metaanalyses on the three outcomesthe severity of global IBS symptoms, the severity of abdominal pain, and the quality of life assessment.
The moderator analyses showed that participants from different continents reported differential effects, while participants from the Asia region reported the largest effect size (Fig. 3A).Additionally, we found that study duration affected the effect size, and a longer study duration was associated with a smaller effect size (Fig. 3B).We also found that longer treatment duration was associated with a smaller effect size (Fig. 3C).The type of outcome impact also had an impact, and probiotics had a larger effect size on abdominal discomfort (SMD − 1.55, 95% CI − 2.97 to − 0.15, P = 0.017; Fig. 3D).Other factors had no significant impact on the effect sizes of the probiotics (Table 2).
In the moderator analysis, we found that the study duration affected the effect sizes.RCTs with a study duration shorter than 4 weeks (SMD, − 1.71, 95% CI − 2.62 to − 0.8) had significantly larger effect sizes than other RCTs with a study duration longer than 4 weeks (P < 0.001; Table 2).The proportion of females also affected the effect sizes, RCTs with a higher proportion of females were associated with smaller effect sizes (coefficient estimate 0.02, 95% CI 0.001-0.046,P = 0.04).For other factors, no significant impact was found (Table 2).
The moderator analysis showed that the treatment duration and the types of probiotics affected the effect size.A treatment duration within 4 weeks showed a significantly larger effect than a treatment duration between 4 and 8 weeks and a treatment duration longer than 8 weeks (Table 2).The probiotic strains containing Bacillus and Bifidobacterium showed significantly larger effect sizes than those containing Saccharomyces (Table 2).

Discussion
By pooling all the effect sizes from the included RCTs, we found a general medium effect size (with an SMD larger than 0.5) of probiotics on the improvement of IBS symptoms compared with placebo, and a large effect size (with an SMD larger than 0.8) of probiotics on the abdominal pain and the scores of quality-of-life assessments.We found that the treatment duration and study duration were the most important moderators of effect, and a longer study duration or treatment duration was associated with a smaller effect size.When the treatment duration was longer than 8 weeks and the study duration was longer than 12 weeks, the effect sizes dropped to − 0.02 and 0.02 (extremely small effect size), respectively.
Our meta-analysis included a larger number of studies than the previous and recent systematic reviews that assessed the efficacy of probiotics for IBS [6,7] because we used the transformation between odds ratios and SMDs, which has been suggested and reported in the Cochrane handbook [16] and methodological reports [9,14] to increase the statistical strength of the meta-analysis.The ability to include more studies might also be attributed to the application of the three-level meta-analysis model, which prompts the estimation of the general effect of the probiotics on
Placebo; q.d.; 6 weeks Relief of IBS symptoms score Sadrin et al. [81] France, RCT, Multicenter  [86] Germany  [88] France This column lists numerous abbreviations for the probiotic components which were named as the article reported or according to the nomenclature or naming system for bacteria suggested by the International Committee on Systematics of Prokaryotes.
IBS and confirms a medium effect size of probiotics on the improvement of global IBS symptoms (SMD 0.56)suggesting a possible generalization to routine practice.Shorter treatment duration or study duration being associated with larger treatment effects of probiotics was one of the major findings of our meta-analysis.A network meta-analysis published in 2022 reported that treatment duration could affect the efficacy of probiotics in the relief of abdominal pain and strain, and it showed that using Bacillus coagulans for 8 weeks was the most efficacious [89] .Dale and colleagues found that longer treatment duration might be associated with better efficacy in the treatment of IBS with probiotics [90] , while the other two studies showed that a shorter treatment duration of probiotics would be more efficacious [91,92] .Our meta-analysis, with a larger number of included trials and the inclusion of more efficacy data (through a third-level model), confirmed that shorter treatment duration was associated with a larger treatment effect.Several hypotheses were proposed for this phenomenon.First, trials with small sample sizes and singlecenter design were more likely to have short treatment Figure 2. The three-level meta-analysis concept and results.IBS, irritable bowel syndrome; SMD, standardized mean difference.Note: The figure shows the concept of the three-level meta-analysis model and the results of the three outcomes.The three-level meta-analysis was conceived and developed to solve the problems of correlated data and missing information, which were unsatisfactorily settled in the traditional meta-analysis model [9] .For example, the severity of global IBS symptoms was assessed by differential scales at differential time points, and in most circumstances in traditional meta-analyses, data from one specific scale measured at one time point would be selected, which induced loss of information.We adopted the three-level model to synthesize all data from the relevant scales measured at defined time intervals, and we provided a general effect size using the SMD.According to previous literature, an absolute value of SMD larger than 0.5 would indicate a medium size of effect [15] , meeting the standard of recommendation for clinical practice.Our meta-analysis demonstrated that all SMDs for the three outcomes exceeded that cut-off point of 0. duration and study duration, which could be the consequence of a shortage of study funding, so the larger treatment effects could be explained by small-study effects [93] .Second, the larger effect of probiotics on IBS might also be attributed to a powerful placebosimilar to the effect of sham device or sham acupuncture reported in previous studies [94][95][96] .In a meta-analysis investigating the magnitude of placebo response in IBS trials, short treatment duration was found to be associated with a large placebo effect [97] .Third, the moderator effect of short treatment duration might reflect the association between the severity of the IBS symptoms and the effect of the assessed intervention.Patients with severer IBS symptoms might report a smaller treatment effect of probiotics, and the physicians might be inclined to suggest a longer treatment duration, especially for those with refractory IBS [98] .
Regarding that, the diagnostic criteria of refractory IBS are difficult to define and the severity of IBS disease is determined by several factorshealth-related quality of life, psychosocial factors, healthcare utilization behaviors, and burden of illness [99] , so it is impossible to test this hypothesis based on the included trials in this meta-analysis since most of the included studies did not classify the disease severity owing to the lack of standard criteria.This informs that there is an urgent need for a standard scale to estimate the overall severity of IBS to minimize the heterogeneity caused by the study population in future meta-analyses on probiotics for IBS.Additionally, future RCTs are encouraged to report symptom severity of IBS using scales like IBS-SSS in baseline evaluation to facilitate subgroup or meta-regression analysis for clarifying the relationship between severity of IBS and probiotic treatment duration.Although we did not find an impact of different types of probiotics on the improvement of IBS symptoms, we found that Bacillus strains led to better improvement in abdominal pain than other strains and were significantly better than the Saccharomyces strain.This finding was consistent with a recent network meta-analysis comparing differential probiotics for the treatment of IBS [7] , which implies that the Bacillus strains might be developed for the treatment of functional abdominal pain and warrants further clarification.
Our study had several limitations.First, the certainty of the evidence was low because most of the included trials were classified as with some concerns or a high risk of bias.Many trials had some concerns in the randomization process (mainly the problem of the transparency of allocation concealment) and the measurement of the outcomes.Second, the large heterogeneity in the meta-analysis was also a concern.The heterogeneity might be caused by the difference in the study population and the intervention protocols.We ran the moderator analysis and confirmed that the duration of treatment and study, the study regions, and the types of outcomes might be the source of heterogeneity.Third, although the method of transforming between OR and SMD enlarged the sample size of the meta-analysis, it made the explanation of the results difficult for clinical practitioners, who might transform it back to the original scale by multiplying the SMD generated from the meta-analysis by the standard deviation of the specific scale [14] .Fourth, forty-eight reports were excluded for the unavailability of full-text copies, which were mainly abstracts of conference presentations and supplementary issues.These reports, known as grey literature, might be valuable for our metaanalysis and might change the conclusion of our study.Updated systematic review and meta-analysis might therefore be warranted while many of them were available with sufficient data for analysis.

Conclusions
Our meta-analysis suggested a medium short-term effect of probiotics on the improvement of global IBS symptoms and abdominal pain.We found that the treatment duration, study regions, the types of outcomes, and the types of probiotics might be major effect moderators, which warrants further investigation.

Figure 3 .
Figure2.The three-level meta-analysis concept and results.IBS, irritable bowel syndrome; SMD, standardized mean difference.Note: The figure shows the concept of the three-level meta-analysis model and the results of the three outcomes.The three-level meta-analysis was conceived and developed to solve the problems of correlated data and missing information, which were unsatisfactorily settled in the traditional meta-analysis model[9] .For example, the severity of global IBS symptoms was assessed by differential scales at differential time points, and in most circumstances in traditional meta-analyses, data from one specific scale measured at one time point would be selected, which induced loss of information.We adopted the three-level model to synthesize all data from the relevant scales measured at defined time intervals, and we provided a general effect size using the SMD.According to previous literature, an absolute value of SMD larger than 0.5 would indicate a medium size of effect[15] , meeting the standard of recommendation for clinical practice.Our meta-analysis demonstrated that all SMDs for the three outcomes exceeded that cut-off point of 0.5.

Table 1
Trial characteristics , body mass index; B.i.d, twice a day.CFU, colony forming units; FODMAP, fermentable oligosaccharides, disaccharides, monosaccharides and polyols; GSRS, gastrointestinal symptom rating scale; IBS, irritable bowel syndrome; IBS-C, constipation-predominant irritable bowel syndrome; IBS-D, diarrhea-predominant irritable bowel syndrome; IBS-M, mixed irritable bowel syndrome; IBS-U, un-subtyped irritable bowel syndrome; IBS-SSS, irritable bowel syndrome symptom severity score; IBS-QOL, evaluation of the irritable bowel syndrome quality of life; O.d, once every two days.Q.d, four times a day; RCT, randomized controlled trial; T.i.d, three times a day.VSL#3, a combination of three types of Bifidobacterium (Bifidobacterium longum, Bifidobacterium infantis, and Bifidobacterium breve).*This column is arranged according to years.† (I, P) refers to the sample sizes for probiotics and control groups, respectively, whereas the I refers to the probiotic group and the P refers to the control group.‡ BMI

Table 2
Moderator analysis of the outcome measurements.The P values were estimated as the reference categories being compared with the reference category. *