Optimal resistance training prescriptions to improve muscle strength, physical function, and muscle mass in older adults diagnosed with sarcopenia: a systematic review and meta-analysis

Abstract

Background and objectives

Resistance training is widely recommended for managing sarcopenia, but evidence on optimal prescriptions remains limited. This study aimed to assess the effects of different resistance training prescriptions on strength, function, and muscle mass in older adults with sarcopenia.

Methods

We searched PubMed, Embase, Web of Science, and CENTRAL to June 2025. Eligible studies were RCTs in adults aged ≥ 60 with sarcopenia, comparing resistance training with usual care or no intervention, and reporting outcomes on strength, physical function, or muscle mass. Risk of bias was assessed using RoB 2. Meta-analyses were conducted using the meta package in R, and Bayesian dose–response models were fitted using the brms package.

Results

Twenty-four randomized controlled trials involving 951 participants were included. Random-effects model showed that resistance training significantly improved handgrip strength, gait speed, knee extension strength, timed up and go test (TUG) and and five-times sit-to-stand test (5STS) performance. However, no significant improvements were observed in the short physical performance battery (SPPB), appendicular skeletal muscle mass index and appendicular skeletal muscle mass. Subgroup analyses revealed significant differences across resistance type, frequency, and setting, although meta-regression identified no significant sources of heterogeneity. A nonlinear Bayesian random-effects model suggested an optimal dose of 1220 MET-min/week for improving handgrip strength, while a minimal effective dose of 600 MET-min/week may suffice to achieve clinically meaningful improvements in gait speed.

Conclusion

Resistance training probably improves muscle strength and physical function in older adults with sarcopenia. However, improvements in grip strength, gait speed, TUG, 5STS, and SPPB did not exceed their MID thresholds, indicating little to no clinical benefit. Resistance type and training frequency were key effect modifiers. Individualized resistance programs within the optimal dose range, emphasizing higher frequency and appropriate resistance types, may help optimize outcomes.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s40520-025-03235-w.

Introduction

Sarcopenia is an age-related condition characterized by the progressive loss of skeletal muscle mass and function. It is closely associated with adverse health outcomes, including functional impairment, mobility limitations, increased risk of falls and fractures, higher rates of hospitalization, and increased mortality [1, 2]. The prevalence of sarcopenia is estimated to be around 10% among adults over 60 years old, rising to over 50% in those aged 80 and above [3]. With the ongoing global trend of population aging, the prevalence of sarcopenia is expected to increase continuously, imposing greater burdens on healthcare systems and resulting in escalating healthcare costs [4].

Sarcopenia has become a significant issue in geriatric medicine, posing a serious threat to older adults’ independence and quality of life [5]. To date, there are no approved pharmacological treatments for the routine clinical management of sarcopenia [6]. Resistance training, as a non-pharmacological intervention, has been widely demonstrated to safely and effectively improve muscle strength, physical function, and muscle mass in older adults [7, 8]. However, there is still controversy regarding the optimal resistance training protocol for patients with sarcopenia. A review by Smith et al. [9] reported that, despite the variability in diagnostic criteria for sarcopenia, international exercise prescription guidelines for older adults and other common clinical populations with similar characteristics are, in principle, highly consistent. Thus, regardless of whether an individual is formally diagnosed with sarcopenia or which diagnostic criteria are used, it may not be necessary to develop a dedicated resistance training program.

Nevertheless, it must be acknowledged that, compared with the general older population, patients with sarcopenia more frequently present with pronounced deficits in strength, balance, flexibility, and cognitive function [10, 11], and often have multiple comorbidities (e.g., osteoarthritis, osteoporosis, obesity, and diabetes) [12]. These factors tend to reduce exercise tolerance, increase fatigue sensitivity, and limit the types and intensities of exercise that can be safely implemented [13, 14]. Furthermore, most existing resistance training guidelines are based on studies in healthy older adults and focus on prevention rather than treatment of sarcopenia, lacking systematic strategies to address the functional impairments and comorbidities unique to this population [15]. Therefore, clarifying and optimizing training parameters, especially frequency, duration, and intensity, remains a key priority for the standardized management of sarcopenia.

A preliminary investigation by Chen et al. [16] explored resistance training variables in patients with sarcopenia, demonstrating that training modality, frequency, intensity, and duration all have a regulatory effect on the benefits of resistance training in this population. However, limited by the number and sample size of included studies, their analysis could only combine various outcome measures (such as muscle strength, physical function, and muscle mass) rather than systematically distinguishing training adaptations for specific outcomes. To address this, the present study comprehensively updates the literature, utilizing subgroup and regression analyses to evaluate further the impact of different training variables on the major outcomes of sarcopenia. In addition, we extend prior work by using a Bayesian framework to model the nonlinear dose–response relationship between training volume and outcome improvement. This approach flexibly captures nonlinearity via splines, stabilizes estimates in sparse data regions with weakly informative priors, and enables clinically meaningful estimation of key parameters, such as the minimum effective and optimal training doses. Together, these advantages strengthen the evidence base for optimizing resistance training prescriptions in older adults with sarcopenia.

Methods

Protocol and registration

The protocol of this systematic review and meta-analysis was registered in PROSPERO (CRD420251075410). This study was conducted in accordance with the PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [17].

Search strategy and study selection

We systematically searched PubMed, Embase, the Cochrane Central Register of Controlled Trials (CENTRAL), and Web of Science from inception to June 2025. Search strategies combined Medical Subject Headings (MeSH) and relevant free-text terms, including: (“Sarcopenia“[MeSH Terms]) AND (“Resistance training“[MeSH Terms]) AND ((“Older“[Title/Abstract]) OR (“Aged“[MeSH Terms]) OR (“Elderly“[Title/Abstract])). Detailed search strategies are provided in Supplementary 1. Three independent reviewers (GS, BW, and LX) conducted the literature search and screened eligible studies. Discrepancies were resolved through discussion with a fourth reviewer (YE). The reference lists of included articles and relevant systematic reviews were also screened to identify potentially eligible studies.

Eligibility criteria

Eligibility was assessed using the PICOS framework (Participants, Interventions, Comparators, Outcomes, and Study design) [18]. Studies were included if they met al.l of the following criteria: (1) Participants: adults aged ≥ 60 years diagnosed with sarcopenia, regardless of specific diagnostic criteria, including but not limited to those proposed by EWGSOP, AWGS, or other authoritative bodies, or self-defined criteria, provided the diagnosis was based on at least one of the following core components: low muscle mass, low muscle strength, or impaired physical function; (2) Intervention: Structured resistance training delivered either alone or as the primary component alongside minor ancillary elements (e.g., balance or flexibility exercises), with no restrictions on training type, intensity, or equipment; (3) Comparator: health education, usual care, or no intervention; (4) Outcomes: Based on the revised EWGSOP2 consensus, which defines low muscle strength, low muscle mass, and low physical performance as the core diagnostic criteria for sarcopenia, the following were prespecified as primary outcomes: handgrip strength, gait speed, and appendicular skeletal muscle mass index (ASMI), calculated as appendicular skeletal muscle mass divided by height squared (kg/m²) and measured using BIA or DXA. The following were considered secondary outcomes due to their complementary or alternative roles in sarcopenia assessment: knee extension strength, timed up and go test (TUG), and appendicular skeletal muscle mass (ASM). In addition, in line with peer-review feedback and the EWGSOP2 recommendations, we included the five-times sit-to-stand test (5STS) and the short physical performance battery (SPPB) as additional secondary outcomes.; (5) Study design: randomized controlled trials (RCTs).

Studies were excluded if they met any of the following criteria: (1) Participants with sarcopenia secondary to specific health conditions such as cancer, diabetes, stroke, HIV/AIDS, chronic obstructive pulmonary disease, chronic kidney disease, cirrhosis, other severe diseases, or recent organ transplantation; (2) Non-English publications; (3) Studies lacking sufficient data; (4) Studies for which the full report could not be obtained through databases or other means.

Data extraction

For each eligible study, data were independently extracted using a predefined form, including study characteristics (first author, year of publication, country, setting, diagnostic criteria for sarcopenia), population characteristics (age, sex, sample size), intervention characteristics (intensity, duration, frequency, and session length), and outcome data (means and standard deviations for continuous outcomes). If multiple follow-up time points were reported, data from the longest follow-up duration were extracted to reflect the sustained effects of the intervention. For studies with missing data, the corresponding author was contacted thrice over three weeks. Two independent reviewers (GS and BW) performed data extraction, with a third reviewer (LX) verifying and adjudicating discrepancies.

Measures of treatment effect

This meta-analysis evaluated treatment effects using changes in mean differences (MD) and standard deviations (SD). When SD were not directly reported in the original studies, they were estimated using alternative parameters such as standard error (SE), 95% confidence intervals (CI), p-values, or t-statistics, following established statistical methods [19]. For studies that did not report the standard deviation of the change from baseline, we estimated it using the following formula:

Where r represents the assumed correlation coefficient between baseline and post-intervention measurements, in this analysis, we adopted a conservative estimate of r = 0.5, reflecting moderate test–retest reliability as widely accepted in the literature. This value was selected to balance potential variability between time points and enhance the pooled estimates’ robustness and credibility [19].

Quality assessment of evidence

The risk of bias in included RCTs was comprehensively evaluated using the Cochrane Risk of Bias tool (ROB 2.0), covering domains such as random sequence generation, allocation concealment, blinding, incomplete outcome data, and selective reporting [20]. A study was rated as having an overall low risk of bias if all domains were judged low risk; if any domain was judged high risk, the study was considered high risk; otherwise, the risk was judged as “some concerns.” Two reviewers assessed the risk of bias independently, resolving disagreements through consensus.

Evidence quality was further assessed using the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) approach with GRADEpro GDT (www.gradepro.org). Five domains were considered for rating the certainty of evidence as high, moderate, low, or very low: risk of bias, inconsistency, indirectness, imprecision, and publication bias [21]. All assessments were performed independently, and discrepancies were resolved by consensus. To assess small-study effects and publication bias, funnel plots were generated and analyzed for each direct comparison.

Statistical analysis

Meta-analyses, subgroup analyses, and meta-regression were conducted using the meta package in R software (version 4.3.1) [22]. Restricted maximum likelihood (REML) was used to estimate random-effects model parameters for more robust effect estimates [23]. Given that several included studies had small sample sizes and moderate to high heterogeneity, the Hartung–Knapp method was further applied to adjust confidence intervals and improve estimate precision [24]. Overall effect testing was performed using t-statistics, with statistical significance set at p < 0.05 [24]. To assess the expected range of effects in future comparable studies, 95% prediction intervals (PIs) were calculated [25]. Intervals entirely on one side of zero suggested consistent and potentially meaningful effects. Intervals spanning zero were interpreted with context: wide intervals indicated substantial heterogeneity and uncertainty, whereas narrow intervals centred around zero suggested a likely absence of effect with high certainty.

For continuous outcomes, we reported either mean differences (MD) or standardized mean differences (SMD) with standard deviations and 95% confidence intervals. MDs were used when outcome measures were reported in consistent units across studies, while SMDs were applied when studies used different scales or units (e.g., knee extension strength). Endpoint values were prioritized, but pre-post change values were used if available. Owing to the generally small sample sizes, Hedges’ g was used to estimate effect sizes and classified as small (g = 0.2), medium (g = 0.5), large (g = 0.8), or very large (g = 1.2) [26]. Heterogeneity was assessed by the I² statistic: <40% indicated low, 40–75% moderate, and >75% high heterogeneity [27]. Egger’s test evaluated potential publication bias in addition to funnel plot inspection [28]. Sensitivity analyses were performed by sequentially omitting each study to assess its impact on the overall effect and heterogeneity [29]. The minimal important difference (MID) for key outcomes was referenced from previous studies: handgrip strength (5.0 kg) [30], gait speed (0.10 m/s) [31], TUG (1.6 s) [32], 5STS (2.3 s) [33], SPPB (1.0 point) [34].

Additional data analyses

When substantial heterogeneity was observed, subgroup and meta-regression analyses were conducted to explore potential effect modifiers. Subgroup analyses of categorical variables were performed based on setting (institutional vs. community), resistance training type (variable, constant, or combined), load (≥ 70% 1RM vs. <70% 1RM), repetition speed (high-speed vs. low-speed), diagnostic criteria (EWGSOP 2010 vs. AWGS 2014 vs. Study-defined criteria) and training frequency (2 vs. 3 times/week). A p-value < 0.1 for interaction was considered statistically significant [35]. For continuous moderators with at least 10 studies, random-effects meta-regression models were used to examine the influence of factors such as percentage of males, sample size, mean age, training frequency, session duration, intervention period, number of sets, repetitions, and rest intervals. Statistical significance was set at p < 0.05 [27].

To assess the dose-response relationship between exercise dose and changes in handgrip strength and gait speed, Bayesian random-effects regression models with nonlinear functions were constructed using the brms package in R (version 2.18.0) [36]. Results were visualized with tidybayes and ggplot2. MD was used as the dependent variable, with natural spline functions (k = 4) modeling weekly exercise dose (METs-min/week) to reflect nonlinear associations with muscle strength and gait speed [37]. In this study, METs-min/week was calculated as the product of metabolic equivalents (METs), session duration, and weekly frequency based on the 2011 Compendium of Physical Activities, which includes 821 activity codes covering nearly all exercise types [38]. For resistance training, METs of 3.5, 5.0, or 6.0 were used based on training modality and intensity (Table 1). Frequency was defined as the total number of weekly exercise sessions, including multiple daily sessions. If exercise duration was gradually increased during the intervention, the mean value over the intervention period was used. For interpretability, two thresholds were defined from the modelled dose–response curve. The minimum effective dose was the lowest weekly training volume at which the lower bound of the 95% credible interval of the predicted effect exceeded zero, indicating a statistically credible improvement. The optimal dose corresponded to the weekly volume at which the posterior mean effect reached its maximum.

Table 1.

TMETs for Resistance Training

Code	Mets	Major Heading	Specific activities
2050	6.0	Conditioning exercise	Resistance training (weight lifting, free weight, nautilus or universal), power lifting or body building, vigorous effort (Taylor Code 210)
2052	5.0	Conditioning exercise	Resistance (weight) training, squats, slow or explosive effort
2054	3.5	Conditioning exercise	Resistance (weight) training, multiple exercises, 8–15 repetitions at varied resistance

Results

Literature selection and study characteristics

A total of 4,195 potential records were identified through systematic searches. After removing duplicates, 3,022 articles remained for title and abstract screening. Of these, 81 articles underwent full-text review based on eligibility criteria. Ultimately, 24 studies were included in this review and meta-analysis, comprising 951 participants with a mean age of 73.45 ± 6.42 years and a male proportion of 19.9%. Sarcopenia was diagnosed using the EWGSOP 2010 criteria in eight studies, the AWGS 2014 criteria in three, the CDC guidelines in one, and self-defined criteria based on muscle mass, strength, or function in twelve. The complete screening and selection process is presented in Fig. 1, and the detailed characteristics of the included studies are summarized in Supplementary 2.

Fig. 1.

PRISMA flow diagram of the study selection process

Risk of bias and certainty of evidence

The risk of bias for each trial is presented in Fig. 2. Overall, 8 studies (33.3%) were classified as low risk of bias, 12 studies (50.0%) as having unclear risk, and 4 studies (16.7%) as high risk (Supplementary 3). All studies explicitly described random sequence generation, but 14 did not mention allocation concealment and were rated as “some concerns” in this domain. Four studies had a high attrition rate for missing outcome data without reporting reasons or handling methods and were rated as “some concerns.” Regarding selective reporting bias, 13 studies did not provide pre-specified analysis plans nor register their protocols and were thus rated as “some concerns.”

Fig. 2.

The overall risk of bias is the percentage of each risk of bias item across included studies

Evidence quality was systematically assessed using the GRADE framework (Table 2). Studies on handgrip strength, knee extension strength, TUG, 5STS, SPPB and ASM were downgraded for risk of bias. For knee extension strength TUG and ASMI, high heterogeneity (I² >75%) led to downgrading for inconsistency, even though subgroup analysis explained some variance. For 5STS, SPPB ASMI and ASM, small sample sizes and confidence intervals crossing the line of no effect resulted in downgrading for imprecision. Overall, the certainty of evidence for most outcomes ranged from moderate to high.

Table 2.

GRADE summary of evidence

Primary outcomes

Seventeen studies (n = 646) reported changes in handgrip strength. Results showed that resistance training significantly improved handgrip strength in older adults with sarcopenia compared with usual care (MD = 2.30 kg, 95% CI: 1.33 to 3.28), with moderate heterogeneity (I² = 60.0%). However, this effect did not exceed the MID of 5.0 kg. (Fig. 3a).

Fig. 3.

Forest plots of primary outcomes (a) Handgrip strength (b) Gait speed (c) ASMI

Twelve studies (n = 501) reported gait speed. Resistance training led to a significant improvement in gait speed compared with usual care (MD = 0.08 m/s, 95% CI: 0.03 to 0.14), with moderate heterogeneity (I² = 51.7%). However, this effect did not exceed the MID of 0.1 m/s.(Fig. 3b).

Seven studies (n = 285) reported changes in ASMI. Resistance training did not significantly improve ASMI compared with usual care (MD = 0.26 kg/m², 95% CI: -0.03 to 0.66), with high heterogeneity (I² = 83.7%).

Secondary outcomes

Eleven studies (n = 436) reported changes in knee extension strength. Resistance training significantly improved knee extension strength compared with usual care (SMD = 1.04, 95% CI: 0.59 to 1.49), with moderate heterogeneity (I² = 73.6%) (Fig. 4a).

Fig. 4.

Forest plots of secondary outcomes (a) Knee extension strength (b) TUG (c) ASM

Ten studies (n = 408) reported TUG test results. Resistance training was associated with a significant improvement in TUG performance (MD = − 1.36 s, 95% CI: − 1.94 to − 0.79) compared with usual care, with high heterogeneity (I² = 84.3%). However, this effect did not exceed the MID of 2.1s.(Fig. 4b).

Ten studies (n = 405) reported changes in ASM. Resistance training showed no effect on ASM compared with usual care (MD = 0.02 kg, 95% CI: − 0.10 to 0.15), with zero heterogeneity (I² = 0%) (Fig. 4c).

Three studies (n = 111) reported changes in 5STS performance. Resistance training significantly improved 5STS compared with usual care (MD = − 1.29 s, 95% CI: − 1.70 to − 0.88), with no heterogeneity (I² = 0%). However, this effect did not exceed the MID of 2.3s.(Fig. 5a).

Fig. 7.

Leave-one-out sensitivity analyses for (a) handgrip strength, (b) gait speed: Impact on heterogeneity (I²) and effect size

Three studies (n = 101) reported SPPB results. Resistance training showed no effect on SPPB compared with usual care (MD = 0.16 point, 95% CI: − 0.38 to 0.70), with no heterogeneity (I² = 0%). (Fig. 5b).

Fig. 5.

Forest plots of secondary outcomes (a) 5STS (b) SPPB

Dose-Response relationships

We observed a nonlinear, reverse J-shaped dose–response relationship between resistance training dose and improvements in handgrip strength (Fig. 6). The minimum effective dose was 490 METs-min/week. The magnitude of handgrip strength improvement increased progressively with higher weekly training doses, reaching a peak at approximately 1220 METs-min/week. Beyond this dose, further increases in training volume did not confer additional benefits, and the improvement in handgrip strength began to decline. In contrast, a positive dose–response relationship was observed between resistance training dose and improvement in usual gait speed (Fig. 6). The minimum effective dose was 200 METs-min/week, and gait speed continued to increase with higher training doses, with no apparent plateau. Clinically meaningful improvements may be achieved when the training dose exceeds approximately 600 METs-min/week.

Fig. 6.

Dose–Response Relationship Between Weekly Exercise Dose (METs-min/week) and Handgrip Strength and Gait Speed

Subgroup analysis

Subgroup analyses identified several important effect modifiers (Table 3); related forest plots in Supplementary 4). For handgrip strength, training frequency showed a significant subgroup difference (χ² = 5.00, p = 0.03): interventions with three sessions per week (MD = 3.18 kg, 95% CI: 2.08 to 4.28) achieved greater improvements than two sessions per week (MD = 1.42 kg, 95% CI: 0.35 to 2.50). For knee extension strength, both resistance type (χ² = 19.68, p < 0.001) and setting (χ² = 6.45, p = 0.01) were significant subgroup factors. Constant resistance training (SMD = 1.53, 95% CI: 0.80 to 2.25) yielded greater improvement than combined resistance (SMD = 0.50, 95% CI: − 0.24 to 0.77). Institutionalized participants showed greater improvement (SMD = 1.04, 95% CI: 0.59 to 1.49) than community participants. For gait speed, resistance type was also significant (χ² = 8.49, p = 0.01): combined resistance training (MD = 0.13 m/s, 95% CI: 0.09 to 0.18) was superior to constant resistance (MD = 0.08 m/s, 95% CI: − 0.06 to 0.22) and variable resistance (MD = − 0.01 m/s, 95% CI: − 0.09 to 0.08). For ASMI, training frequency was identified as a significant effect modifier (χ² = 0.58, p = 0.08). Interventions delivered three times per week (MD = 0.40 kg/m², 95% CI − 0.03 to 0.83) produced greater improvements than those delivered twice per week (MD = 0.00 kg/m², 95% CI − 0.07 to 0.07), although the within-subgroup effects did not reach statistical significance. Notably, significant improvement in ASMI was observed exclusively in trials employing constant resistance training. By contrast, no evidence of subgroup effects was found for ASM.

Table 3.

Summary of subgroup analysis results

Outcome	Number of participants	Heterogeneity		Meta analysis		Subgroup differences
		P value	I2	Effect estimate (95%CI)	P value
Meta-analysis results by type of resistance
Handgrip strength	646	<0.001*	60.00%	3.28(1.60,4.95)	<0.001*	0.36
Constant resistance	240	0.12	39.50%	1.89(0.61,3.17)
Mixed resistance	330	0.001*	72.50%	2.30(1.33,3.28)
Variable resistance	76	0.78	0.00%	1.56(-0.87,3.99)
Knee extension strength	436	<0.001*	51.70%	0.08(0.03,0.14)	<0.001*	<0.001*
Constant resistance	161	0.01*	66.80%	1.29(0.64,1.93)
Mixed resistance	229	0.6	0.00%	0.50(0.24,0.77)
Variable resistance	46			1.04(0.59,1.49)
Gait speed	501	0.02*	51.70%	0.08(0.03,0.14)	0.002*	0.01*
Constant resistance	126	0.02*	68.60%	0.08(-0.6,0.22)
Mixed resistance	253	0.84	0.00%	0.13(0.09,0.18)
Variable resistance	122	0.39	0.00%	-0.01(-0.09,0.08)
Timed up and go test	408	<0.001*	84.30%	-1.34(-1.94,-0.79)	<0.001*	0.86
Constant resistance	72	<0.001*	85.6%%	-1.08(-2.63,0.47)
Mixed resistance	214	<0.001*	89.70%	-1.42(-2.38,-0.47)
Variable resistance	122	0.69	0.00%	-1.56(-2.32,-0.80)
Appendicular skeletal muscle mass index	285	<0.001*	83.70%	0.26(-0.03,0.55)	0.07	0.75
Constant resistance	73	0.2511	27.60%	0.27(0.01,0.54)
Mixed resistance	136	<0.001*	93.60%	0.60(-0.66,1.87)
Variable resistance	76	0.8606	0.00%	0.16(-0.20,0.52)
Appendicular skeletal muscle mass	405	0.7989	0.00%	0.02(-0.10,0.15)	0.71	0.8
Constant resistance	76	0.857	0.00%	0.09(-0.30,0.48)
Mixed resistance	253	0.342	11.20%	0.01(-0.12,0.14)
Variable resistance	76	0.846	0.00%	0.02(-0.58,1.18)
Meta-analysis results by load
Handgrip strength	244	0.0831*	46.30%	3.29(1.77,4.81)	<0.001*	0.72
≥ 70%1RM	98	0.9232	0.00%	2.71(1.35,4.07)
< 70%1RM	146	0.0512*	57.60%	3.30(0.36,6.23)
Knee extension strength	146	0.0081*	71.00%	1.42(0.69,2.16)	<0.001*	0.94
≥ 70%1RM	86	0.0052*	0.872	1.51(-0.44,3.46)
< 70%1RM	60	0.081	0.602	1.44(0.65,2.22)
Meta-analysis results by frequency
Handgrip strength	646	<0.001*	60.00%	3.28(1.60,4.95)	<0.001*	0.03*
2 times/week	313	0.08*	44.20%	1.42(0.35,2.50)
3 times/week	333	0.11	39.40%	3.18(2.08,4.28)
Knee extension strength	436	<0.001*	51.70%	0.08(0.03,0.14)	<0.001*	0.12
2 times/week	346	0.03*	55.30%	0.76(0.42,1.10)
3 times/week	90	0.02*	74.60%	1.65(0.58,2.71)
Gait speed	501	0.02*	51.70%	0.08(0.03,0.14)	0.002*	0.97
2 times/week	336	0.27	21.50%	0.08(0.03,0.14)
3 times/week	165	0.005*	73.00%	0.09(-0.03,0.20)
Timed up and go test	408	<0.001*	84.30%	-1.34(-1.94,-0.79)	<0.001*	0.89
2 times/week	140	<0.001*	93.90%	-1.36(-1.94,-0.79)
3 times/week	268	<0.001*	82.90%	-1.40(-2.04,-0.77)
Appendicular skeletal muscle mass index	285	<0.001*	87.30%	0.26(-0.03,0.55)	0.07	0.08
2 times/week	91	0.6189	0.00%	0.00(-0.07,0.07)
3 times/week	194	0.034	61.60%	0.40(-0.03,0.83)
Appendicular skeletal muscle mass	405	0.7989	0.00%	0.02(-0.10,0.15)	0.71	0.8
2 times/week	247	0.4297	0.00%	0.00(-0.14,0.13)
3 times/week	158	0.914	0.00%	0.18(-0.14,0.50)
Meta-analysis results by repetition speed
Knee extension strength	197	<0.001*	81.70%	1.17(0.45,1.88)	0.001	0.77
High speed	43	0.2083	36.80%	1.05(0.22,1.88)
Low speed	154	<0.001*	88.30%	1.25(0.20,2.30)
Meta-analysis results by setting
Handgrip strength	646	<0.001*	60.00%	3.28(1.60,4.95)	<0.001*	0.94
Community	281	0.008*	61.60%	2.26(0.61,3.90)
Institutionalized	365	0.008*	63.50%	2.34(1.11,3.57)
Knee extension strength	436	<0.001*	51.70%	0.08(0.03,0.14)	<0.001*	0.01*
Community	266	0.39	4.30%	0.59(0.33,0.86)
Institutionalized	170	<0.001*	84.60%	1.81(0.91,2.71)
Gait speed	501	0.02*	51.70%	0.08(0.03,0.14)	0.002*	0.48
Community	268	0.44	0.00%	0.11(0.06,0.16)
Institutionalized	233	0.007*	66.00%	0.08(0.03,0.14)
Timed up and go test	408	<0.001*	84.30%	-1.34(-1.94,-0.79)	<0.001*	0.008*
Community	136	0.74	0.00%	-2.19(-2.79,-1.59)
Institutionalized	272	<0.001*	80.90%	-1.03(-1.64,-0.42)
Appendicular skeletal muscle mass index	285	<0.001*	87.30%	0.26(-0.03,0.55)	0.07	0.4
Community	45	0.9793	0.00%	0.10(-0.23,0.44)
Institutionalized	240	<0.001*	89.10%	0.34(-0.11,0.79)
Appendicular skeletalmuscle mass	405	0.7989	0.00%	0.02(-0.10,0.15)	0.71	0.8
Community	259	0.6275	0.00%	0.06(-0.22,0.33)
Institutionalized	146	0.618	0.00%	0.02(-0.12,0.15)
Meta-analysis results by diagnostic criteria
Handgrip strength	565	0.0013*	61.50%	2.39(1.27,3.52)	<0.001*	0.21
EWGSOP 2010	209	0.4726	0.00%	3.32(1.91,4.73)
AWGS 2014	94	0.22	33.50%	3.32(1.33,5.32)
Study-defined criteria	262	0.0042*	70.90%	1.28(-0.67,3.23)
Knee extension strength	400	<0.001*	73.20%	0.97(0.52,1.43)	<0.001*	0.82
EWGSOP 2010	131	0.0086*	74.30%	1.04(0.29,1.79)
Study-defined criteria	269	0.0031*	74.90%	0.93(0.30,1.56)
Gait speed	485	0.0117*	56.00%	0.08(0.03,0.14)	0.0038*	0.38
EWGSOP 2010	179	0.0303*	62.60%	0.05(-0.04,0.14)
Study-defined criteria	306	0.0377*	57.60%	0.10(0.03,0.18)
Timed up and go test	408	<0.001*	84.30%	-1.36(-1.94,-0.79)	<0.001*	0.43
EWGSOP 2010	123	0.58	0.00%	-1.09(-1.52,-0.66)
Study-defined criteria	285	<0.001*	89.80%	-1.45(-2.26,-0.64)
Appendicular skeletal muscle mass index	225	<0.001*	78.20%	0.15(-0.04,0.35)	0.13	0.77
EWGSOP 2010	45	0.9793	0.00%	0.10(-0.23,0.44)
Study-defined criteria	180	<0.001*	86.90%	0.17(-0.07,0.42)
Appendicular skeletal muscle mass	389	0.8533	0.00%	0.02(-0.11,0.14)	0.79	0.25
EWGSOP 2010	157	0.9987	0.00%	0.32(-0.21,0.86)
Study-defined criteria	232	0.4571	0.00%	0.00(-0.13,0.13)

Meta-regression

We performed multiple univariate random-effects meta-regression analyses to assess the impact of continuous covariates on treatment effects (Table 4). Training frequency was a significant covariate for both handgrip strength (β = 1.7570, p = 0.0255) and knee extension strength (β = 0.8682, p = 0.0407), suggesting that higher training frequency may be associated with greater strength improvements. Other covariates (sample size, mean age, male proportion, session duration, intervention period) showed no significant associations with primary outcomes (Supplementary 5).

Table 4.

Random-effects meta-regression analyses by potential covariates

Outcome	Covariate	Univariate meta-regression
		β	95% lower	95% upper	P-value	R2
Handgrip strength	Mean age	-0.0521	-0.2082	0.104	0.4823	0.00%
	Period	0.0157	-0.1444	0.1757	0.8478	0.00%
	Duration	0.0361	-0.0397	0.112	0.3506	0.00%
	Frequency	1.757	0.2151	3.2989	0.0255*	60.25%
	Male%	2.0246	-1.5799	5.6292	0.2709	10.83%
	Sample size	-0.0338	-0.0831	0.0154	0.1784	14.87%
	Rest interval	0.0065	-0.021	0.034	0.6425	0.00%
	Set	0.234	-0.5271	0.9951	0.5468	0.00%
	Repetitions	-0.0406	-0.3406	0.2595	0.7911	0.00%
Gait speed	Mean age	0.0028	-0.0069	0.0125	0.5721	0.00%
	Period	-0.0026	-0.0135	0.0083	0.6375	0.00%
	Duration	-0.0014	-0.0087	0.006	0.7185	0.00%
	Frequency	0.0075	-0.1052	0.1203	0.8959	0.00%
	Male%	0.148	-0.0273	0.3234	0.098	11.93%
	Sample size	0.0005	-0.0021	0.0032	0.6917	0.00%
	Set	0.0321	-0.0556	0.1199	0.4729	0.00%
	Repetitions	0.0044	-0.0162	0.025	0.6763	0.00%
Knee extension strength	Mean age	-0.0205	-0.0895	0.0484	0.5592	0.00%
	Period	-0.0167	-0.1138	0.0803	0.7356	0.00%
	Duration	-0.0088	-0.0574	0.0399	0.7232	0.00%
	Frequency	0.8682	0.0027	1.7391	0.0407*	42.71%
	Male%	1.5005	-0.1122	3.1132	0.0682	23.72%
	Sample size	-0.0128	-0.0328	0.0071	0.2062	9.93%
	Set	-0.1083	-0.4145	0.198	0.4884	0.00%
	Repetitions	0.0054	-0.1271	0.1379	0.9361	0.00%
Timed up and go test	Mean age	-0.0922	-0.2244	0.04	0.1717	24.67%
	Period	-0.0562	-0.2717	0.1592	0.6089	0.00%
	Duration	0.0186	-0.0974	0.1345	0.7538	0.00%
	Frequency	-0.2074	-1.608	1.1931	0.7716	0.00%
	Male%	0.1901	-1.2007	1.581	0.7888	0.00%
	Sample size	0.0033	-0.0289	0.0356	0.8392	0.00%
	Set	0.2648	-0.6489	1.1785	0.57	0.00%
	Repetitions	-0.1968	-0.6656	0.2719	0.4104	0.00%
Appendicular skeletal muscle mass	Mean age	-0.0023	-0.0504	0.0458	0.9259	0.00%
	Period	0.0028	-0.0864	0.0919	0.9515	0.00%
	Duration	-0.0271	-0.0666	0.0124	0.1783	0.00%
	Frequency	0.1884	-0.1578	0.5346	0.2861	0.00%
	Male%	0.078	-0.3438	0.4999	0.7169	0.00%
	Sample size	-0.003	-0.0094	0.0033	0.3521	0.00%
	Repetitions	0.0435	-0.0393	0.1263	0.3031	0.00%

Sensitivity analysis

Leave-one-out sensitivity analyses were conducted to evaluate the impact of individual studies on pooled effect estimates and heterogeneity. Results showed that removing any single study did not materially change the overall effect or statistical significance, and I² values remained stable, supporting the robustness of the findings (Fig. 7 and Supplementary 6). Additionally, as three studies reporting handgrip strength did not include upper limb resistance training, we excluded them and performed an additional sensitivity analysis. The pooled estimates and significance levels remained largely unchanged, further reinforcing the robustness of this outcome (Supplementary 6).

Publication bias

Egger’s test was used to evaluate potential publication bias for each outcome. Significant publication bias was detected for knee extension strength (p = 0.0398). The trim-and-fill method identified one potentially missing study, but adjusted effect sizes remained significant (SMD = 0.93, 95% CI: 0.43 to 1.42), suggesting the overall robustness of the results. No evidence of publication bias was found for other outcomes: handgrip strength (p = 0.4054), gait speed (p = 0.3345), ASMI (p = 0.3205), TUG (p = 0.2426), and ASM (p = 0.1657).

Adverse events

Four included studies reported adverse events related to resistance training, all of which were mild, reversible, and did not result in study withdrawal. Specifically, Vasconcelos et al. [39] reported 4 events among 14 participants (28.6%), Rufino et al. [40] reported 3 events among 20 participants (15.0%), and Kim et al. [41] reported 2 events among 32 participants (6.3%). Iranzo et al. [42] mentioned adverse events but did not provide specific numbers. Most of the reported events were transient symptoms such as joint discomfort, muscle soreness, or fatigue at the beginning of the intervention, which typically resolved after adjusting training intensity or technique. These findings support the safety and tolerability of resistance training in older adults with sarcopenia.

Discussion

This meta-analysis provides robust evidence that resistance training probably improves muscle strength and physical function in older adults with sarcopenia, although its effect on muscle mass remains limited. Further subgroup and meta-regression analyses showed that resistance type was a significant moderator for handgrip strength, knee extension strength, and gait speed. Training frequency significantly affected improvements in handgrip strength. The training setting also influenced knee extension strength and TUG performance. In contrast, other variables, such as training load, repetitions, tempo, number of sets, intervention duration, and participant characteristics, including sex, age, and sample size, were not identified as significant effect modifiers.

Among all training variables, load is traditionally considered one of the most important factors influencing resistance training outcomes. However, only 10 of the 24 included studies clearly reported training load. High-load resistance training is generally believed to maximize neuromuscular adaptations and is an effective strategy for improving muscle strength [43]. Jenkins et al. [44] confirmed that, compared with low-load training at 30% 1RM, high-load training at 80% 1RM results in lower neural activation requirements for achieving equivalent submaximal torque, with greater gains in muscle strength, while the effects on muscle hypertrophy are similar. Previous meta-analyses have also shown that high-load resistance training is superior to moderate or low-load interventions for enhancing muscle strength, even in frail older adults [45]. Although high-load resistance training may confer greater neuromuscular benefits under controlled conditions, its implementation in frail or functionally impaired older adults may be impractical due to equipment requirements, supervision needs, and increased safety risks [15]. Moreover, adherence to such high-load protocols in this population is often limited, further reducing their feasibility [46]. High relative intensities can, however, be achieved through lower-cost modalities such as bodyweight or elastic resistance exercises, which may offer more feasible and scalable alternatives [47, 48].

In addition, previous studies have found that power-based resistance training, which emphasizes fast movement, is often more effective than slow-velocity training for improving muscle strength and function in older adults. This effect is usually observed at low to moderate intensities (40–60% 1RM) [49, 50]. In daily life, functional capacity often depends more on rapidly producing force than absolute maximal strength. For example, Ramírez-Campillo et al. [49] reported that high-speed resistance training produced greater improvements than low-speed training for leg press (36.2% vs. 29.2%), maximal gait speed (14.1% vs. 8.7%), and TUG (17.6% vs. 9.7%) in older women. However, our analysis did not observe a significant difference between high- and low-speed training in improving knee extension strength among older adults with sarcopenia. Given the limited number of studies reporting repetition tempo, these results should be interpreted cautiously and require further confirmation in larger, high-quality studies.

Some evidence also suggests that variable resistance training may be more beneficial than constant resistance training for improving physical function in older adults [51]. Our findings support this view. We found that constant resistance training was less effective than variable resistance training in improving gait speed and TUG, although it had a greater effect on muscle strength. Walker et al. [52] found that variable resistance training and constant resistance training produced similar gains in muscle strength and hypertrophy, but variable resistance training was superior in enhancing fatigue resistance. This adaptation may help older adults better maintain their capacity for daily movement and functional tasks. Smith et al. [53] also reported that variable resistance training significantly reduced the electromechanical delay of the knee reflex, leading to faster neuromuscular response and better coordination. These neurological adaptations are important for improving motor control, reducing the risk of falls, and supporting independent living in later life. In our analysis, combined resistance training, which included constant and variable resistance, was most effective for improving gait speed. This approach leverages the advantages of both modalities and provides diverse neuromuscular stimulation, leading to more comprehensive adaptations that benefit functional tasks such as walking, rising, and turning. Therefore, for patients with sarcopenia, it is advisable to match the structure of resistance training with functional goals and emphasize training diversity and specificity rather than focusing on load alone.

Previous research has shown that the dose-response relationship between training volume and skeletal muscle hypertrophy follows a U-shaped curve [54]. A recent review by Bernárdez-Vázquez et al. [55] identified a similar pattern and recommended a minimum of 10 sets per muscle group per week to achieve optimal muscle hypertrophy. Notably, a recent study among previously untrained older women found that, in the early phase of moderate-intensity resistance training, once the minimum effective training volume was reached, improvements in muscle mass, maximal strength, and neuromuscular recruitment were similar regardless of whether training consisted of one or three sets [56]. Furthermore, a meta-regression analysis showed that moderate training volume (about 24 repetitions per session) was more effective than lower (< 24) or higher (>24) volumes in improving muscle strength [50]. It is important to note that the number of repetitions is determined mainly by training load, with fewer repetitions usually reflecting a higher load, which can lead to greater strength gains. However, some studies suggest that when the total training volume is matched, differences in load do not substantially affect hypertrophy or strength gains [57]. In addition, current evidence does not support training to failure as providing extra physiological benefits for older adults [58]. Performing the maximum number of repetitions possible at 50–70% 1RM under good conditions is sufficient to promote neuromuscular adaptation and reduce the risk of improper technique and injury.

As a key component of the total weekly training volume, training frequency also deserves attention. Existing studies show that resistance training frequency has little independent effect on muscle hypertrophy but may have some influence on muscle strength [59]. This difference may depend mainly on whether the total training volume is equivalent. Training frequency shows a dose-response relationship with strength improvement when the volume is not matched [55]. However, when total training volume is consistent, frequency does not significantly affect muscle strength or hypertrophy gains [60]. Therefore, training frequency itself is not the main factor determining adaptation. Its primary significance lies in spreading training across more sessions, allowing for a higher weekly training volume. Our study found an “optimal dosage window” for improving handgrip strength, with improvement beginning at 490 METs-min/week, peaking around 1220 METs-min/week, and decreasing thereafter. Meta-regression analysis also identified training frequency as a significant moderator, suggesting that when total training volume is the same, increasing frequency and spreading the total workload over more days can optimize muscle strength adaptation, lower per-session intensity and fatigue, and provide a safer and more effective intervention. More research is needed to determine whether increasing frequency (from two to three times per week) can yield additional adaptive benefits under equal total volume conditions.

Our study found that older adults receiving long-term institutional care experienced greater knee extension strength than community-dwelling participants. At the same time, improvements in functional outcomes such as TUG were greater in community dwellers. This difference may reflect that improvements in physical function rely on integrating multiple systems, including cardiorespiratory endurance, balance, flexibility, and activities of daily living. Institutionalized older adults often have poorer health, more comorbidities, and a higher rate of cognitive impairment, which can limit functional improvements and make it more difficult for muscle strength gains to translate into daily activity. Our results support the view proposed by Coelho-Júnior et al.¹¹, calling for resistance training prescriptions to be based on each patient’s functional status and real needs, with dynamic adjustments in intensity, frequency, and content to achieve stratified, graded, and individualized interventions.

Clinical implications

This systematic review and meta-analysis suggests that resistance training probably improves muscle strength and physical function in older adults with sarcopenia. However, for outcomes such as handgrip strength, gait speed, the TUG, SPPB and 5STS test, between-group effects reached statistical significance but did not exceed MID thresholds, indicating that additional clinically perceptible benefits over usual care are uncertain. Nevertheless, with appropriate optimisation of training dose and structure, resistance training appears able to achieve greater effects and reach clinically meaningful improvements.

Our Bayesian dose–response model showed that even at the optimal dose (approximately 244 min per week of moderate-intensity resistance training), handgrip strength improvements remained below the established MID, underscoring limited clinical relevance. Although handgrip strength is a reliable prognostic marker, it showed limited responsiveness to training‐induced adaptations, likely because most included trials relied on general resistance programmes rather than grip‐specific exercises [61]. In line with the principle of task specificity, greater gains would be expected when training directly targets grip function [62]. This raises a critical issue: if handgrip strength is used both as a diagnostic criterion and as the primary endpoint in intervention studies, patients who achieve meaningful improvements in gait speed, lower‐limb strength, or quality of life may still be classified as sarcopenic. Such a mismatch highlights the need to distinguish between “diagnostic” and “monitoring” concepts. While handgrip strength remains useful for case finding and risk stratification, our analysis showed that gait speed, the five‐times sit‐to‐stand test, and knee extension strength demonstrated larger and more consistent improvements, making them better markers of training adaptation. Given that these lower-limb strength and functional outcomes are closely related to mobility, independence, and fall risk, they may warrant consideration as candidate endpoints in future revisions of sarcopenia diagnostic criteria [63]. Notably, gait speed showed a greater likelihood of achieving clinically meaningful improvements compared with usual care when training doses exceeded 600 MET·min per week. Based on the dose–response relationship of gait speed, clinical practice may consider initiating resistance training at approximately 120 min per week of moderate intensity, delivered in three sessions, and progressively increasing training frequency or weekly exercise dose as tolerated, to maximize the likelihood of achieving clinically meaningful improvements.

In addition, the choice of training modality should align with the primary objective: mixed resistance training is likely preferable when the goal is to enhance functional performance, while constant resistance training may be more suitable when the focus is on gains in muscle strength and mass. Although these recommendations are relevant for clinical practice, the evidence base remains limited. Future trials should test higher-volume or multimodal interventions and include participant-centred outcomes, such as fall incidence and health-related quality of life, to better determine the clinical value of resistance training in the management of sarcopenia in older adults.

Strengths and limitations

The main strength of this systematic review and meta-analysis is that it provides the most comprehensive synthesis of the effects of resistance training on primary outcomes in older adults with sarcopenia and the associations with key training variables. For the first time, we used Bayesian random-effects regression modeling to describe the nonlinear dose-response relationship between training dose and primary outcomes such as handgrip strength and gait speed.

Nonetheless, several limitations must be noted. First, the number of studies evaluating muscle mass was limited, restricting our ability to build robust models for this outcome. Some included studies did not report important training variables such as load, repetitions, rest intervals, range of motion, or tempo, which may affect the precision of the results. Second, The training parameters reported by most studies were also relatively homogeneous (duration ≤ 26 weeks, frequency 2–3 sessions per week, and session duration mainly 50–60 min), which may limit the generalizability of the dose-response findings. Future research should focus on improving the standardization and reporting of resistance training protocols and diversifying training parameters to enhance future meta-analyses’ quality and practical value [64]. Third, in response to peer-review feedback and to further align with the EWGSOP2 recommendations, we replaced ASMI with SMI and added 5STS and SPPB as secondary outcomes. Although not prespecified in the original registration, these outcomes were added with transparent reporting to enhance the comprehensiveness and clinical relevance of the study. Finally, our study focused on key training variables to provide general clinical recommendations without stratified or individualized analysis for different subtypes or functional statuses of sarcopenia. As previous studies have pointed out, patients with sarcopenia vary greatly in baseline fitness, health, and disability [11]. Future studies should consider patient-specific characteristics and design stratified, graded, and individualized training interventions to offer more targeted and clinically relevant resistance training prescriptions.

Conclusion

This systematic review and meta-analysis demonstrates that resistance training probably improves muscle strength and physical function in older adults with sarcopenia. However, improvements in grip strength, TUG, and gait speed did not reach their respective MID thresholds. Specifically, resistance training results in little to no difference in grip strength, and probably results in little to no difference in TUG and gait speed, suggesting that the clinical relevance of these improvements may be limited. Notably, training frequency and resistance type were identified as the main effect modifiers, whereas other training variables had limited influence. Based on the current evidence, we recommend implementing individualized resistance training programs within the optimal dosage range, prioritizing higher training frequencies, and selecting appropriate resistance types according to specific training goals to further enhance muscle strength and functional performance in older adults with sarcopenia.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

RX and YH participated in the conception or design, acquisition, analysis, or interpretation of the data, and drafting and revising the manuscript. RF and JX participated in the acquisition, analysis, or interpretation of the data. SJ participated in revising the manuscript and supervision. DY and JS participated in the acquisition, analysis, or interpretation of the data. All authors have read and approved the final version of the manuscript and agree with the authorship order.

Funding

This research received no external funding.

Data availability

All data generated or analyzed during this study are included in this published article (and its supplementary files).

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

Not applicable.

Human and animal rights

Not applicable.

Informed consent

Not applicable.