A comprehensive literature search was performed in PubMed, Embase, the Cochrane Library, and Web of Science. The search strategy combined controlled vocabulary terms with free-text keywords and covered the period from database inception to November 1, 2025, with no language restrictions applied. Search terms included “sarcopenia,” “sarcopenias,” “obesity,” “aged,” “older adults,” “sports,” “athletics,” “resistance training,” “physical exercise,” “proteins,” “protein gene products,” and “randomized controlled trial” (Supplementary Table S2).

Inclusion Criteria:

Exclusion criteria:

Two investigators independently extracted data from eligible randomized controlled trials in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, with all entries cross-checked for accuracy. Any discrepancies were resolved through discussion with a third author until consensus was reached. From each study, the following information was extracted: first author, year of publication, mean participant age, geographic location, follow-up duration, intervention protocols for the experimental and control groups, and the corresponding outcome measures. For continuous outcomes (BMI, GRIP, FM, PBF, and SMI), priority was given to extracting the mean change from baseline and its standard deviation (SD). When only baseline and post-intervention values were reported, change scores and their SDs were calculated using standardized formulas. If the within-participant correlation coefficient (r) was not reported and could not be derived, an assumed value of r = 0.5 was applied. The robustness of this assumption was subsequently evaluated through sensitivity analyses.

The methodological quality of the included randomized controlled trials was assessed using the Cochrane Risk of Bias 2.0 (ROB 2) tool. This instrument evaluates potential sources of bias across five domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each domain, as well as the overall risk of bias, was classified in accordance with Cochrane guidance as low risk, some concerns, or high risk.

The analyses of all networks meta-data were executed using Stata MP version 17.0. For outcomes that were measured using the same scale and units across various studies, we computed the mean differences (MDs) along with their respective 95% confidence intervals (CIs). In cases where the outcomes were measured on different scales or with varying units, we utilized standardized mean differences (SMDs) with their 95% CIs. The core analyses were conducted assuming consistency and employed random-effects models. The between-study variance (τ²) was estimated using the restricted maximum likelihood (REML) technique. In networks with closed loops, we evaluated global inconsistency through the design-by-treatment interaction model. On the other hand, we assessed local inconsistency using the node-splitting method. When the p value fell below 0.10, this raised a red flag about potential inconsistency. To dig deeper into loop-specific consistency, we turned to the inconsistency factor (IF); if the 95% CI for the IF included zero, we took that as a green light indicating no statistically significant disagreement between direct and indirect evidence. To bring the network geometry to life, we created network plots where the size of each node corresponded to the total sample size for a given intervention, while the thickness of the edges showed how many studies directly compared different interventions. To really separate the wheat from the chaff among competing interventions, we applied multiple ranking metrics—SUCRA, PreBest, and mean rank—which gave our results both more robustness and clearer interpretation. When we had more than 10 studies in the mix, we used comparison-adjusted funnel plots to sniff out any publication bias or small-study effects. Our sensitivity analyses employed a leave-one-out approach, where we systematically removed each study one at a time and re-estimated the random-effects consistency model, keeping a close eye on whether the direction and size of our pooled effect estimates changed—a telltale sign of result stability. Finally, we conducted univariable network meta-regression analyses to see how study-level covariates might be influencing treatment effects. The results are presented as regression coefficients complete with 95% CIs and Wald test p values, with anything below 0.05 hitting the statistical significance mark and suggesting a real effect-modifying association.

The certainty of evidence for outcomes derived from the network meta-analysis was evaluated using the GRADE framework, operationalized through the Confidence in Network Meta-Analysis (CINeMA) approach. As all included studies were randomized controlled trials, evidence certainty was initially rated as high and subsequently assessed across six domains: within-study bias, indirectness, imprecision, heterogeneity, inconsistency, and across-study bias (including publication bias and small-study effects). Within-study bias was assessed on a domain-specific basis using the RoB 2.0 tool and summarized using the CINeMA contribution matrix, which weights each study’s risk of bias according to its contribution to the corresponding network effect estimate. Indirectness was evaluated under the assumption of transitivity, with a priori consideration of potential effect modifiers, including baseline disease severity, intervention intensity, and follow-up duration. Imprecision was judged with reference to the minimal important difference (MID). For continuous outcomes, a standardized mean difference (SMD) threshold of 0.5 was applied, and judgments were based on the position of the 95% confidence interval relative to the null line or MID threshold. Heterogeneity was evaluated using the between-study variance (τ²) from random-effects models, together with the location of the prediction interval relative to the MID. When closed loops were present within the network, inconsistency between direct and indirect evidence was examined using node-splitting (SIDE) and/or design-by-treatment interaction models. To evaluate potential across-study bias, we cast a wide net by pulling together data from trial registration records, digging through grey literature, and taking a close look at comparison-adjusted funnel plots to sniff out any small-study effects. For each domain, we gave it a clean bill of health, raised some red flags, or sounded the alarm—translating to zero, one, or notches down the credibility ladder, respectively, following the GRADE playbook. In the final analysis, we slapped a high, moderate, low, or very low label on the overall certainty of evidence for each outcome.

The initial literature search yielded 418 records. After removal of duplicates and screening of titles and abstracts, 331 articles proceeded to full-text assessment. Of these, 24 randomized controlled trials met the eligibility criteria and were included in the network meta-analysis (Figure 1). The included studies comprised a total of 1,298 participants and evaluated nine exercise- and nutrition-related intervention strategies: multicomponent training combined with nutritional supplementation (MC-NS), nutritional supplementation alone (NS), high-protein supplementation (HP), resistance training (RT), aerobic training (AT), combined resistance and aerobic training (RT-AT), multicomponent training (MC), resistance training combined with high-protein supplementation (RT-HP), and usual care (UC). Across the included trials, mean participant age ranged from 55 to 79.9 years. Studies were conducted in 12 countries, with a relatively high proportion originating from China. Follow-up durations ranged from 3 to 9 months. Detailed characteristics of the included studies are presented in Table 1. Overall, baseline BMI levels were broadly comparable across the included studies. (Supplementary Tables S43–S44).

Study quality was assessed using the Cochrane Risk of Bias 2.0 (ROB 2) tool. Of the 24 randomized controlled trials included, 13 were judged to be at low overall risk of bias, while 11 were rated as having some concerns. Overall, the methodological quality of the included studies was considered acceptable. Most trials reported clearly defined randomization procedures, appropriate intervention implementation, relatively complete outcome data, and objective outcome measures. With respect to the randomization process, the majority of studies described adequate allocation concealment and employed random number tables or computer-generated randomization methods. However, a small number of trials did not clearly report sequence generation or allocation concealment procedures, introducing potential selection bias. Regarding deviations from intended interventions, most studies explicitly adopted an intention-to-treat (ITT) approach, thereby reducing bias related to non-adherence or protocol deviations. Nonetheless, several trials did not implement participant blinding and relied on single-blind designs, leaving open the possibility of performance bias related to behavioral modification. When it comes to the issue of data gaps, the completeness of outcomes was mostly on point, and ITT-based strategies were often employed to tackle any missing info, which kept the threat of attrition bias to a minimum. Yet, a few studies fell short in giving the nitty-gritty details on how they dealt with data holes. As for measuring outcomes, the majority of the trials relied on objective assessment gadgets and ensured blinding at the very least during the outcome evaluation phase, which kept the chance of detection bias low. When you look at the big picture, the risk of bias across the board was quite low, and the evidence’s quality was considered up to snuff, giving us a lot of faith in the compiled findings. For a deeper dive into the risk-of-bias assessments, check out Figure 2.

To assess the impact of individual studies on the pooled effect estimates, we performed leave-one-out sensitivity analyses across all outcomes. This approach involved systematically removing one study at a time and re-running the network meta-analysis on the remaining data under a random-effects model while maintaining the consistency assumption throughout each iteration. Across all analyses, the direction of pooled effects for exercise and nutritional interventions relative to UC remained unchanged. Variations in effect size magnitude were modest, confidence intervals overlapped substantially, and inferences regarding statistical significance were not materially altered, indicating robustness of the findings (Supplementary Tables S14–S18). To explore potential sources of heterogeneity, univariable network meta-regression analyses were performed for the five outcomes, with geographic region, follow-up duration, and mean participant age included as study-level covariates. None of these covariates showed a statistically significant association with the relative treatment effects of any intervention compared with UC (Supplementary Tables S19–S33).

Moreover, we crafted funnel plots that accounted for comparisons to evaluate BMI, grip strength, forearm muscle, percentage body fat, and skeletal muscle index, as well as overall effectiveness. A quick glance revealed a roughly balanced spread with no striking anomalies, indicating a slim chance of significant publication bias (refer to Supplementary Figures S6–S10).

The certainty of evidence for the primary outcomes was evaluated using the CINeMA framework, revealing variability across outcomes. Detailed evaluations are provided in Supplementary Tables S34–S38. For BMI, among the 28 pairwise comparisons, 4 (14%) were rated as high certainty, 10 (36%) as moderate certainty, 13 (46%) as low certainty, and 1 (4%) as very low certainty. For GRIP, also across 28 comparisons, 19 (68%) were judged moderate certainty, 8 (28%) low certainty, and 1 (4%) very low certainty. For FM (36 comparisons), 2 (6%) were high certainty, 16 (44%) moderate certainty, 16 (44%) low certainty, and 2 (6%) very low certainty. The overall certainty for PBF (36 comparisons) was comparatively lower, with 5 (14%) rated moderate certainty, 21 (58%) low certainty, and 10 (28%) very low certainty. For SMI (10 comparisons), 2 (20%) were moderate certainty, while the remaining 8 (80%) were low certainty.

With respect to BMI, MC was the only intervention that achieved a statistically significant reduction compared with UC (MD = −1.08, 95% CI −1.86 to −0.29). Although RT-HP and RT-AT demonstrated downward trends, these effects did not reach statistical significance. This pattern suggests that, in older adults with sarcopenic obesity, single-modality exercise interventions may be insufficient to elicit consistent improvements in weight-related indices, whereas multicomponent programs—by concurrently integrating aerobic, resistance, and functional training elements—may be more likely to produce measurable reductions in BMI (22, 23), Importantly, BMI changes alone provide an incomplete representation of body composition. In sarcopenic obesity, clinically meaningful improvement is defined by fat loss accompanied by preservation or accretion of muscle mass, underscoring the necessity of interpreting BMI alongside FM, PBF, and SMI.

In terms of muscle strength, both RT and MC significantly improved handgrip strength, with RT demonstrating the larger effect size (RT: MD = 3.96, 95% CI 2.15–5.77; MC: MD = 2.13, 95% CI 0.25–4.01) and ranking highest in SUCRA analyses. These findings are biologically plausible, as resistance training enhances muscle strength through externally imposed loads, progressive overload, and varied contraction velocities, thereby promoting neuromuscular adaptation and increasing force-generating capacity (24, 25). Given the well-established associations between handgrip strength and fall risk, functional decline, and mortality in older adults, resistance training should be regarded as a foundational component of exercise prescriptions for individuals with sarcopenic obesity.

For adiposity-related outcomes, the FM analysis indicated that only RT, relative to UC, produced a statistically significant reduction in fat mass (MD = −2.30, 95% CI −3.63 to −0.98) and ranked prominently among competing interventions. In contrast, for PBF, both MC and RT were associated with significant reductions (MC: MD = −3.53, 95% CI −5.70 to −1.36; RT: MD = −2.30, 95% CI −3.98 to −0.62). These results suggest that both intervention types may meaningfully contribute to reductions in relative adiposity. From a mechanistic standpoint, RT may decrease fat mass indirectly by increasing lean tissue and improving metabolic efficiency, whereas MC may exert broader effects on body fat percentage through its more comprehensive and functionally oriented training structure (22, 23, 26). Nevertheless, the detection of a potential inconsistency within the MC–RT–UC closed loop in the PBF network necessitates caution when interpreting rankings and effect estimates for this outcome. Accordingly, the certainty of evidence for PBF was downgraded in the network GRADE assessment. To explore potential sources of this inconsistency, a clinical heterogeneity assessment was undertaken focusing on exercise intensity and baseline participant characteristics across studies contributing to the MC–RT–UC loop. Although interventions were broadly categorized as resistance training or multicomponent training, substantial variability existed in prescribed exercise intensity. Some RT and MC programs employed moderate-intensity protocols according to ACSM criteria (e.g., RPE 12–13 or 40–60% 1RM), whereas others adopted moderate-to-vigorous or high-intensity regimens (e.g., ≥70–80% 1RM or higher perceived exertion), potentially leading to differential effects on body fat percentage. In addition, baseline BMI distribution varied markedly across studies, ranging from overweight or mildly obese populations (mean BMI ≈ 25–27 kg/m²) to individuals with established obesity (mean BMI ≥ 33–35 kg/m²). Given that baseline adiposity level may modify the responsiveness of PBF to exercise interventions, such heterogeneity in participant characteristics may violate the transitivity assumption and partly explain the observed inconsistency within the MC–RT–UC closed loop for PBF outcomes.

Evidence regarding muscle mass outcomes was comparatively limited. Only five trials reported SMI, and although MC-NS showed a non-significant trend toward increasing SMI relative to UC (MD = 0.41, 95% CI −1.10 to 1.93) and ranked favorably, the available evidence does not support definitive conclusions (27). Improvements in skeletal muscle mass typically require longer intervention durations, adequate and sustained protein intake, and progressively increasing training loads. The short follow-up periods, modest sample sizes, sparse network connections, and imprecision of effect estimates in the included trials likely contributed to the absence of statistically significant findings for SMI. In the present network meta-analysis, evidence regarding changes in SMI was limited, and none of the evaluated interventions produced statistically significant improvements in this outcome. Although MC-NS ranked favorably and showed a non-significant upward trend in SMI relative to usual care, the magnitude and robustness of this effect remain uncertain. These findings are likely attributable to a combination of physiological and methodological factors. A principal biological explanation is age-related anabolic resistance, a well-recognized phenomenon whereby older adults exhibit a blunted skeletal muscle protein synthesis response to both resistance exercise and protein intake (28). Accordingly, in older adults with sarcopenic obesity, the protein doses used in many trials may have been insufficient to overcome this physiological barrier (29). Moreover, prior meta-analyses have indicated that longer intervention durations and sufficiently high training loads are critical for inducing muscle hypertrophy, whereas the relatively short intervention periods and heterogeneity in training protocols across most included trials may have limited the detectability of changes in SMI (30).

The results of this study clearly demonstrate that older adults battling sarcopenic obesity can significantly boost their handgrip strength and trim down their body fat percentage by engaging in exercise programs, with resistance training proving to be particularly effective in this regard. These results are broadly consistent with those reported by Chen et al., who likewise observed favorable effects of exercise on muscle strength and adiposity-related outcomes (31). However, their meta-analysis included only 12 trials and focused exclusively on resistance training, without considering alternative exercise modalities. In contrast, the current network meta-analysis incorporated a wider spectrum of exercise-based strategies and explicitly evaluated the potential modifying roles of nutritional supplementation and high-protein intake. This broader analytical framework enabled a more comprehensive comparison of rehabilitation approaches for sarcopenic obesity. Similarly, Contillo et al. reported that exercise interventions were associated with reductions in body fat and improvements in handgrip strength (32). Nevertheless, their analysis did not distinguish the relative effectiveness of specific exercise modalities. By applying a network approach with SUCRA-based ranking, the present study extends prior work by demonstrating that resistance training is most consistently associated with superior outcomes in both adiposity reduction and muscle strength enhancement in this population. In the meta-analysis conducted by Contillo et al., interventions were primarily categorized according to exercise modality (AT, RT, RT-AT) and the presence or absence of concomitant protein or amino acid supplementation. Building on this framework, the present network meta-analysis extends the scope of comparative evaluation by incorporating not only direct contrasts between different exercise modalities, but also a systematic comparison of multicomponent training combined with nutritional supplementation against alternative intervention strategies. This expanded approach allows for a more comprehensive assessment of the relative effectiveness of integrated intervention models. With respect to outcome measures, the two analyses demonstrate substantial concordance, with both focusing predominantly on muscle strength and body composition outcomes, including handgrip strength, skeletal muscle mass, fat mass, and percentage body fat. These measures are widely regarded as core clinical endpoints for evaluating intervention efficacy in sarcopenic obesity. From a methodological perspective, the present network meta-analysis further strengthens the interpretability of its findings by applying the GRADE framework to systematically rate the certainty of evidence across key outcomes. This approach enhances transparency, improves clinical credibility, and provides a structured hierarchy of evidence to better inform clinical decision-making.

When the totality of outcomes is considered, the present findings suggest that resistance training yields more consistent benefits for improving GRIP and reducing FM, whereas multicomponent training appears more advantageous for improving BMI and PBF. Given the consistent advantages of MC in reducing BMI and PBF, it is clinically appropriate to translate these findings into practical recommendations. For older adults with sarcopenic obesity, MC may reasonably integrate aerobic and resistance training within a single intervention framework, with each component accounting for approximately 40–60% of the total training volume to maximize their complementary effects. Aerobic exercise may be prescribed at moderate intensity (approximately 60–75% of maximal heart rate) for 20–30 min per session to enhance energy expenditure and promote fat loss, whereas resistance training may be performed 2–3 times per week using progressive loads corresponding to roughly 60–80% of one-repetition maximum, targeting major muscle groups. In clinical practice, MC programs should be individualized according to baseline functional status and comorbid conditions, and progressively advanced to ensure both safety and adequate training stimulus. MC may be particularly appropriate for individuals in whom adiposity reduction is a primary therapeutic objective, whereas resistance training alone should remain the cornerstone intervention for patients with pronounced muscle weakness or functional impairment. From a clinical perspective, resistance training may therefore be regarded as a foundational intervention, particularly for older adults with sarcopenic obesity who exhibit pronounced muscle weakness or functional impairment. Aerobic and functional components may be incorporated as appropriate to augment energy expenditure and enhance performance in activities of daily living.

Nutritional supplementation and high-protein intake are biologically plausible adjuncts for preserving or augmenting muscle mass; however, substantial heterogeneity exists across studies with respect to supplement type, dosage, adherence, and dietary context. As a result, the current evidence base supports cautious consideration rather than strong clinical endorsement of these strategies. Exercise prescriptions should also be tailored to individual comorbidity profiles and risk factors—such as osteoporosis, knee or hip osteoarthritis, and cardiopulmonary limitations—and implemented with graded intensity, appropriate exercise selection, and safety monitoring, preferably under professional supervision.

The strengths of this study include its focus on sarcopenic obesity as a distinct and clinically burdensome phenotype, the inclusion of a relatively large number of trials spanning multiple countries, and the use of network meta-analysis to enable systematic comparison and ranking of multiple interventions. We further conducted comprehensive assessments of consistency and inconsistency, along with leave-one-out sensitivity analyses and network meta-regression, to confirm the robustness of the findings. Evaluation of evidence certainty using the CINeMA framework revealed substantial variability across outcomes, with particularly low certainty for PBF and SMI, underscoring the need for cautious interpretation of these results.

Several limitations merit consideration. First, strict double blinding is inherently difficult in exercise-based interventions, potentially introducing performance and detection bias; however, most primary outcomes were objectively measured, which likely mitigated this concern. Second, some comparisons were limited by small sample sizes and sparse network connectivity, particularly for SMI, which was reported in only five trials, resulting in imprecise effect estimates. Third, heterogeneity in exercise intensity, training frequency, and nutritional supplementation protocols may have influenced the transitivity assumption and the reliability of indirect comparisons. Fourth, the detection of potential loop inconsistency within the PBF network highlights the need for additional head-to-head trials to clarify these relationships. Finally, most studies had follow-up durations of 3–9 months, restricting assessment of long-term sustainability, adverse events, and adherence; moreover, safety and compliance were not consistently reported.