Open access
how judgments are reached. Any disagreements that arise between the reviewers will be resolved by consensus or by consulting a fifth reviewer. Response options (ie, judg- ments about risk) are low risk, high risk and unclear risk. Hence, studies will not be excluded on the results of crit- ical appraisal; however, study quality will be considered when analysing and interpreting results.
standard Forest plots for each domain, and the heteroge- neity will be evaluated by inconsistency index (I 2 statistic).
Subgroup, meta-regression and sensitivity analyses To investigate possible sources of trial heterogeneity, addi- tional analyses of the main outcome will be performed. We will stratify the available trials according to trial characteristics: (1) type of diagnosis, (2) type of inter- vention and (3) type of concurrent anti-inflammatory treatment, respectively, using univariate REML-based meta-regression. We will also explore the statistical effect of trial design according to the following subgroups: (I) Trials at risk of bias (on an item-by-item basis); (II) trials with shorter vs longer follow-up and (III) Comparator (active vs placebo). Dealing with missing data When there are missing data, we will attempt to contact the authors of the study to obtain the relevant missing data. Important numerical data will be carefully evalu- ated. If missing data cannot be obtained, an imputation method will be used. Anticipating ROB-ME (risk of bias due to missing evidence), the structured approach for assessing the risk of bias that arises when entire studies, or particular results within studies, are missing from a meta-analysis because of the p-value, magnitude or direc- tion of the study results. 67 To visualise the potential for bias in the meta-analysis result, we will generate Forest plots displaying trials with results, along with infor- mation on trials that did not report specific outcome domains. Rating certainty of the evidence The GRADE system will be used 79 to summarise the certainty of evidence on an outcome-by-outcome basis as high, moderate, low or very low. 80 Quality of evidence will not be downgraded for risk of bias if subgroup anal- ysis indicates no association of treatment effects with risk of bias. Additionally, when there are a minimum of 10 studies for meta-analysis, we will attempt to evaluate the risk of publication bias through visual assessment of funnel plot asymmetry. Presentation of results Depending on the type of outcome, the quantitative synthesis will encompass both ORs and SMD with corre- sponding 95% CIs for each outcome domain. Forest plots will be used for graphical representation. Exact p values will be reported, and statistical significance will be defined as p<0.05. Sensitivity and subgroup analyses will be reported alongside the main analyses to allow full interpretation of the robustness of findings. Protocol deviations Any modifications to this protocol during the study’s execution will be documented in the final manuscript and on PROSPERO.
Data synthesis Primary and secondary endpoint analyses
Meta-analyses will be conducted separately for each primary and secondary outcome domain. For trials with multiple intervention arms, the number of patients in the shared control group will be divided by the number of comparisons to avoid double-counting and to produce correct estimates with appropriately increased standard errors. Binary outcomes 66 : Most primary endpoints are antic- ipated to be binary. ORs will be computed such that a result greater than one indicates greater effectiveness of microbiota-targeted therapeutics compared with placebo/ sham. In trials utilising continuous outcomes instead of discrete ones, if the means and SD of the placebo and treatment groups on these trial outcomes are available, they will be transformed into the corresponding OR using a method outlined by Chinn. 76 Whenever feasible, all participants who drop out will be considered as non- responders, adhering to the intention-to-treat principle. ORs are also calculated for withdrawals, withdrawals due to adverse events, number of SAEs and deaths. Continuous outcomes 77 : For continuous outcome domains, standardised mean differences (SMDs) will be used to combine results measured with different instru- ments. Mean differences (MD) at follow-up will be used when the change from baseline is not available. SMDs are calculated by dividing the difference in mean values by the pooled SD for the given outcome; a correction will be applied by default by calculating Hedges’s g 78 and the variance (SE 2 ) will be calculated based on the SMD and number of patients in each group. This calculation involves dividing the difference between the intervention and comparator mean changes in each trial (ie, the MD by the estimated within-group SD for that trial). Accord- ingly, an SMD <zero will indicate a beneficial effect of the experimental intervention (eg, a reduction in the patient’s global assessment) compared with control comparator. The SMD will be converted to ORs by the conversion proposed by Hasselblad and Hedges: ln(OR)= SMD π √ 3 Evidence synthesis methods All meta-analyses will be performed using R Foundation for Statistical Computing or Stata, with effect estimates pooled using random-effects models based on restricted maximum likelihood (REML). Fixed-effect models will be used in sensitivity analyses to test the robustness of results. Statistical heterogeneity will be investigated using
7
Kragsnaes MS, et al . BMJ Open 2025; 15 :e101593. doi:10.1136/bmjopen-2025-101593
Powered by FlippingBook