INTRODUCTION

Ulcerative colitis (UC) is a chronic, relapsing, and remittent inflammatory disease of the colon occurring at the interface between luminal contents and the mucosal immune system. Increasing evidence implicates the colonic microbiome in the pathogenesis of UC, with microbial antigens contributing to aberrant immune activation. Patients with UC have less diverse microbiota compared with healthy subjects whatever the level of disease activity (1). This is predominantly attributable to a loss of immune-protective symbionts and increase in proinflammatory bacteria, particularly overabundant species including Escherichia coli and other Enterobacteriaceae (2). The usual medical therapies targeting the microbial environment (antibiotics, probiotics, and prebiotics) are not effective enough or ineffective and not recommended to induce or maintain remission (3). Fecal microbiota transfer (FMT) has been shown to be clinically effective in patients with Clostridioides difficile infection, where modification of the colonic ecosystem alters the disease process (4–6). Recently, FMT has been extensively studied in active UC patients in randomized controlled trials (RCTs) with various protocols. Published results from systematic reviews and meta-analysis have been largely positive (7,8), raising hope for new promising therapeutic approaches to achieve remission in active UC (9–11). The methodologies and results of these studies are not consistent, with product preparation, dosing, and route of administration representing sources of heterogeneity (7). By conceptually altering the disease process by modifying the colonic ecosystem, FMT may be expected to restore homeostasis of biochemical and antigenic drivers of immune-mediated diseases.

The objective of this systematic review with meta-analysis was to compare the efficacy of multidonor (MDN) and single-donor (SDN) strategies for FMT product preparation in achieving response to treatment of active UC. In this study, the intervention was considered to be FMT. The strategies differed in that at any individual treatment timepoint, a patient with UC was treated with a product manufactured from a single donor (SDN strategy) and from at least 2 different donors (MDN strategy).

METHODS

The systematic review and meta-analysis were conducted as per the Preferred Reporting Items for Systematic Reviews and Meta-Analyses method (12). The protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO, registration number: CRD42020210649). The statistical analysis plan was locked before statistical analysis.

Search strategy and selection criteria

Included studies had to report at least 1 efficacy endpoint and meet the following eligibility criteria.

Patients: diagnosis of UC, irrespective of follow-up duration, concomitant medication, sex, age, or language; these variables were recorded and used as meta-regressive moderators; Interventions: considered to be microbiotherapy based on an ecological and complete ecosystem such as FMT with product administered at any individual treatment timepoint produced by SDN/MDN strategies; Comparators: placebo (control treatment): Autologous FMT: patient microbiota reconstituted at the treatment timepoint from patient's own stool donation at study inclusion. This perfectly controls the process and identifies the benefit of exogenous microbiota. Changes in patient's gut physiology between donation and treatment timepoint could introduce bias, in which case the autologous treatment could be beneficial, have no impact, or have a negative impact on the patient; Saline buffer control: saline buffer administered into the patient gut to control the enema process is the best inert control compared to heterologous FMT. The quality of the blinding process is not as good for saline buffer, but no activity bias is expected.

Two investigators independently searched the current literature for articles, books, and abstracts related to the efficacy and safety of microbiota-derived drugs, irrespective of language, and checked selected references manually. The investigators searched scientific articles on Scopus, PubMed, Web of Science, and Registers, and patents on Orbit Intelligence, for documents containing clinical data assessing FMT in inflammatory bowel disease, and identified records of interest by searching websites of organizations and citation searching (see Supplemental Digital Content 1, https://links.lww.com/CTG/A910). The last search was completed on 28 June 2022.

Two investigators (B.L. and M.F.) independently assessed abstracts of selected references for eligibility; any disagreement was resolved by a third investigator (P.L.). Eventually, studies reporting on use of FMT in patients with UC were selected. If several articles reported results of the same clinical trials, the article reporting the most extensive information was selected. Potentially relevant articles were evaluated in more detail using predesigned forms to assess eligibility independently, according to predefined criteria. Studies were excluded after endpoint evaluation and because included patients did not meet the eligibility criteria for this analysis. Selected records were retrieved and further assessed for eligibility. The selection process is summarized in Figure 1 (12).

Figure 1.: PRISMA flow diagram showing the study selection process. n, number of records; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses. From Page et al. (12).Figure 2.:

Forest plot. CI, confidence interval; MDN, multidonor; SDN, single donor; RR, risk ratio. Comparison of SDN and MDN was performed by using placebo as the null reference. RR SDN/placebo and MDN/placebo.

Study selection

B.L. and P.L. first read the Material and Methods sections of each selected article to decide whether the study met the eligibility criteria for the meta-analysis and reached a consensus through face-to-face discussion. Subsequently, the authors attempted to identify articles based on the same raw data. The reasons for excluding studies were summarized and documented. Finally, only studies with a control arm were included in the review and meta-analysis.

Data collection

All data from publications were systematically reviewed. All authors evaluated each publication. Several tables were constructed including a summary table of study characteristics (author, publication date, interventions, study design, and endpoints) (Table 1). After discussion between authors and preliminary reviews, a list of endpoints and moderators was set up. Each author built a data matrix containing results reported for the planned endpoints; these were compared and reconciled.

Table 1. - Study characteristics Study Country Year Design Mayo12 Antibiotica PEGb Durationc FMTd Notese Routef Paramsothy et al. (13) Australia 2017 RCT 8 0 0 8 40 2 1 Moayyedi et al. (5) Australia 2015 RCT 6.4 0 0 7 6 4 1 Costello et al. (14) Australia 2019 RCT 7 0 1 8 3 2 1 Sood et al. (15) India 2019 RCT 6 0 1 48 7 4 1 Rossen et al. (6) Netherlands 2015 RCT 6 0 0 12 2 4 2 Crothers et al. (16) United States 2021 RCT 6.3/6.7 1 0 12 84 1 + 3 Haifer et al. (17) Australia 2021 RCT 5/7 1 0 8 56 3 Pai et al. (18) Canada 2021 RCT PUCAI 0 0 6 12 1 Březina et al. (19) Czech Republic 2021 RCT 6 0 0 6 10 1 Sarbagili Shabat et al. (20) Italy and Israel 2022 RCT 6 (SCCAI) 0 1 2 3 1 Subhadra (21) United States 2016 CC 8 0 0 7 8 3 Kump et al. (22) Austria 2017 CC 8 1 1 13 5 1 Scaldaferri et al. (23) Italy 2015 CC 6.4 0 0 12 3 3 1 Ishikawa et al. (24) Japan 2019 CC 8 1 0 4 1 3, 4 1 Borody et al. (4) Australia 2003 C 6.4 1 1 6 5 4 1 Angelberger et al. (1) Austria 2013 C 7.3 1 0 12 3 4 2 Kump et al. (25) Austria 2013 C 8.9 0 0 12 1 4 1 Kunde et al. (26) United States 2013 C P 0 0 4 5 1 Ren et al. (27) China 2015 C 7.3 0 0 4–30 1 4 2 Suskind et al. (28) United States 2015 C P 1 0 12 1 2 Damman et al. (29) United States 2015 C 6.2 0 0 12 1 4 1 Wei et al. (30) China 2016 C 5.8 1 1 12 1 1 Vermeire et al. (31) Belgium 2016 C 8 0 0 8 2 4 2 Karakan et al. (32) Turkey 2016 C - 0 1 12 1–6 1 Mizuno et al. (33) Japan 2017 C 8 0 1 12 1 4 1 Nishida et al. (34) Japan 2017 C 6 0 1 8 1 1 Jacob et al. (35) United States 2017 C 7.5 0 1 4 1 2 1 Uygun et al. (36) Turkey 2017 C 10 0 0 12 1 4 1 Karolewska-Bochenek et al. (37) Poland 2017 C PUCAI 0 0 4 8 2 Goyal et al. (38) United States 2018 C PUCAI 0 0 26 1 2 Tian et al. (39) China 2019 C 5 0 1 18 5 1 2 Steube et al. (40) Germany 2019 C 8.3 1 0 12 600 1, 2 3 Cold et al. (41) Denmark 2019 C 7 0 0 12 1,250 2, 4 3 Ding et al. (42) China 2019 C 10.3 0 0 12 1–3 2, 4 2 Adler et al. (43) United States 2019 C 8 0 0 6 60 3 Sood et al. (44) India 2019 C 8.9 0 0 14 7 1 Chen et al. (45) China 2020 C 5.9 0 1 12 3 1 Chen et al. (46) China 2020 C 10.2 0 1 - 3 - Dang et al. (47) China 2020 C - 0 0 - 1 1 Ren et al. (48) China 2021 C 9.6 0 1 18 2 1 Seth and Jain (49) India 2022 C 6.4 0 1 12 3 1 Smith et al. (50) United States 2022 C 6.5 0 0 6 6 3 Zhang et al. (51) China 2022 C 7.0 0 1 6 1 1 5-ASA, 5-aminosalicylic acid; AFM, amoxycillin + fosfomycin + metronidazole; C, cohort; CC, controlled cohort; FMT, fecal microbiota transfer; Mayo12, Mayo Score (range 0–12) at baseline; MDN, multidonor; PEG, polyethylene glycol (bowel preparation); PUCAI, Pediatric Ulcerative Colitis Activity Index; RCT, randomized controlled trial; SCCAI, Simple Clinical Colitis Activity Index; SDN, single donor; ST, standard therapy; Trt1, active FMT treatment group; Trt2, control treatment group (autologous product used only in the study conducted by Rossen et al. (6)); TRT, number of donors per FMT product (1 = SDN, >1 = MDN).

aNumber of responders and total number of patients reported on the left and right columns, respectively.

bAntibiotic/PEG used: 0 (none); 1 (1 or more).

cDuration of follow-up in weeks.

dNumber of FMT administrations during the study.

e1: Mayo score replace by the probability of score <2 (considered as relief). Pnorm(2,mean,sd)*n where n = number of patients. 2: Studied treatment where >1 donor is considered as MDN; 3: AFM assimilated with placebo ST, standard therapy; 4: Baseline value was used. When not available, estimation based on protocol selection calculated as the truncated mean of distribution assumed to be N(6,2) following the expression sum(dnorm(q:12,6,2)*q:12)/sum(dnorm(q:12,6,2)), where M and m are the minimum and maximum values in the selection. z<-seq(m,M,.01); sum(dnorm(z,6,2)*z)/sum(dnorm(z,6,2)).

f1 = Lower route; 2 = upper route; 3 = capsules.

Data were extracted using a predetermined computer data entry template including author, publication year, trial start date, countries where the trial was conducted, treatment group, sample sizes (randomized/analyzed, by group), inclusion/exclusion criteria, mode of administration, dose and duration of treatments, outcomes, and source of funding.

Collected data were

1. Proportions originating from categorical or binary variable requiring the category count and the group sample size; 2. For continuous data, mean (SD) or median (interquartile range), and sample size by group. Outcome assessment

After preliminary examination of eligible studies, the investigation was limited to the main endpoint, concentrating on the therapeutic response as a success/failure binary endpoint as reported by the investigator (remission in most cases). Very few other outcomes were consistently reported in the studies.

Data availability

All data generated/analyzed during this study are included in this published article and its supplementary information.

Meta-analysis

All studies were analyzed for certainty of evidence based on the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) approach (52). The Cochrane Collaboration's risk-of-bias tool (53) was used to evaluate bias and filled in by 3 reviewers on the 14 controlled studies (Table 2). Each study was separately analyzed for risk of bias or indirectness. Heterogeneity (using a random model), imprecision, and publication bias (using funnel plots) were evaluated at the meta-analysis level. Meta-regression was performed using evidence level as moderator.

Table 2. - Summary of certainty of evidence including risk of bias and directness for each study Study Bias risk Directness ΣBD Blinding ITT Selected Design ΣBR P I C O ΣD Rossen et al. (6) + + + + 0 + + + + 0 H Moayyedi et al. (5) + + + + −1 + − + + 0 H Scaldaferri et al. (23) − + + + −1 + + + + −1 M Kump et al. (22) − + − − −1 + + + + −1 M Ishikawa et al. (24) − − + + −1 + + + + −1 M Sood et al. (15) + + + − −1 + + + + −1 H Subhadra (21) − + − − 0 + + + + 0 M Paramsothy et al. (13) + − + + 0 + − + + 0 H Costello et al. (14) + + + + 0 + + + + 0 H Crothers et al. (16) + + + − −1 + + + - −1 M Haifer et al. (17) + + + + 0 + + + + 0 H Pai et al. (18) + + + − −1 + + + - −1 M Březina et al. (19) − + + − −1 + + + + 0 M Sarbagili Shabat et al. (20) − + + + −1 + − + - −1 M Heterogeneity I 2 = 0% A nonsignificant heterogeneity was demonstrated + Precision OIS = 492 Sample size (656) exceeds the OIS, and all results were significant + Publication bias P = 0.51 No asymmetry was observed on the funnel plots + Total Results with high level of certainty of evidence considered

∑BR, summary of bias risk; ∑BD, summary of bias risk and directness; ∑D, summary of directness; H, high certainty of evidence; ITT, intention to treat; M, moderate certainty of evidence; n, number of patients; OIS, optimal information size; PICO, patient selection, intervention, control, and outcomes.

The significance of the indirect estimate of the difference between MDN and SDN treatments was sought through a network approach (54). The risk ratio (RR) was calculated as the main calculation of effect size. A random-effects model was assumed to be most likely where difference may be expected among studies, and the fixed model was performed for sensitivity purposes. All results were compared with an alternative fixed statistical model, and heterogeneity tests were used. Correlated pairwise comparisons in multiarm studies were corrected by the weight reduction approach (55). Model fit was assessed by generalized Cochrane Qt (56,57). Treatment ranking by P-scores measured the extent of certainty that any one treatment was better than another, averaged over all competing treatments (58). Statistical analyses were performed using R statistical packages (version 3.2.4) and the meta-library Netmeta (59).

Data values provided as SEM were converted into SDs as per the formula SD = SEM*sqrt(n). For endpoint calculation and effect size, given the heterogeneity of the studies in their clinical definition, the following transformations were also needed for direction and measurement.

1. Severity scores (higher values meaning higher severity) were converted into improvement scores; 2. Two alternative methods were used to aggregate scales based on quantitative values or proportions: