This systematic review and meta-analysis followed the PRISMA (Preferred Reporting Items for Systematic Review and Meta-analysis) guideline [11].

The PICOS criteria were applied to formulate the question: Population (adults aged 18 and older), Intervention (soluble fiber supplementation), Compression (not using soluble fiber supplementation), Outcome (changes of SBP and DBP), and Study design (parallel and crossover clinical trials).

Systematic literature searches for published articles were conducted in PubMed/Medline, Scopus and Web of Science without time and language restrictions through Aug 2022. A comprehensive description of the search strategy is provided in Table S1. In addition, we performed a reference list check of relevant articles, reviews, and meta-analyses to avoid missing any relevant literature.

We included studies based on the following inclusion criteria: (1) RCTs with either parallel or crossover design; (2) studies conducted on the adult population (≥ 18 years); (3) studies that involved comparison groups who received either a placebo without soluble fiber or an isolated soluble fiber treatment as part of controlled consumption studies or ad libitum supplementation interventions that were conducted in free-living subjects; and (4) studies that reported adequate baseline and follow up data in both treatment and control groups.

We excluded observational studies, animal or in-vitro studies, and those conducted on children, pregnant, or lactating women. In addition, studies without a control group, those that reported insufficient data on the selected outcomes in the soluble fiber or control groups, and trials that examined the effect of soluble fiber in combination with other components were not included. Additionally, trials that included soluble fiber supplementation as part of a dietary mixture or mixed within a dietary substance were excluded from analysis if the effects of soluble fiber alone could not be determined.

Abstracts and full texts of qualified studies were screened independently by two reviewers (SB and AG), who were blinded to the studies' authors or results. A chief reviewer (GA) was consulted to reach a consensus when necessary.

The following information was extracted from each eligible trial: first author's name, year of publication, study location, length of intervention, study design, characteristics of enrolled participants (numbers, mean age, sex, and health status of them), and soluble fiber intervention (form of adminstration, type, dosage, fermentability and viscosity), comparator and background diet. The mean and standard deviation (SD) of SBP and DBP before and after intervention alongside changes between baseline and post intervention were extracted. The mean and SD were extracted or calculated from available reported data [reported as 95% confidence intervals (CI), standard error (SE), median and (interquartile range (IQR)] using a standardized formula [12] at baseline and at the end of the respective treatment periods and were incorporated in the meta-analysis. When outcome measures were reported in figures/plots alone (e.g., no mean data reported), WebPlotDigitzer 4.5 was utilized to estimate equivalent numerical values. For multi-arm trials, intervention groups were isolated to determine the independent effects of soluble fiber supplementation across treatments. Paired analyses were conducted for all crossover trials [13].

Quality assessment was undertaken to assess the rigor of RCTs as determined by the risk of bias utilizing the Cochrane Risk of Bias Assessment Tool Evaluation of RCT quality (bias) was based on seven domains: participant randomization sequence generation, supplement allocation concealment, blinding of participants and research personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other bias sources. The risk of bias for each study was then rated as low (adequate information provided), unclear (if certain information was unclear or indeterminate), and high (if there was a serious concern in the criteria). The overall quality of studies was graded as good if there were low risk of bias for more than two items, fair if there were low risk of bias for two items and poor if there was a low risk of bias for less than two items [14].

The Grading of Recommendations Assessment, Development and Evaluation [15] method was employed to evaluate the certainty of the evidence for outcomes. The quality of the assessed evidence was rated as high, moderate, low, and very low. High grades suggest high confidence that the actual effect is commensurate with the estimated effect. Moderate grades suggest that the actual effect is likely to be close to the estimate of the effect; however, there exists a small possibility of substantial differences. A low grade suggests a greater likelihood that the true effect may be substantially different from the estimate of the effect, and very low grades suggest the true effect is likely different from the estimated effect. Further, RCTs with an initial high quality of evidence evaluation may be downgraded based on study limitations, including risk of bias inconsistency (substantial unexplained heterogeneity, I2 > 50%; p < 0.05) and imprecision (95% CI for effect estimates are wide or cross the minimally significant threshold difference for clinical benefit). Minimal thresholds for clinically important changes and consider indirectness of outcomes (primary outcomes presented are surrogate rather than patient-important outcomes [15] and other considerations (publication bias and dose–response gradient usage).

The mean and SD changes of SBP and DBP were used to calculate the pooled effect sizes. Studies that did not report the SD of the mean differences in each group required manual calculation as follows [16];

$$_=\sqrt_^+_^-2\times R\times _\times _}$$

where R represents a correlation coefficient of 0.5 [17]. Because pretest to posttest R was not reported in RCTs, an R-value of 0.5 was utilized throughout this meta-analysis [18]. Also, we conducted sensitivity analyses using different correlation values (0.25 and 0.75) and report and interpretation meta-analysis results. Due to the observed variation in study treatments and protocol, a random-effects model was used to estimate mean difference (MD) and 95% confidence interval (CIs) [18].

We conducted meta-regressions to assess outcomes in relation to the following prespecified factors: intervention duration, daily dose of soluble fiber, soluble fiber category (type, fermentability and viscosity), gender, mean baseline BMI, mean age, and mean baseline SBP and DBP and health status. Factors were selected on the basis of the likelihood of influence on outcomes of interest. The intervention duration was defined as the time period (weeks) when participants received the treatment or placebo; the soluble fiber dose was defined as daily grams of fiber treatment as provided during the intervention; and the categorization of soluble fibers was based on their physicochemical properties (type, viscosity, and fermentability) [19]. Mean BMI was categorized to obese (≥ 30 and non-obese < 30), mean baseline SBP and DBP was categorized to hypertention (SBP ≥ 130 mmHg and DBP ≥ 80) [20], classified trials based on mean age (equal and more than 50 years and bellow 50 years), and according to different health status (individual with hypertension, metabolic syndrome, hypercholesterolemia, hyperlipidemia, diabetes, overweight-obese, and otherwise healthy).

We assessed the between-study heterogeneity of soluble fiber intervention effects using the I2 statistic. The following I2 interpretive categories were used: bellow I2 < 50% was considered “moderate”, (50% <  = I2 <  = 75%), was considered “substantial” heterogeneity and “considerable” heterogeneity (I2 >  = 75%) [18]. Sensitivity analyses assessed the impact of each study on the pooled effect size [21]. The risk of publication bias was assessed using visual inspection of the funnel plot and Begg’s test and Egger's test [22]. Based on Crippa and Orsini's method, the mean and corresponding SD of change in liver function test, and the number of participants in each study arm, was used to conduct a random-effects model for each 5 g/d increase in soluble fiber supplementation in the intervention group on changes in liver function test [23]. Additionally, we conducted a dose–response meta-analysis to clarify the shape of the effect of different doses of soluble fiber on blood lipids [24]. All statistical tests were performed using the Stata software (Version 17.0, Stata Corp, College Station, TX), and a P-value less than 0.05 considered significant.