A comprehensive literature search in PubMed, Web of Science, Embase and CNKI databases was conducted to retrieve relevant articles published from database inception up to 31 March 2025 (data last searched), with the following keywords or free-text terms: (“ultra-processed foods” OR “UPFs” OR “ultra-processed food” OR “UPF” OR “NOVA food classification”) AND (“non-alcoholic fatty liver disease” OR “NAFLD” OR “fatty liver disease” OR “metabolic dysfunction-associated fatty liver diseases” OR “MASLD”). Searches were limited to human studies without any restrictions regarding publication date and language. Furthermore, reference lists of eligible articles and prior systematic reviews of relevant topics were also reviewed to lessen the chance of missing potentially relevant articles. The literature search was conducted by two independent authors (X.-P. C and L. S.). The details of search strategy are described in the Supplementary Table S1.

Two authors (X.-P. C and J.-H. Z.) independently examined the titles and abstracts of all articles retrieved in the initial literature search to identify studies that reported the correlation between UPF consumption and risk of NAFLD. To be included in this updated systematic review and meta-analysis, the following criteria had to be met: (1) observational studies, including cohort, case–control, nest case–control, case-cohort or cross-sectional studies, performed in adults (≥18 years); (2) considered UPF consumption as the exposure variable; (3) UPF was defined based upon the NOVA food classification; (4) considered NAFLD as the outcome variable; (5) studies reporting the association between UPF consumption and risk of NAFLD, providing the estimates of relative risks (RRs), hazard ratios (HRs), odds ratios (ORs) along with their corresponding 95% confidence intervals (CIs) (or sufficient data to calculate them); (6) If the original data in the retrieved studies lacked sufficient detail, the corresponding author will be contacted by email. Additionally, studies were excluded if they met one of the following criteria: (1) animal, cell culture, and in vitro studies; (2) non-observational studies, including congress abstracts, reviews, editorials, case reports, book chapters, and letters; (3) did not use the NOVA food classification system to define UPF (assessed the only specific food or food groups, such as sugar-sweetened drinks, processed meat); (4) Secondary steatosis, e.g., fatty liver disease, non-alcoholic steatohepatitis, viral hepatitis; (5) did not provide the HRs, RRs or ORs with corresponding 95%CIs; (6) Irrelevant articles; (7) studies performed in the pediatric participants (aged < 18 years). The PECOS criteria for this meta-analysis is described in Table 1.

For each eligible article, we used a predefined form to extract the required information on first author’s name, year of publication, study design, study region, sample size, number of total participants and NAFLD cases, mean age/age range, follow-up duration for cohort studies, dietary assessment method, and confounding variables that were adjusted for in the statistical analysis. In this study, two authors (J.-H. Z. and L. S.) independently utilized the Newcastle-Ottawa Scale (NOS) to judge the methodological quality of all included studies. The NOS consisted of three domains: selection of participants, comparability of participants, and ascertainment of outcome/exposure of interest, which was designed for non-randomized studies in meta-analyses (33). The total NOS score ranges from 0 to 9 points, with higher scores indicating higher study quality. Studies with total NOS scores ≥7 points were judged as high quality (34). Any discrepancies arising during data extraction and quality assessment were resolved via joint discussion and/or consultation with the corresponding author (X.-P. C).

According to the NOVA classification, all food and beverage items were divided into four different groups, including unprocessed or minimally processed foods, processed culinary ingredients, processed foods and UPFs (11). Existing literature indicates that UPFs are typically ready-to-eat, hyper-palatable, cheap and characterized by high energy density, high intake of added sugars, salt, saturated and trans-fats, as well as low consumption of dietary fiber, vitamins and minerals (12). Typical examples of UPF include biscuits and cakes, crisps, sugar-sweetened beverages, pizza, processed meats, desserts, etc.

The RRs and their 95%CIs were considered as the effect estimate in this updated systematic review and meta-analysis. Meanwhile, we assumed that the HRs were approximately equal to RRs (35). For cross-sectional or case–control studies, ORs were converted to RRs by the following formula: RR = OR/[(1-P₀) + (P₀*OR)], where P₀ shows the incidence of NAFLD in the non-exposed group (36). The pooled RRs and 95%CIs of NAFLD were calculated by comparing the highest and the lowest categories of UPF consumption and for each 10% increment in UPF consumption. The Cochran’s Q test and I-squared (I²) statistic were used to evaluate between-studies heterogeneity with estimates associated with p-values of Cochran’s Q test <0.05 and I² ≥ 50% considered to be substantial heterogeneity (37). If there was substantial heterogeneity between studies, RRs were calculated using a random-effects meta-analysis with the DerSimonian and Laird method. Otherwise, fixed-effects model was used to pool the RRs (38). Given the expected heterogeneity of the included studies, leave-one-out sensitivity and subgroup analyses were used to detect the potential sources of heterogeneity. Subgroup analyses were performed based on study design (cohort or cross-sectional studies), dietary assessment method (FFQ or 24-h dietary recall), study region (Western countries or other countries), sample size (≥5,000 or <5,000), study quality (≥7 or <7), and mean age (≥50 or <50). Sensitivity analysis was conducted to confirm whether the pooled RRs were robust or sensitive to the impact of a certain study. Univariable meta-regression analyses were also performed to test the effect of specific variables (i.e., sample size, study area, and dietary assessment methods) on the effect size for the association between UPF consumption and NAFLD. Publication bias was assessed via the visual inspection of funnel plots and quantified by Begg’s test and Egger’s regression asymmetry test (39). If the results indicated evidence of publication bias, the trim-and-fill method was used to re-calculate the results (40). Finally, we also performed a dose–response meta-analysis to estimate the trend from the correlated log RRs across the categories of UPF consumption. A two-stage GLST model based on generalized least squares was used to examine potential linear or non-linear dose–response association between UPF consumption and risk of NAFLD. We used UPF consumption modeling and restricted cubic splines with three knots at fixed percentiles (10, 50, and 90%) of the distribution. All statistical analyses were carried out using STATA, version 17.0 (StataCorp, College Station, TX, United States), and a two-sided p-value of ≤0.05 was regarded as statistically significant.

Figure 1 summarized the results of the literature search and study selection. During the initial literature screening, we identified 64,270 potentially relevant articles through four databases and other sources. After removing 625 duplicates, 63,645 records were preliminarily screened. Subsequently, 63,578 articles were excluded based on the examining of titles and abstracts of retrieved articles and irrelevant articles. The remaining 67 full-text articles were reviewed in details by two independent authors (X. P. C and L. S.) and 57 articles were excluded for the following reasons: systematic review and/or meta-analysis (n = 37), did not use the NOVA food classification (n = 5), report the association between UPF consumption and risk of NAFLD in adolescents (n = 2), the outcome of interest was features of NAFLD (n = 4), the main exposure was specific foods, such as sugar-sweetened beverages, desserts, and processed meats (n = 7), reported data as β coefficient (n = 1), and reported the same participants (n = 1). Ultimately, a total of 10 articles fulfilled the inclusion criteria and were included in the final analysis.

The main characteristics of these studies included in the meta-analysis are presented in Table 2. A total of 10 articles (involving 513,440 participants and 20,637 NAFLD cases), published between 2021 and 2025, were included in the final analyses. All the eligible studies had an observational design. Six of all included studies were cohort studies (18, 20, 22, 24, 25, 27), and the remaining four were cross-sectional studies (19, 21, 23, 26). Two of included studies were from United Kingdom (18, 20), two from United States (19, 21), two from Israel (23, 26), one from China (22), one from Brazil (25), one from Korea (27), and one from Spain (24). The follow-up duration of cohort studies ranged from 4.2 to 10.5 years. Sample size of included studies ranged from 286 to 173,889. The age of participants across studies ranged from ages 20 to above. Five studies used fatty liver index for the diagnosis of NAFLD (19, 24–27), two studies used ultrasonography (22, 23), two used hospital records (18, 20) and one study used the controlled attenuation parameter (21). UPF consumption was self-reported through food frequency questionnaires (FFQs) and 24-h dietary recalls across all published studies. Specifically, six articles used an FFQ (22–27), and the remaining four studies used 24-h dietary recalls (18–21). All the included studies classified UPF consumption basing on the NOVA food classification systems (18–27). As reported on Table 3, eight out of all included studies received NOS scores ≥7 points, and were classified as of high-quality (18–22, 24, 25, 27). Additionally, the remaining two studies were classified as of medium-quality (23, 26).

Ten articles involving a total of 513,440 participants and 20,637 NAFLD cases, were included in this meta-analysis. Highest UPF consumption was associated with a 22% increased risk of NAFLD compared to the lowest consumption (RR = 1.22, 95%CI: 1.14–1.31, p < 0.001) (Figure 2), with significant heterogeneity (I² = 78.5%; P_{heterogeneity} < 0.001). A 10% increment in UPF consumption was associated with a 6% higher risk of NAFLD (RR = 1.06; 95%CI: 1.04–1.09, I² = 75.9%; p < 0.001) (Figure 3).

Six studies (4 cohort and 2 case–control studies) were included in the dose–response analysis for the association between UPF consumption and risk of NAFLD (Figure 4). The results showed a linear trend association between UPF consumption and risk of NAFLD (RR = 1.02; 95%CI: 0.98–1.07, P_{dose–response} = 0.295, P_nonlinearity = 0.541).

Given the substantial heterogeneity in this meta-analysis (I² = 78.5%, p < 0.001), subgroup analyses were undertaken to further explore the potential sources of heterogeneity across studies (Table 4). In our analyses, subgroup analyses were stratified based on study design (cohort or cross-sectional studies), dietary assessment method (FFQ or 24-h dietary recalls), study region (Western countries or other countries), sample size (≥5,000 or <5,000), study quality (≥7 or <7), and mean age (≥50 or <50). The results of subgroup analyses suggested that high UPF consumption was significantly associated with an increased risk of NAFLD in the subgroups of 24-h dietary recalls (RR = 1.35, 95%CI: 1.24–1.46, p < 0.001), and no heterogeneity was found (p = 0.559; I² = 0.0%). When analyzed separately by study design, the results showed a positive association between UPF consumption and NAFLD risk in cross-sectional studies (RR = 1.29; 95%CI: 1.11–1.31, p = 0.001). There was less evidence of heterogeneity between studies (p = 0.188; I² = 37.3%). For sample size, we observed a positive association between UPF consumption and NAFLD risk in studies with sample size <5,000 (RR = 1.20; 95%CI: 1.02–1.42, p = 0.030), and there was less heterogeneity (p = 0.315; I² = 13.5%). Meanwhile, a positive association between UPF consumption and risk of NALFD was also observed in the subgroups of age < 50 (RR = 1.29; 95%CI: 1.13–1.48, p < 0.001), and heterogeneity decreased to 55.0%. As shown in Supplementary Table S2, the results of univariable meta-regression analyses to explore the effects of potential moderator variables revealed no significant effects of study region and sample size on the association between UPF consumption and risk of NAFLD (p > 0.05).

As shown in Supplementary Figure S1, an examination of the funnel plot revealed little evidence of asymmetry. Similarly, Begg’s and Egger’s tests did not show the presence of significant publication bias (highest versus lowest categories of UPF consumption: Begg’s test: p = 0.640; Egger’s test: p = 0.108). Additionally, the association between UPF consumption and NAFLD risk might be affected by single study based on the results of sensitivity analysis. As shown in Supplementary Figure S2, the study by Konieczna et al. (24) exceeded the range and might be a potential source of heterogeneity. After excluding the studies of Konieczna et al. (24) and Canhada et al. (25) in the repeat analyses (Supplementary Figure S3), sensitivity analysis revealed a slight increase in the pooled RRs on the association between UPF consumption and risk of NAFLD (RR = 1.29; 95%CI: 1.21–1.39, p < 0.0001). Moreover, the heterogeneity also decreased from 78.5 to 27.5%.

This systematic review and meta-analysis has several important strengths. First, as discussed previously, this systematic review and updated meta-analysis is the largest study to date that comprehensively assesses the association between UPF consumption and NAFLD risk. Compared with the aforementioned meta-analysis (28), we not only provided timely updates, but also included more participants and NAFLD cases, thereby enhancing the statistical power to determine a more reliable estimate of the association between UPF consumption and risk of NAFLD. Furthermore, our findings add to the existing evidence and support the emphasis on the importance of reducing UPF consumption in the prevention of NAFLD. Second, visual inspection of the funnel plots and statistical tests (e.g., Begg’s and Egger’s tests) did not indicate any significant publication bias. Third, we strictly screened the articles according to the predetermined inclusion and exclusion criteria. Fourth, all included studies adopted the NOVA classification system to define UPF, avoiding misclassification errors. Finally, despite substantial heterogeneity, we undertook subgroup and sensitivity analyses to explore potential sources of heterogeneity. In addition, dose–response analysis was conducted to further strengthen the association between UPF consumption and risk of NAFLD. Notwithstanding the aforementioned strengths, our study also has some limitations that should be mentioned. First, the observational design of the included studies does not allow for the establishing a causal association between high UPF consumption and the risk of NAFLD. Therefore, further prospective cohort studies or intervention trials are needed to confirm the association between UPF consumption and NAFLD risk. Second, most of the included studies used FFQs to gather dietary information, which may led to an under-or over-estimation of UPF consumption. Additionally, FFQs and 24-h dietary recalls were not explicitly designed to capture the degree of processing, and not validated to precisely measure UPF consumption based on the NOVA food classification. Third, even with some potential confounding variables adjusted, the possibility of residual confounding by certain unmeasured factors could not be excluded. Additionally, our analysis did not search gray literature extensively, which may introduce potential selective reporting bias. Fourth, significant heterogeneity (I² = 78.5%) was observed in this meta-analysis. Despite conducting subgroup and sensitivity analyses to explore potential sources of heterogeneity, we were unable to fully account for the sources of inter-study heterogeneity. Finally, eight of the included studies originated from developed countries and two studies from developing countries, which might limit the generalizability of our findings.