This evidence-based analysis followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) 2020 statement (21) and was prospectively registered in PROSPERO (CRD42024513246). We systematically searched the Cochrane, Embase, PubMed, and Web of Science databases up to 31 September, 2024. The search keywords: fish oils, Metabolic dysfunction-associated steatotic liver disease. The search strategy was ((“Fish Oils”[Mesh]) OR ((((((Oils, Fish) OR (Fish Oil)) OR (Oil, Fish)) OR (Fish Liver Oils)) OR (Liver Oils, Fish)) OR (Oils, Fish Liver))) AND ((“Non-alcoholic Fatty Liver Disease”[Mesh]) OR (((((((((((((Non alcoholic Fatty Liver Disease) OR (MASLD)) OR (Nonalcoholic Fatty Liver Disease)) OR (Fatty Liver, Nonalcoholic)) OR (Fatty Livers, Nonalcoholic)) OR (Liver, Nonalcoholic Fatty)) OR (Livers, Nonalcoholic Fatty)) OR (Nonalcoholic Fatty Liver)) OR (Nonalcoholic Fatty Livers)) OR (Nonalcoholic Steatohepatitis)) OR (Nonalcoholic Steatohepatitides)) OR (Steatohepatitides, Nonalcoholic)) OR (Steatohepatitis, Nonalcoholic))). The search results from four databases are presented in Supplementary Table S1. This trial had no language or geographical restrictions.

Inclusion criteria were as follows: (1) The study design is RCT; (2) Participants diagnosed with MASLD; (3) Intervention group received fish oil supplementation, while the control group received placebo or other treatments. Title and abstract screening for eligibility was independently conducted by two authors (L.K.Z and D.M.S). Disagreements were resolved by consulting a senior author (H.Q.B). Reviews, letters, editorial comments, case reports, conference abstracts, non-human studies, unpublished articles, those with incomplete data, and non-English articles were excluded.

Two authors (L.K.Z and D.M.S) independently screened literature and extracted data from the included trials, including the first author’s last name, number of participants, publication year, country, and outcome data for both intervention and control groups. Any disagreements were resolved by a third investigator (H.Q.B.) for a final decision. The outcomes focused on were: (1) biochemical markers, including serum markers of liver injury (ALT, AST, GGT and CK18-M30) and lipid profiles (TC, TG, HDL and LDL), adiponectin, and UA; (2) insulin, HOMA-IR and FBS; and (3) anthropometric parameters, such as obesity estimated by BMI, waist circumference, hip circumference, and WHR.

Two authors (L.K.Z and D.M.S) used tool (22) to assess the risk of bias in the included studies, with the third author (H.Q.B) responsible for confirming the judgment results. RCTs were assessed for high, low, or unclear risk of bias in six domains: randomization method, allocation concealment, blinding, completeness of results data, selective reporting of results, and other sources of bias. If a study did not provide data, it was rated as having an unclear risk of selective reporting bias (see Figures 1, 2).

The methodological quality assessment was based on JADAD score (23), including the following: sequence generation, allocation concealment, blinding, withdrawals and drop outs, and randomization efficacy. The evaluation process was independently performed by two of the authors (L.K.Z and D.M.S). Disagreement was resolved by third author ((H.Q.B; Supplementary Table S2).

All analyses were conducted using Review Manager software version 5.4 (Nordic Cochrane Center, the Cochrane Collaboration, 2020). We anticipated clinical heterogeneity, so we chose a random-effects model. Dichotomous variables are presented as risk ratios (RR) with their 95% confidence intervals (95% CI). Continuous variables are expressed as standardized mean differences (SMD) with their 95% CI. Statistical heterogeneity between studies was assessed by calculating the I² statistic. An I² value greater than 50% was considered to indicate significant heterogeneity (24), a random-effect model was used to estimate the combined SMD when significant heterogeneity was detected (I² > 50%). Otherwise, the fixed-effect model was applied. A p-value less than 0.05 was considered statistically significant. To assess the impact of individual studies on the pooled outcomes exhibiting notable heterogeneity, we further conducted one-way sensitivity analyses. The existence of publication bias was visually assessed using funnel plots generated in Review Manager 5.3 (Cochrane Collaboration, Oxford, United Kingdom). Additionally, Egger’s regression tests (25) were conducted in Stata 12.0 (Stata Corp, College Station, TX, United States) to further evaluate potential bias in outcomes with at least 10 included studies. A p-value less than 0.05 was deemed statistically significant, indicating the presence of publication bias. Our study conducts subgroup analysis based on the dosage of fish oil and the duration of intervention to explore the stability of the results and potential sources of heterogeneity.

After excluding duplicate literature, 623 literature identified in our research. Figure 3 shows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses flowchart. After screening of titles and abstracts, 616 records were excluded. Seven RCTs, comprising 439 participants, were included after reviewing the full texts. Of these articles, there were 340 males and 99 females, with a minimum sample size of 34 and a maximum sample size of 74, a minimum mean age of 33.6 years and a maximum of 60.88 years, and a minimum mean body mass index of 26.0 and a maximum of 33.3. The study lasted for a minimum of 12 weeks and a maximum of 1 year. The characteristics of the included RCTs (26–32) were presented in Table 1.

The analysis involved data from 7 RCTs for AST (26–32), 6 for ALT and GGT (27–32), and 2 for CK18-M30 (26, 27). Significant statistical heterogeneity was observed among studies for AST (I² = 62%, p = 0.007) and CK18-M30 (I² = 88%, p = 0.005). The meta-analysis revealed that compared to the control group, the fish oil group exhibited a significant improvement in AST (SMD: −0.29, 95% CI: −0.48 to −0.10), whereas no significant differences were observed in ALT (SMD: −0.15, 95% CI: −0.45 to 0.15), GGT (SMD: −0.07, 95% CI: −0.26 to 0.12), or CK18-M30 (SMD: −0.74, 95% CI: −1.95 to 0.47; see Figure 4).

The meta-analysis included data from five RCTs for HDL (27, 29–32), six for LDL (26, 27, 29–32) and total cholesterol (TC) (26, 27, 29–32), and seven for triglycerides (TG) (26–32). Significant heterogeneity was observed among studies for total cholesterol (TC: I² = 69%, p = 0.002). The fish oil group exhibited a significant improvement in triglycerides (SMD: −0.40, 95% CI: −0.58 to −0.21). However, no significant changes were found in HDL (SMD: 0.02, 95% CI: −0.18 to 0.22), LDL (SMD: −0.01, 95% CI: −0.20 to 0.18), or TC (SMD: −0.34, 95% CI: −0.70 to 0.01; see Figure 5).

The meta-analysis included data from five for FBS (26, 27, 29, 30, 32), four for insulin and HOMA-IR values (26, 27, 29, 32). Significant heterogeneity was observed among studies for FBS (I² = 52%, p = 0.06), insulin (I² = 91%, p < 0.00001), and HOMA-IR (I² = 97%, p < 0.00001). The meta-analysis showed a significant improvement in HOMA-IR with fish oil supplementation, no significant change in insulin levels, and a significant increase in FBS. The combined SMD for FBS was 0.08 (95% CI: −0.13 to 0.30), the combined SMD for insulin was 0.23 (95% CI, −0.52 to 0.97), and the SMD for HOMA-IR was −2.06 (95% CI, −3.36 to −0.49; see Figure 6).

The analysis included data from seven RCTs for BMI (26–32), 2 for hip circumference (Hip-C) (27, 32), four for waist circumference (Waist-C) (27, 28, 30, 32), and three for waist-to-hip ratio (WHR) (27, 29, 31). No statistical heterogeneity was observed among the studies. The meta-analysis revealed a significant improvement in Waist-C (SMD: −0.31, 95% CI: −0.54 to −0.08) in the fish oil group. However, there were no significant improvements in BMI (SMD: 0.16, 95% CI: −0.34 to 0.02), Hip-C (SMD: −0.10, 95% CI: −0.36 to 0.17), or WHR (SMD: −0.07, 95% CI: −0.34 to 0.19; see Figure 7).

The analysis included data from two RCTs for adiponectin (26, 27), three for uric acid (UA) (27, 31, 32), and three for tumor necrosis factor-alpha (TNF-α) (27, 31, 32). Heterogeneity was observed among studies for adiponectin (I² = 91%, p = 0.0009) and TNF-α (I² = 82%, p = 0.0001). The meta-analysis revealed a significant improvement in TNF-α levels in the fish oil group compared to the control group (SMD: −0.76, 95% CI: −1.35 to −0.18). However, there were no significant changes in adiponectin (SMD: 0.70, 95% CI: −0.72 to 2.12) or UA (SMD: −0.14, 95% CI: −0.37 to 0.09; see Figure 8).

Our investigation found that Egger’s test results indicated no publication bias for certain outcomes: AST (p = 0.302), ALT (p = 0.358), GGT (p = 0.638), HDL (p = 0.452), LDL (p = 0.964), TG (p = 0.387), TC (p = 0.616), Hip-C (p = 0.236), Waist-C (p = 0.475), WHR (p = 0.864), insulin (p = 0.431), FBS (p = 0.449), TNF-α (p = 0.303), and UA (p = 0.051). However, publication bias was suggested by the test for BMI (p = 0.034) and HOMA-IR (p = 0.024). Funnel plots also revealed publication bias in ALT, BMI, FBS, insulin, and HOMA-IR. Figures 9–11 demonstrate the visual assessment of funnel plots.

One-way sensitivity analyses were conducted to assess result stability and evaluate heterogeneity’s impact on the study outcomes. ALT, Total Cholesterol, insulin, TNF-α, and HOMA-IR were compared, and each study’s influence on the combined SMD was examined through sequential removal. Sensitivity analyses indicated consistent combined SMDs after excluding individual studies for FBS, ALT, and insulin. Total Cholesterol, TNF-α, and HOMA-IR showed significant fluctuations in the combined SMD, indicating result instability. Exclusion of the Shojasaadat-2019 and Song-2020 studies led to a shift from non-significant to significant findings for Total Cholesterol. Likewise, exclusion of the Qin-2015 and Guo-2022 studies reversed significant findings to non-significant for TNF-α. In terms of HOMA-IR, excluding the Guo-2022 study led to a transition from significant to non-significant outcomes (see Figure 12).

Subgroup analyses were conducted based on the dosage of fish oil and the intervention duration to explore the stability of the results and potential sources of heterogeneity. For ALT, we found that there was a statistically significant difference when the intervention duration was greater than 12 weeks (p < 0.0001), while there was no statistical significance when it was less than 12 weeks (p = 0.1). The heterogeneity among subgroups decreased (I² = 0). However, the dose subgroup analysis did not show statistical significance, and heterogeneity decreased in the subgroup less than 2000 mg (I² = 36%). The subgroup analysis of AST indicated that there was statistical significance when the intervention dose was less than 2000 mg (p = 0.0006) and when the intervention duration was greater than 12 weeks (p = 0.0004); there was no statistical significance when the intervention dose was greater than 2000 mg (p = 0.45) and when the treatment duration was less than 12 weeks (p = 0.56). The subgroup analysis of GGT showed statistical significance when the intervention dose was less than 2000 mg (p = 0.02), but no statistical significance when it was greater than 2000 mg (p = 0.2). The subgroup analysis based on intervention duration did not reveal any statistical significance (see Table 2).