The protocol for this systemic review was registered with PROSPERO (International Prospective Register of Systemic Reviews, CRD42041241243). The aim of this study is to identify the effect of bariatric intervention on MASLD, determined by histologic findings and/or MRI findings. This study was conducted along with others that identified the effect of exercise intervention on MASLD [16], or validated the accuracy of noninvasive scoring system in assessing liver fibrosis [17]. Study selection followed the Preferred Reporting Items for Systemic Review and Meta-analyses extension for Diagnostic Test Accuracy (PRISMA-DTA) statements [18]. We searched the Ovid-MEDLINE, EMBASE, KMBASE, Korean Studies Information Service System (KISS), and Cochrane library, covering the period from database inception, through January 1, 1997 to October 31, 2023. We only included papers published in English. The search keywords used were: metabolic dysfunction-associated steatotic liver disease, MASLD, metabolic dysfunction-associated steatohepatitis, MASH, nonalcoholic fatty liver disease, nonalcoholic fatty liver, NAFLD, nonalcoholic steatohepatitis, NASH, obesity/obese, bariatric surgery, intragastric balloon, and gastric banding.

Two researchers (Y.C. and S.H.K.) reviewed the screened papers independently, by their titles and abstracts, in the first screening. In the secondary screening, full text of the papers that passed the first screening was examined, providing specific reasons for any exclusions. During the process of literature selection, any discordance between the two clinical researchers was resolved through mutual consultation. In cases where a consensus could not be reached, the final decision was made by the research committee through a formal meeting. The study design, outcome measure (histology vs. MRI-PDFF), sample size, intervention, mean age, duration of follow-up, and BMI changes in each study were extracted as basal characteristics. The entire search process was administered by a professional statistician (M.Y.C. and D.L.).

The following criteria were required for studies to be selected: (1) patients who underwent bariatric surgery or intragastric balloon/gastric banding for obesity; and (2) those who were diagnosed with histologically proven MASLD or MASH designs included randomized controlled trials, cross-sectional studies, and cohort studies, both prospective and retrospective. Studies were excluded based on the following criteria: (1) case reports; (2) case series, in which less than five patients in total were involved; (3) reviews; (4) cell or animal studies; (5) studies on chronic viral hepatitis, such as hepatitis B or hepatitis C; (6) studies on human immunodeficiency virus; (7) studies on population with significant alcohol consumption; (8) studies with no histology result provided; or (9) pediatric studies.

Studies that reported at least one histologic or MRI variable were included in the analyses. All included studies presented baseline and follow-up BMI. Histologic variables are as follows: (1) NALFD activity score (NAS), which assesses the severity of inflammation and hepatocellular injury in liver biopsy [19], (2) Histologic features of MASH, including steatosis, lobular inflammation, ballooning degeneration, and fibrosis, and (3) worsening MASH. MRI variable includes the change in steatosis and BMI.

We performed the quality assessment of the final selection of papers, based on key questions, using RoBANS tool for non-randomized controlled studies and Cochrane risk-of-bias tool for randomized controlled study. QUADAS-2 evaluation tool was used to evaluate the diagnostic accuracy. The following factors were checked: comparability of participants, selection of participants, confounding variables, measurement of exposure, blinding of outcome assessment, outcome evaluation, incomplete outcome data, and selective reporting. Publication bias was evaluated by a funnel plot for studies investigating populations with more than ten individuals. The certainty of the evidence was evaluated by Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach, which considers factors such as risk of bias, inconsistency of results, indirectness of evidence, imprecision, and publication bias.

The process of meta-analysis with paired ratio of means data and proportion data involved the estimation of Hedge’s corrected standardized mean difference, under the assumption of a random-effects model, to account for heterogeneity across the studies. The pooled effect estimates between pre- and post-operation within each study’s specific time frame were derived to measure the effect size, using Freeman-Tukey variant of the arcsine square root transformation, along with a 95% confidence interval (CI). To address the differences in follow-up periods across the selected studies, we categorized the studies according to the length of the follow-up time in further analyses. Analysis of each pooled data is shown in a forest plot. Heterogeneity of the studies was assessed by Cochrane’s Q test; I² statistics higher than 50% was considered indicative of significant heterogeneity, and more than 75% was considered “high” heterogeneity. Next, sensitivity analyses and meta-regression were conducted to assess the influence of other factors on diagnostic accuracy across the studies with significant heterogeneity. Cochrane Review Manager version 5.4 (London, UK), R Foundation for Statistical Computing version 4.1.2 (Vienna, Austria), and GRADEpro GDT were used in the analysis, with a significance level set at a P-value less than 0.05.

From January 1, 1997 to October 31, 2023, a total of 1,546 papers were identified through preliminary data searching using search keywords described in the Methods section. The numbers of papers from each database are as follows: Ovid-MEDLINE (n=593), KMBASE (n=5), EMBASE (n=845), KISS (n=11), and Cochrane library (n=92). After removal of duplicates (n=1,069), 246 papers were excluded based on the title only. In the first screening, two researchers independently reviewed the screened papers by their titles and abstracts (n=231), followed by exclusion of 156 papers after review. Furthermore, we thoroughly examined the full-text articles of the remaining studies (n=75), resulting in an additional exclusion of 46 studies. Finally, thirty studies were incorporated into this analysis (Fig. 1). Out of all these papers, twenty-four were evaluated through histological examination while the remaining six were assessed using MRI. General characteristics of the selected studies are presented in Table 1. A total of 3,134 patients from the studies were analyzed. It was noteworthy that only two were randomized controlled studies [20], whereas others employed a cross-sectional design. To address the difference in follow-up duration, we categorized the data into three time periods, as follows: 3–6 months, 12–24 months, and 36–60 months, for further analyses.

A total of thirty studies showed a significant reduction in BMI except for one randomized controlled trial with small sample size [20]. BMI was reduced by 19% at a rate of 0.81 (95% CI, 0.71–0.92) at 3–6 months, by 28% at a rate of 0.72 (95% CI, 0.68–0.76) at 12–24 months, and by 27% at a rate of 0.73 (95% CI, 0.68–0.79) at 36–60 months after intervention (Fig. 2A). The overall NAS, before and after bariatric intervention, with the ratio of their mean values, was examined in seven studies (Supplementary Table 1). NAS was reduced by 60% at a rate of 0.40 (95% CI, 0.30–0.54) at 3–6 months, by 40% at a rate of 0.60 (95% CI, 0.40–0.89) at 12–24 months, and by 50% at a rate of 0.50 (95% CI, 0.35–0.70) at 24–60 months (Fig. 2B). In the study conducted by von Schönfels et al. [21], post-NAS revealed a value of 0, rendering the calculation of confidence interval unattainable. The results of the histology were evaluated according to Brunt’s criteria and Kleiner score [19]. Nineteen studies revealed that the proportion of patients with steatosis decreased by 44% at 3–6 months, 37% at 12–24 months, and 29% at 36–60 months (Fig. 2C); lobular inflammation by 33% at 3–6 months, 36% at 12–24 months, and 51% at 36–60 months (Fig. 2D); ballooning degeneration by 20% at 3–6 months, 38% at 12–24 months, and 18% at 36–60 months (Fig. 2E). Remarkably, the proportion of patients with stage 2 fibrosis or higher (≥F2), or with more advanced form of MASLD, was found to decrease by 18% at 12–24 months, and by 17% at 36–60 months after intervention, compared to that before the operation (Fig. 2F, Supplementary Table 2). Most of these analyses revealed the I² statistics exceeding 75%, signifying substantial heterogeneity across the studies. Subsequent sensitivity analysis and meta-regression were conducted; however, the source of heterogeneity remained unclear (data not shown).

Six studies showed significant reduction of intrahepatic fat in MRI-PDFF at six months after bariatric intervention (Table 2). Ratio of the means of pre-operative to post-operative MRI-PDFF was 0.28 (95% CI, 0.24–0.33), which implied that 72% of intrahepatic fat was reduced after intervention with 21% of BMI reduction (Fig. 3). We observed a borderline heterogeneity in BMI, but no heterogeneity was identified in steatosis measured by MRI-PDFF.

Two studies reported mortality rate after the bariatric surgery, with pooled mortality rate of 13% (95% CI, 0.07–0.21). The other three studies reported cases in which MASLD worsened based on liver histologic examination after the surgery, with a pooled proportion of 6% (95% CI, 0.04–0.09) (Supplementary Table 3).

Out of a total of thirty studies, most studies showed low risk of bias, whereas a few showed moderate/high (Supplementary Figs. 1 and 2). To visually inspect the potential publication bias, we also created funnel plots for studies investigating populations with more than ten individuals where BMI was measured within 12–24 months. We found significant publication bias (Supplementary Fig. 3). To address this bias, we applied trim and fill method, confirming the correction of publication bias in the modified result. Notably, the adjusted effect size values remained consistent with the trend observed in the analysis using raw values. For populations with a follow-up period of 12–24 months, the GRADE assessment was used to rate the quality of evidence (Supplementary Table 4).