This study was performed in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, ensuring rigorous and transparent reporting of systematic review and meta-analysis methodologies (21). The study utilized publicly available data from databases including PubMed, Embase database, Web of Science and Cochrane databases, and did not involve human participants. Thus, it did not require submission for institutional review board approval or informed consent.

A thorough search was performed covering literature from inception through May 2024. The detailed search strategy incorporated a combination of the MeSH term “Dietary Supplements” OR the keywords “Dietary Supplement*,” “Food Supplement*,” “Oral nutritional supplements,” “Nutraceutical*”; MeSH term “Neoplasms” OR the keywords “Tumor*,” “Neoplasm*,” “Cancer*” and “Neoplasia*”; the keywords “Surgery,” “Operation,” “Resection” OR “Microsurgery”; MeSH term “Body Weight” OR the keywords “Body Weight*.” PubMed and Cochrane utilize the same MeSH for indexing. In Embase, however, searches are based on Emtree. If the terms do not match, we also expand to the closest related terms. This search is language-and study-type agnostic. This meta-analysis exclusively includes randomized controlled trials (RCTs). Two independent researchers manually searched additional references from the reference lists of pertinent studies as well as review articles to identify further references and reviewed all pertinent texts tables, and figures to extract data.

Studies meeting the following inclusion criteria would be selected: (1) patients who have been diagnosed with malignant tumor and have subsequently undergone surgical treatment; (2) interventions including ONSs; (3) Comparisons were made based on usual, standard or regular diet categories; (4) outcomes assessed as mean differences in body weight; and (5) RCTs. Duplicate publications, animal studies, and articles without original data (including reviews, abstracts only, letters, comments, editorials, meetings, and case reports) were excluded. Studies without comparison groups or complete data were also excluded.

Two researchers separately assessed the titles, abstracts, and complete documents of the articles based on pre-defined inclusion criteria. To ensure a thorough and objective evaluation of the data, any discrepancies or disagreements between the reviewers were resolved through a process of discussion and consensus-building. When agreement could not be achieved, a collaborative review involving all researchers was carried out to determine article inclusion. Data extraction was carried out using specialized forms designed for this purpose. To ensure the meta-analysis’s accuracy and completeness, comprehensive checks were performed on the references from each included study to identify and remove any potential overlaps.

Extracted parameters from the trials included the first author, year of publication, study country, study design, type of ONS nutrients, controls conditions, intervention duration, characteristics of the study population, such as patients count, age distribution, female ratio, disease type, daily kilocalorie intake, TNM stage and body weight data. For studies lacking standard deviation reports, pooled standard deviations were estimated for each group.

Mean differences (MD) along with their associated 95% confidence intervals (CIs) were computed to assess the continuous variable of body weight loss (BWL). To obtain the pooled overall MD and 95% CIs for the outcomes, a random-effects model was utilized, analyzed with restricted maximum likelihood (REML) estimation (22). The choice between a fixed-effects model and a random-effects model was based on the level of heterogeneity assessed. Heterogeneity among included studies was valuated using Cochrane’s Q test and the Higgins’ I² values. An I² value of less than 50% was interpreted as low heterogeneity, indicating that a fixed-effects model was appropriate. Conversely, an I² value greater than 50% suggested substantial heterogeneity, warranting the use of a random-effects model. Statistical significance of heterogeneity was determined with Cochran’s Q test, where a p-value below 0.05 was considered indicative of significant heterogeneity (23).

To investigate factors moderating BWL, we conducted meta-regression analyses on continuous variables including patient number, age, percentage of females and intervention duration. For categorical variables such as daily calorie intake, type of disease, and country, we performed meta-analysis of variance. REML estimation was employed to assess variance in true effects and a examine potential moderating factors. p-values of 0.05 or less, or 95% CI that did not include zero (mean difference = 0), were deemed statistically significant. In this meta-analysis, R software (R Foundation for Statistical Computing; version 4.4.1) was utilized for data processing and the results were presented in the form of figures.

Publication bias was evaluated through the use of a funnel plot, with a symmetrical distribution of studies within the plot suggests indicating a lack of bias. To more precisely quantify the extent of publication bias, we employed Egger’s test as an additional analytical method (23–25).

The quality of included RCTs was evaluated using the Cochrane Collaboration’s Risk-of-Bias (RoB) 2.0 tool. This comprehensive assessment tool evaluates potential bias across five key domains. Each domain was systematically rated to determine the risk of bias as high, low, or some concerns. In the RoB 2.0 item, an overall risk rating of “high” is assigned if any one of the five domains is rated as “high,” or if two or more domains are rated as “some concerns” (26).

The initial search identified 361 publications across PubMed, Web of Science, Embase databases and the Cochrane Library. Among them, 78 articles were excluded due to duplication. After reviewing titles as well as abstracts, 259 articles were also eliminated for various reasons. The remaining 24 full-text studies were subjected to a comprehensive and detailed review. 12 articles were further removed, including 9 articles due to lack of quantitative outcomes and 3 articles due to inappropriate timing of intervention. Ultimately, twelve studies fulfilled the selection criteria and were incorporated into this meta-analysis (27–38) (Figure 1). These 12 cohort studies encompassed a total of 2,268 cancer patients diagnosed with stomach, colorectal, head and neck, or bladder cancer. The studies were conducted in Asia and the Americas, specifically with five from Japan, four from China, one from Thailand, one from Brazil, and one from the United States. The key characteristics of these included RCTs were summarized in Table 1.

This analysis demonstrated a statistically significant reduction in BWL in the oral nutritional supplement (ONS) group compared to the control group with a mean difference (MD) of 1.11 (95% CI: 0.52–1.70). Due to significant heterogeneity (I² = 97.0%, p < 0.01), a random-effects model was applied for the pooled analysis of BWL. To further assess the impact of daily ONS consumption on BWL, a subgroup analysis was carried out. In the subgroup consuming less than 500 kilocalories per day, the pooled MD was 1.37 (95% CI: 0.60–2.14), indicating statistical significance (p < 0.01), though the Higgins’ I² value was 98.0%, reflecting considerable heterogeneity. Conversely, in the subgroup consuming over 500 kilocalories per day, the pooled MD between the ONS group and control groups was 0.50 (95% CI: 0.39–0.61), also statistically significant (p = 0.01) with a Higgins’ I² value of 0.0%, indicating low heterogeneity (Figure 2).

Meta-regression and meta-analysis of variance were conducted to investigate potential moderators influencing BWL, such as daily nutritional intake as well as the study country was conducted (Table 2). The analysis revealed significant heterogeneity in the two variables, daily consumption and study country. Patients in the ≤500 kcal group experienced greater BWL compared to those in the >500 kcal group (p = 0.0287). Additionally, there were statistically significant differences in BWL between countries (p < 0.0001), with China (MD = 1.70, 95% CI = 0.52–2.88) and the United States (MD = 1.90, 95% CI = −7.61–11.41) showing higher mean differences than Japan (MD = 0.86, 95% CI = 0.78–0.94), Thailand (MD = 0.32, 95% CI = −2.04–2.68), and Brazil (MD = 0.50, 95% CI = 0.39–0.61). Furthermore, differences were observed between disease types, with patients in the gastrointestinal tumor group showing better improvement in BWL (MD = 1.14, 95% CI = 0.52–1.75) compared to those in the non-gastrointestinal cancer groups (MD = 0.41, 95% CI = −1.88–2.70), although this result did not achieve statistically significant (p = 0.5494), probably due to insufficient sample size. Longer intervention duration was associated with reduced BWL (β = 0.46), though this finding also lacked statistical significance (p = 0.1512).

The assessment of publication bias associated with this study was presented using a funnel plot (Supplementary Figure S1), which visually indicated that the mean differences were symmetrically distributed. Egger’s test further supported the absence of publication bias (p = 0.4562).

The stability of the mean difference for body weight loss was tested using the trim and fill method under a random-effects model. After imputing two hypothetical studies, the adjusted pooled MD (MD = 0.70, 95% CI = 0.0034–1.39, p = 0.0489) was not significantly different from the original pooled MD, indicating that the initial conclusions remained robust (Supplementary Figure S2). Additionally, the pooled MD showed no substantial change regardless of which study was excluded in the sensitivity analysis (Supplementary Figure S3).

Twelve studies were assessed for risk of bias using the five domains of the RoB 2.0 tool (Figure 3). The assessment results showed that the overall risk of bias of 11 studies was rated as “high,” except for 1 study which was rated as “some concerns.” This outcome is largely attributed to the RoB 2.0 criteria, where a study is classified as high risk if any domain is deemed high risk. Specifically, in the domain of bias related to deviations from intended interventions (D2), eleven studies were rated as “high risk,” while 1 study was rated as “some concerns,” primarily because most studies were unable to blind the oral nutritional supplement (ONS) intervention due to its nature. Additionally, in the area of bias related to selective reporting (D5), 2 studies assessed as having “high risk” and while another two were evaluated as “some concerns.”