The effects of metformin and exercise training on cardiorespiratory, blood pressure, and metabolic adaptations across the spectrum of glucose dysregulation: a systematic review and meta-analysis

Summary

Background

Metformin and structured exercise are routinely co-prescribed across the spectrum of glucose dysregulation, under the assumption of a cardiometabolic potentiating effect. Experimental data, however, suggest that metformin might blunt some exercise-induced adaptations, but this potential interaction has not been systematically quantified.

Methods

We conducted a systematic review and meta-analysis of controlled trials in adults with conditions across the spectrum of glucose dysregulation, comparing exercise plus metformin with the same exercise programme plus placebo or no metformin, including studies published between 1 January 2000 and 12 December 2025. Randomised and non-randomised controlled trials were eligible. Outcomes included peak oxygen uptake (VO₂peak), body composition, blood pressure, glycaemic markers, and lipids. Mean differences (MDs) with 95% confidence intervals (CIs) were pooled using random-effects models, and the certainty of evidence was assessed with GRADE. Heterogeneity was evaluated through Cochran's Q test and Higgins' I² statistic, while publication bias was estimated based on the visual examination of the funnel plot, Egger's test, and selection models. This systematic review with meta-analysis was registered in PROSPERO (CRD420251167325).

Findings

Nine studies (n = 827; 14 publications) met the inclusion criteria; all contributed to the meta-analyses. Compared with exercise alone, metformin was associated with smaller improvements in VO₂peak (MD −1.19 [95% CI −2.33 to −0.04]), attenuated reductions in systolic blood pressure (3.76 [0.63–6.89]) and attenuated reductions in diastolic blood pressure (1.98 [0.42–3.55]). No significant between-group differences were observed for changes in body weight, body mass index, waist circumference, fasting glucose, glycated haemoglobin, fasting insulin, homoeostatic model assessment of insulin resistance, or lipid outcomes. Based on GRADE assessment, the certainty of evidence was moderate for VO₂peak, blood pressure, and most anthropometric measures, and low to very low for most glycaemic and lipid markers. Most outcomes showed little heterogeneity, and no evidence of publication bias was observed.

Interpretation

Across controlled trials, adding metformin to structured exercise modestly attenuated improvements in cardiorespiratory fitness and blood pressure, while glycaemic and lipid indices were unchanged. These findings suggest that co-prescribing metformin and structured exercise may attenuate exercise-induced adaptations. It may therefore be advisable to consider strategies that prioritise physical exercise and support a more individualised approach to sequencing, dosing, and monitoring when metformin is co-initiated with exercise. Future research should examine the effects of combining metformin and exercise on exercise-induced adaptations, including cardiorespiratory fitness, strength, body composition, cardiovascular parameters, and metabolic biomarkers.

Funding

None.

Research in context.

Evidence before this study

Metformin and exercise are commonly co-prescribed, based on the assumption that metformin will amplify the benefits of training. However, emerging evidence suggests that metformin may attenuate some exercise-induced adaptations. To address this research gap, this systematic review with meta-analysis aimed to clarify whether metformin influences exercise-induced physiological adaptations. On 12 December 2025, we conducted a search for studies published from 2000 onwards in the PubMed, Scopus, and Web of Science databases. The search strategy included terms related to metformin (e.g., “metformin”) and physical exercise (e.g., “exercise” or “training”). No language filters were applied. A total of nine studies (n = 827, 14 publications) met the inclusion criteria. Overall, the outcomes assessed in the meta-analysis had low to moderate certainty, with no evidence of publication bias.

Added value of this study

This systematic review with meta-analysis suggests that metformin, when combined with structured exercise, is associated with reduced gains in cardiorespiratory capacity and blood pressure, with no significant effect on body composition or glycaemic and lipid markers. Moreover, it identifies research gaps that may guide future research.

Implications of all the available evidence

Co-prescribing metformin with structured exercise does not reliably confer a potentiating effect and may modestly attenuate key adaptations, notably gains in cardiorespiratory fitness and reductions in blood pressure. It might therefore be reasonable to consider “exercise-first” strategies or to prescribe metformin based on the physical exercise training, although further high-quality evidence is needed. Clinicians should therefore avoid assuming potentiating effects and instead monitor fitness and blood pressure explicitly when metformin is co-initiated with exercise. Further randomised controlled trials are warranted to examine the effects of combining metformin and exercise on exercise-induced adaptations in specific subgroups within the spectrum of glucose dysregulation.

Introduction

The current global age-adjusted prevalence of diabetes is 6.1%, and projections estimate that by 2050, 1.31 billion people will be living with diabetes worldwide, most of whom will have type 2 diabetes.¹ In the United States, more than one-third of adults are estimated to have prediabetes.² Type 2 diabetes is characterised by impaired glucose metabolism due to insulin resistance and progressive β-cell dysfunction, and it is commonly accompanied by abnormalities in lipid metabolism.³

Metformin is widely prescribed as first-line pharmacotherapy for glycaemic management, although its mechanisms of action remain incompletely understood.⁴ Current recommendations support its use not only in people with type 2 diabetes, but also in selected individuals with prediabetes or conditions across the spectrum of glucose dysregulation.^5,6 The hypoglycaemic effects of metformin are likely mediated through multiple molecular pathways.⁷ Evidence indicates that metformin exerts significant antidiabetic effects through various mechanisms, including suppression of hepatic gluconeogenesis, increased peripheral utilisation of plasma glucose, and reduction in the rate of appearance of ingested glucose in plasma.8, 9, 10

Metformin prescription is typically accompanied by recommendations for regular physical activity and structured exercise training.^11,12 Exercise induces multiple skeletal muscle adaptations central to metabolic health, including increased glucose uptake (via increased GLUT4 translocation), mitochondrial biogenesis and improved oxidative enzyme capacity, greater fatty acid oxidation, improved insulin signalling, and reduced inflammation and oxidative stress.^13,14 Exercise also promotes vascular adaptations, including improved endothelial function and increased capillary density in skeletal muscle, which support substrate delivery and utilisation.¹³

Although exercise and metformin are frequently combined in clinical practice, their primary sites of action and the signalling pathways they engage are only partially overlapping. Metformin primarily lowers blood glucose by suppressing hepatic glucose production,^7,8 whereas exercise improves peripheral insulin sensitivity through adaptations in skeletal muscle.^13,14 Redox signalling triggered by muscle contraction is an important mediator of exercise-induced metabolic remodelling and regulation of pathways such as mTORC1.¹⁵ In contrast, metformin has been reported to reduce reactive oxygen species (ROS) generation,¹⁶ raising the possibility that it could dampen redox-sensitive signals required for optimal training adaptation.¹⁷ A further point of convergence is AMPK activation, which is a central node in metabolic regulation.¹⁸ Studies evaluating the interaction between metformin and exercise have yielded inconclusive results. Pilmark et al.¹⁹ reported no effect of metformin on exercise-induced AMPK activation in human skeletal muscle, whereas Sharoff et al.²⁰ observed a blunted AMPK activation when metformin was combined with exercise. Together, these partially overlapping and potentially competing pathways create a biological rationale for an exercise–drug interaction.

Metformin and structured exercise are routinely co-prescribed as first-line therapy across the spectrum of glucose dysregulation. This practice implicitly assumes potentiating, or at least non-interfering, benefits. However, emerging evidence indicates that the combination may involve a clinically significant exercise–drug interaction, in which metformin modifies key physiological signals that underpin both acute metabolic responses and longer-term training adaptations.²¹ Some acute studies suggest that taking metformin before aerobic exercise can diminish expected exercise-induced improvements in glycaemic control,²² although other studies have reported no interference¹⁰ or even a potentiating effect.²³ Meanwhile, intervention studies examining metformin alongside structured exercise have produced mixed results, with some reporting attenuation of exercise-induced cardiovascular effects.^24,25

Previous evidence syntheses have been limited by study design and outcome scope. A meta-analysis, mainly based on cross-sectional and parallel comparisons, found no significant differences in peak oxygen uptake (VO₂peak) between metformin users and non-users but could not directly assess metformin's effect on exercise-induced adaptations,²⁶ whereas a network meta-analysis comparing exercise, metformin, and their combination focused predominantly on glycaemic outcomes, providing little insight into broader physiological adaptations.²⁷ Clarifying whether metformin reduces improvements in cardiorespiratory fitness, body composition, and blood pressure—strong predictors of cardiovascular events and mortality—has direct implications for treatment sequencing, dosing, and individualisation, especially when metformin is initiated concurrently with an exercise program. We therefore conducted a GRADE-informed systematic review and meta-analysis of controlled trials to determine whether metformin modifies training-related adaptations beyond glycaemic outcomes.

Methods

This systematic review with meta-analysis was conducted based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.²⁸ The review methodology followed the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions.²⁹ It was registered in PROSPERO (CRD420251167325).

Search strategy

A systematic search was conducted in the PubMed, Web of Science, and Scopus databases on 12 December 2025. A search filter was established to extract publications published since 1 January 2000, to ensure minimal similarity in data analysis and reporting. No language filter was used. We used both controlled vocabulary (e.g., MeSH/Emtree where applicable) and free-text terms for metformin and exercise training, and we screened reference lists of eligible articles and relevant reviews. The full search strategy is presented in Table S1.

Selection criteria

The inclusion criteria were defined following the PICOS model (population, intervention, comparison, outcome, study type).³⁰ Thus, the following criteria were established: (a) patients over 18 years old with type 2 diabetes, prediabetes, insulin resistance, impaired glucose tolerance, metabolic syndrome, or risk factors for type 2 diabetes; (b) studies evaluating the effects of daily metformin intake combined with an exercise training intervention; (c) the comparison group participated in the same physical exercise but did not take metformin; (d) studies assessing adaptations to physical exercise; and (e) randomised controlled trials (RCTs) or non-randomised controlled trials. Reviews, meeting abstracts, and editorials were excluded.

Although the included populations differed in clinical diagnosis (type 2 diabetes, prediabetes, insulin resistance, impaired glucose tolerance, metabolic syndrome, or risk factors for type 2 diabetes), these conditions represent a continuum of glucose dysregulation and share common underlying pathophysiological mechanisms, particularly insulin resistance. Therefore, it was considered clinically appropriate to pool these populations.

Data extraction

After combining the search results from each database, duplicates were removed. Two reviewers (PE-U and MLSA) independently screened the reports by title and abstract, then examined the full texts to exclude articles that did not meet the predefined inclusion criteria. Similarly, two authors (PE-U and MLSA) independently extracted the following information for each publication: general information (e.g., author name, year, country, and study design), participants (e.g., number of individuals, number of males and females, age and glycaemic clinical status), intervention (metformin dose, as well as duration, intensity, and type of exercise intervention), measurement methods, and outcomes.

Risk of bias assessment

Two reviewers (PE-U and MLSA) independently assessed the risk of bias, and any differences were resolved by a third reviewer (MI). The risk of bias in RCTs was assessed using the Revised Cochrane risk-of-bias tool for randomised trials (RoB 2), which includes the following domains: randomisation, deviations from the intended intervention, missing outcome data, outcome measurement, and selection of the reported result. For each domain and for the overall evaluation, there are three possible answers: low risk of bias, some concerns, and high risk of bias.³¹ For non-randomised studies, we used the risk of bias in non-randomised studies of interventions (ROBINS-I) tool, which evaluates risk of bias due to confounding, participant selection, intervention classification, deviations from the intended intervention, missing outcome data, outcome measurement, and selection of the reported result. Each risk of bias was classified as low, moderate, serious, or critical.³² Because relatively few RCTs were available, we included non-randomised controlled trials when they were clinically and methodologically comparable and explored study design (randomised vs non-randomised) as a potential source of heterogeneity in subgroup and sensitivity analyses. Risk of bias and indirectness from non-randomised designs were incorporated into the GRADE judgements for each outcome.

Statistical analysis

Quantitative analyses were performed when at least three included studies reported the same endpoint. VO₂peak was designated a priori as the primary outcome, given its strong association with cardiovascular events and mortality and its centrality as an indicator of exercise training response. Secondary outcomes included anthropometric indices, blood pressure, glycaemic markers, and lipid profiles. In the meta-analysis, when multiple articles from the same study reported the same outcome, the primary analysis was included to avoid double-counting. Heterogeneity was assessed using Cochran's Q test and Higgins' I² statistic, while publication bias was estimated using visual examination of the funnel plot, Egger's test, and selection models.³³ Mean differences (MD) and 95% confidence intervals (CIs) were calculated by a random effects model (Restricted Maximum Likelihood). We also calculated 95% prediction intervals (PIs) to summarise the range of effects that might be expected in similar future populations. In cases where the result was significant, sensitivity analyses were performed using Pearson's r values of 0.2 and 0.8, and multi-level meta-analysis. Because exercise adaptation is heterogeneous and metformin pharmacodynamics vary across phenotypes, when there were at least 10 comparisons, we planned to explore a priori effect modifiers, including glycaemic status, metformin exposure history (naïve vs chronic use), metformin dose, exercise modality, and study design. The significance level for all statistical tests was set at p < 0.05. Statistical analyses were performed using RStudio 2025.05.0 and JASP 0.19.3.0.

Ethics

Informed participant consent was not required because all data came from previously published studies.

Role of the funding source

No funding was received for this work.

Results

Fig. 1 presents the PRISMA flow diagram for study selection. The search identified 5800 records; after removing duplicates, 2649 unique references remained. Title/abstract screening excluded 2611 records, leaving 38 articles for full-text review. Full-text exclusions were due to ineligible participants (n = 7), exercise interventions that did not meet inclusion criteria (n = 9), an ineligible control group (n = 4), outcomes not assessed (n = 2), or non-controlled study designs (n = 2). Ultimately, nine studies^25,34, 35, 36, 37, 38, 39, 40, 41 and five additional analyses^19,24,42, 43, 44 derived from those studies (14 publications) were included in the qualitative synthesis, and all contributed data to the meta-analysis.

Fig. 1.

Study profile. n, number of studies.

The methodological quality of RCTs was assessed using RoB 2. As shown in the traffic light plot (Figure S1) and the summary plot (Figure S2), most RCTs were judged to be at low risk of bias^24,34, 35, 36^,42,43; one study (reported across two articles) raised some concerns, primarily due to missing outcome data.^19,37 Non-randomised controlled trials were assessed using ROBINS-I: three studies were rated at moderate risk of bias,^38,39,41 whereas two studies (three articles) were rated at serious risk of bias (Figures S3 and S4).^25,40,44

Table 1 summarises the characteristics of the included populations.^19,24,25,34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44 Excluding complementary analyses, the evidence base comprised 827 participants (381 men and 446 women). Participants presented a range of metabolic phenotypes, including type 2 diabetes,^38,39 prediabetes,³⁵ metabolic syndrome,^25,41 insulin resistance,⁴⁰ impaired glucose tolerance,³⁷ and other risk factors for type 2 diabetes.^34,36

Table 1.

Qualitative analysis of included publications.

Study information	Publications	Participants	Intervention	Measurement methods	Outcomes
RCT
NCT02552355, United States, RCT	Konopka et al.,34,a 2018	N(M/F): 53(11/42) Age: 62.5 ± 1 MET/PLA: 27/26 At least one risk factor for T2DM	MET: 12-week aerobic exercise at 65–85% HRpeak (3×/wk, 45’/session) combined with metformin, 1000 mg twice daily. PLA: exercise without metformin.	OGTT, CPET, muscle biopsy, multiplexed qPCR measurements	↔ VO2peak, ↗ Vmax, ↔ body weight, ↔ BMI, ↔ fat mass, ↔ trunk fat mass, ↔ leg fat mass, ↔ FFM, ↔ HbA1c, ↔ fasting glucose, ↔ fasting insulin, ↔ HOMA-IR, ↓ whole-body insulin sensitivity (Matsuda index), ↓ oral glucose insulin sensitivity
United States, RCT	Malin et al.,35,a 2012, Malin & Braun,42 2013, Malin et al.,24 2013, Viskochil et al.,43 2017	N(M/F): 16(6/10) Age: 47.25 ± 7.34 MET/PLA: 8/8 Prediabetes	MET: 12-week aerobic (70% HRpeak) and resistance (70% 1RM) training (3×/wk, 60–75′), combined with 2000 mg/day of metformin. PLA: exercise without metformin.	CPET, 1RM tests, DXA, blood analysis	↔ VO2peak, ↔ strength, ↔ body weight, ↓ BMI, ↔ body fat percentage, ↔ central body fat percentage, ↔ waist circumference, ↔ FFM, ↔ fasting Ra, ↔ fasting Rd, ↔ fasting glucose, ↔ fasting insulin, ↔ fasting lactate, ↔ insulin sensitivity, ↔ insulin clearance, ↔ plasma glucose during exercise, ↔ lactate during exercise, ↓ plasma insulin during exercise, ↔ fasting RER, ↔ TC, ↔ LDL-c, ↔ HDL-c, ↔ triacylglycerol, ↔ SBP, ↔ DBP, ↔ mean arterial pressure, ↔ cardiac risk ratio, ↔ hs-CRP, ↔ MetS z-score, ↔ MetS prevalence, ↔ HRpeak, ↔ HR during exercise, ↔ RPE during exercise, ↔ VO2 during exercise
NCT03355469, United States, RCT	Malin et al.,36,a 2025	N(M/F): 72(35/37) Age: 55.68 ± 2.25 MET/PLA: 46/45	MET: 16-week semi-supervised aerobic exercise (two groups: 55% or 85% of VO2peak; 5×/wk) combined with 2000 mg/day of metformin. PLA: exercise without metformin.	DXA, CPET, euglycemic-hyperinsulinemic clamp, Doppler ultrasound, blood analysis	↔ VO2peak, ↔ body weight, ↔ BMI, ↔ body fat percentage, ↔ lean body mass, ↔ waist circumference, ↘ fasting glucose, ↔ fasting insulin, ↔ glucose infusion rate, ↔ triglycerides, ↔ HDL-c
NCT03316690, Denmark, RCT	Pilmark et al.,37,a 2021 Pilmark et al.,19 2022	N(M/F): 29(14/15) Age: 49.55 ± 10.47 MET/PLA: 14/15 Impaired glucose tolerance	MET: 12-week HIIT (4×/wk, 45’/session) combined with metformin, 1000 mg twice daily. PLA: exercise without metformin.	DXA, CPET, MRI scan, muscle biopsy	↔ VO2peak, ↔ maximal work (watt), ↔ body weight, ↔ lean body mass, ↔ total fat mass, ↔ visceral fat content, ↔ fasting glucose, ↔ HbA1c, ↔ fasting insulin, ↔ fasting NEFA, ↔ TC, ↔ HDL-c, ↔ LDL-c, ↔ fasting lactate, ↔ mitochondrial respiration, ↔ skeletal muscle H2O2 emission, ↔ oxidative stress, ↔ AMPK activation
Non-RCT
Portugal; Non-RCT	Baptista et al.,38,a 2018	N(M/F): 254(79/175) Age: 70.86 ± 6.17 MET/PLA: 59/195 Type 2 diabetes	MET: 24-month aerobic (50–70% HRpeak) and resistance (50–70% 1RM) training (3×/wk, 60’/session) combined with metformin, 850 mg twice daily. PLA: exercise without metformin.	Blood analysis, 6MWT	↔ body weight, ↔ waist circumference, ↔ BMI, ↔ waist-to-hip ratio, ↑ fasting glucose, ↔ HbA1c, ↔ TC, ↔ HDL-c, ↔ LDL-c, ↑ TG, ↔ SBP, ↔ DBP, ↔ 6MWD
NCT00195884, Canada; Non-RCT	Boulé et al.,39,a 2013	N(M/F): 225(146/79) Age: 54.24 ± 7.05 MET/PLA: 143/82 Type 2 diabetes	MET: 22 weeks of aerobic (60–75% HRmax), resistance (8–15 RM), or combined training combined with metformin. PLA: exercise without metformin.	CPET	↔ VO2peak, ↔ strength, ↔ HRpeak, ↔ body weight, ↔ waist circumference, ↔ fasting glucose, ↔ HbA1c
Italy; Non-RCT	Cadeddu et al.,40,a 2014, Cadeddu et al.,44 2016	N(M/F): 50(NA) Age: 45.8 ± 12 MET/PLA: 25/25 Insulin resistance	MET: 12-week aerobic exercise at 60–80% of HRR (4×/wk, 60’/session) combined with 1000 mg/day of metformin. PLA: exercise without metformin.	Echocardiography, CPET	↔ VO2peak, ↔ maximum work, ↓ anaerobic threshold, ↔ VE/VCO2, ↔VO2/WORK, ↓ BMI, ↔ HOMA, ↔ EDD, ↔ IVS, ↔ left ventricular mass, ↔ EDV, ↔ EF, ↔ IVRT, ↔ E/E′, ↑ longitudinal left ventricular S wave, ↑ GLS, ↑ GLSR
NCT01676870, Australia; Non-RCT	Ramos et al.,41,a 2020	N(M/F): 65(NA) Age: 56.83 ± 8.55 MET/PLA: 18/47 Metabolic syndrome	MET: 16-week 1HIIT (one bout at 85–95% HRpeak, 3×/wk), 4 HIIT (four bouts at 85–95% HRpeak, 3×/wk), or MICT at 60–70% HRpeak (5×/wk) combined with metformin. PLA: exercise without metformin.	CPET, blood analysis	↔ VO2peak, ↔ body weight, ↔ BMI, ↔ waist circumference, ↔ glucose, ↔ HOMA-IR, ↔ triglycerides, ↔ HDL-c, ↘ SBP, ↔ DBP, ↔ MetS z-score
NCT03019796, Spain; Non-RCT	Moreno-Cabañas et al.,25,a 2022	N (M/F): 63 (32/31) Age: 52.62 ± 7.09 MET/PLA: 34/29 Metabolic syndrome	MET: 16-week HIIT at 70–90% of HRpeak combined with metformin. PLA: exercise without metformin.	Bioelectrical impedance analysis, blood analysis, CPET	↗ VO2peak, ↗ peak aerobic power, ↔ HRpeak, ↔ body weight, ↔ BMI, ↔ fat mass, ↔ FFM, ↔ waist circumference, ↔ fasting glucose, ↔ fasting insulin, ↔ HOMA-IR, ↔ MFO, ↔ TG, ↔ HDL-c, ↘ mean arterial pressure, ↘ SBP, ↔ DBP, ↔ MetS z score, ↔ MetS factors

↔, not statistically significant between-group differences; ↑, statistically significant increase compared to control group; ↓, statistically significant reduction compared to control group; ↗, increase significantly smaller than in placebo; ↘, reduction significantly smaller than in placebo. 6MWD, 6-min walking distance; 6MWT, 6-min walking test; BMI, body mass index; CPET, cardiopulmonary exercise testing; DBP, diastolic blood pressure; DXA, dual-energy x-ray absorptiometry, F, female; FFM, fat-free mass; FPG, fasting plasma glucose; HDL-c, high-density lipoprotein cholesterol; HIIT, high-intensity interval training; HRpeak, peak heart rate; HRR, heart rate reserve; LDL-c, low-density lipoprotein cholesterol; M, male; MET, metformin group; MetS, metabolic syndrome; MFO, maximal fat oxidation; MICT, moderate intensity continuous training; N, number of patients; NA, not available; OGTT, oral glucose tolerance test; PLA, placebo group; Ra, rate of glucose appearance; RCT, randomised controlled trial; Rd, rate of glucose disappearance; RER, respiratory exchange ratio; RM, repetition maximum; RPE, rate of perceived exertion; SBP, systolic blood pressure; TC, total cholesterol; TG, triglycerides; VO₂peak, peak oxygen uptake; x/wk, times per week.

^aPrimary analysis.

All interventions included supervised exercise sessions. The most common intervention duration was 12 weeks,^34,35,37,40 although several trials lasted 16 weeks or longer.^{25,36,38,39,41} Exercise modalities included aerobic training,^34,36,39, 40, 41 resistance training,³⁹ combined aerobic and resistance training,^35,38,39 and HIIT training.^25,37,41 Across studies, both groups undertook the same exercise program, whereas metformin exposure was the primary difference between groups. Most trials prescribed therapeutic metformin doses of 1000–2000 mg/day.34, 35, 36, 37^,39,40

Five studies evaluated cardiorespiratory adaptations to exercise using relative VO₂peak (mL·kg⁻¹·min⁻¹).^{25,35,37,39,41} Although visual inspection of the funnel plot suggested some asymmetry (Figure S5), there was no evidence of publication bias based on the Egger's test (p = 0.674) and selection models (p = 0.918). The pooled analysis showed that metformin was associated with a significantly smaller improvement in VO₂peak compared with exercise alone (MD = −1.186; 95% CI −2.327 to −0.044; I² = 0.00%; p = 0.045; Fig. 2). This finding was partially confirmed by sensitivity analyses (r = 0.2; MD = −1.353; 95% CI −2.522 to −0.184; I² = 0.00%; p = 0.032|r = 0.8; MD = −0.938; 95% CI −1.934 to 0.057; I² = 0.00%; p = 0.059). When applying a multilevel model to adjust for dependence between subgroups within the same study, the effect remained in the same direction, although it did not reach statistical significance (MD = −1.180; 95% CI −3.813 to 1.454; I² = 0.00%; p = 0.170; Figure S6). Both in the main model and in the sensitivity analyses, a consistent difference greater than 1 mL/kg/min was observed, which has been associated with lower mortality and a better glycaemic profile, suggesting a clinically relevant biological impact.^45,46 Of the five studies included, three^25,35,39 reported that metformin reduced the effect of training by 40–60%, corresponding to a lower improvement of approximately 1–2 mL/kg/min.

Fig. 2.

Forest plot showing the impact of metformin on exercise-induced VO₂peak changes. CI, confidence interval; MD, mean difference; PI, prediction interval.

Data on body weight were available from eight trials.^25,34, 35, 36, 37, 38, 39^,41 No evidence of publication bias was observed, as indicated by visual inspection of the funnel plot (Figure S7), Egger's test (p = 0.684) and selection models (p = 0.793). The pooled analysis showed no between-group difference in body weight change (MD = −0.059; 95% CI −1.260 to 1.142; I² = 0.00%; p = 0.911; Figure S8). Subgroup analyses showed no evidence of effect modification by exercise modality (resistance, moderate-intensity continuous training, high-intensity interval training, or combined; F = 1.435; p = 0.603; Figure S9).

Seven studies reported outcomes for body mass index (BMI).^25,34,36,38,40, 41, 42 Although the funnel plot (Figure S10) and Egger's test (p = 0.011) showed significant asymmetry, more robust selection models did not support the presence of publication bias (0.794). There was no significant between-group difference in BMI change (MD = −0.268; 95% CI −1.020 to 0.485; I² = 0.00%; p = 0.342; Figure S11).

Five trials provided data on waist circumference.^{25,36,38,39,41} Despite the funnel plot (Figure S12) and Egger's test (p = 0.025) suggesting significant asymmetry, more robust selection models showed no evidence of publication bias (p = 0.792). Metformin was not associated with a significant change in waist circumference (MD = 0.903; 95% CI −0.301 to 2.107; I² = 0.00%; p = 0.106; Figure S13).

Three studies evaluated fat-free mass.^25,34,35 Visual inspection of the funnel plot (Figure S14) and the Egger's test (p = 0.356) found no significant evidence of publication bias, which was confirmed by selection models (p = 0.845). The pooled effect size was not significant (MD = −1.384; 95% CI −5.525 to 2.758; I² = 0.00%; p = 0.287; Figure S15).

Fasting glucose was reported in eight studies.^25,34, 35, 36, 37, 38, 39^,41 Despite the asymmetry observed in the funnel plot (Figure S16), publication bias was not detected using the Egger's test (p = 0.653) and selection models (p = 0.171). Overall, there was no significant between-group difference in fasting glucose change between exercise plus metformin and exercise alone (MD = 1.648; 95% CI −6.934 to 10.230; I² = 75.93%; p = 0.663; Figure S17). Subgroup analyses found no significant modification of the effect by exercise modality (F = 0.366; p = 0.792; Figure S18).

Four studies evaluated whether metformin modified exercise-induced changes in glycated haemoglobin (HbA1c).^34,37, 38, 39 No evidence of publication bias was observed based on the funnel plot (Figure S19), Egger's test (0.760) and selection models (p = 0.891). The pooled effect showed no significant difference between groups (MD = 0.104; 95% CI −0.225 to 0.432; I² = 69.21%; p = 0.388; Figure S20).

Fasting insulin was available from five studies.^25,34, 35, 36, 37 Despite the slight asymmetry observed in the funnel plot (Figure S21), the Egger's test (p = 0.957) and selection models (p = 0.826) reported absence of publication bias. There was no significant effect of metformin on exercise-induced changes in fasting insulin (MD = 0.688; 95% CI −0.720 to 2.095; I² = 0.00%; p = 0.246; Figure S22).

Similarly, four studies contributed data for homoeostatic model assessment of insulin resistance (HOMA-IR).^25,34,40,41 Visual inspection of the funnel plot (Figure S23), along with the results of the Egger's test (p = 0.487) and selection models (p = 0.930), indicated no significant publication bias. The pooled estimate indicated no significant between-group difference (MD = 0.036; 95% CI −0.144 to 0.216; I² = 0.00%; p = 0.569; Figure S24).

Three studies reported total cholesterol outcomes.^24,37,38 Although the funnel plot showed some asymmetry (Figure S25), the Egger's test (p = 0.918) and selection models (p = 0.601) found no publication bias. Metformin co-administration was not associated with a significant between-group difference in total cholesterol change compared with exercise alone (MD = 8.407; 95% CI −12.044 to 28.858; I² = 0.00%; p = 0.219; Figure S26).

Six trials evaluated high-density lipoprotein cholesterol (HDL-c).^24,25,36, 37, 38^,41 Visual assessment of the funnel plot (Figure S27) and the Egger's test (p = 0.400) suggested no significant evidence of publication bias, which was confirmed by selection models (p = 0.365). The pooled estimate showed no significant effect of metformin on exercise-induced changes in HDL-c (MD = 0.320; 95% CI −1.384 to 2.024; I² = 0.00%; p = 0.649; Figure S28).

Low-density lipoprotein cholesterol (LDL-c) was reported in three studies.^24,37,38 The Egger's test (p = 0.916) and selection models (p = 0.885) indicated no publication bias, with the funnel plot showing substantial asymmetry (Figure S29). Metformin was not associated with a significant difference in LDL-c change (MD = 4.067; 95% CI −15.887 to 24.020; I² = 0.00%; p = 0.473; Figure S30).

Four studies also reported triglyceride outcomes.^25,36,38,41 Although the visual examination of the funnel plot suggested some asymmetry (Figure S31), the Egger's test (p = 0.317) and selection models (p = 0.899) found no significant evidence of publication bias. The pooled effect did not reach statistical significance (MD = 9.470; 95% CI −15.529 to 34.470; I² = 36.39%; p = 0.314; Figure S32).

Five trials reported systolic blood pressure (SBP).^{24,25,36,38,41} Despite the substantial asymmetry observed in the funnel plot (Figure S33), the Egger's test (p = 0.286) and selection models (p = 0.565) reported no indication of publication bias. The pooled estimate suggested a significant higher SBP in the metformin-plus-exercise group compared with exercise plus placebo (MD = 3.761; 95% CI 0.630–6.893; I² = 0.00%; p = 0.029; Fig. 3). This effect remained statistically significant in the sensitivity analyses (r = 0.2; MD = 3.728; 95% CI 0.590–6.866; I² = 0.00%; p = 0.030|r = 0.8; MD = 3.890; 95% CI 0.769–7.011; I² = 4.64%; p = 0.026). Furthermore, the multilevel meta-analysis yielded consistent findings (MD = 4.220; 95% CI 0.261–8.179; p = 0.042; Figure S34). A reduction of approximately 5 mmHg in SBP is associated with a ∼10% lower risk of major cardiovascular events, suggesting that the ∼4 mmHg reduction observed in the pooled analysis could still confer a clinically relevant benefit.⁴⁷

Fig. 3.

Forest plot showing the impact of metformin on exercise-induced SBP changes. CI, confidence interval; MD, mean difference; PI, prediction interval.

Five studies evaluated diastolic blood pressure (DBP).^{24,25,36,38,41} Although the visual examination of the funnel plot suggested publication bias (Figure S35), the Egger's test (p = 0.386) and selection models (p = 0.414) revealed no significant evidence of publication bias. Similar to SBP, metformin was associated with a significantly attenuated exercise-induced reduction in DBP compared with placebo (MD = 1.983; 95% CI 0.416–3.549; I² = 0.00%; p = 0.025; Figure S36). Sensitivity analyses yielded consistent results (r = 0.2; MD = 1.980; 95% CI 0.415–3.545; I² = 0.00%; p = 0.025|r = 0.8; MD = 1.992; 95% CI 0.421–3.564; I² = 0.00%; p = 0.024). Moreover, the multilevel meta-analysis corroborated these findings (MD = 1.919; 95% CI 0.269–3.570; p = 0.033; Figure S37). Of the four studies that evaluated DBP, two^38,41 observed a 25% attenuation of the training effect (equivalent to approximately 2 mm Hg), while the other two^24,25 reported a 50% attenuation (equivalent to approximately 4 mm Hg).

Based on GRADE assessment, the certainty of evidence was moderate for VO₂peak, body weight, BMI, waist circumference, fasting glucose, HDL-c, SBP, and DBP. Certainty was low for fat-free mass, fasting insulin, HOMA-IR, total cholesterol, LDL-c, and triglycerides. Evidence for HbA1c was rated very low (Table S1).

Table 2 summarises the effects of metformin on exercise-induced adaptations, indicating the direction of the effect, the pooled effect, the number of studies, the level of certainty, and the clinical interpretation for each outcome assessed. Adherence to the interventions and adverse events are reported in Table S2.

Table 2.

Evidence-to-practice summary of metformin effects on exercise-induced adaptations.

Outcome (priority)	Direction of effect on exercise responsea	Pooled effect (MD, 95% CI)	Studies (n)	Certainty (GRADE)b	Clinical interpretation
VO2peak (cardiorespiratory fitness)	↓ Smaller improvement	−1.19 (−2.33 to −0.04)	5	Moderate	Metformin co-initiation may blunt fitness gains from training. If fitness is a key target, consider monitoring VO2peak/submax tests and evaluating sequencing/timing/dose in future trials/clinical pathways.
Body weight	↔ No clear difference	−0.06 (−1.26 to 1.14)	8	Moderate	Do not expect potentiating weight loss from adding metformin to training; prioritise exercise dose + diet and monitor weight trajectory.
BMI	↔ No clear difference	−0.27 (−1.02 to 0.49)	7	Moderate	BMI may be insensitive to compartment changes; interpret alongside waist and/or body composition measures.
Waist circumference	↔ No clear difference	0.90 (−0.30 to 2.11)	5	Moderate	No further improvement in waist circumference was observed with metformin; prioritise exercise + an appropriate diet. Ensure that diet is measured in future studies.
Fat-free mass	↔ No clear difference	−1.38 (−5.53 to 2.76)	3	Low	Evidence is limited/low certainty; future trials should standardise resistance training dose and use robust methods (e.g., DXA) to assess lean mass.
Fasting glucose	↔ No clear difference (heterogeneous)	1.65 (−6.93 to 10.23)	8	Moderate	No consistent potentiating (or harmful) effect on fasting glucose overall; phenotype (naïve vs chronic metformin, baseline glycaemia) may matter.
HbA1c	↔ No clear difference (heterogeneous)	0.10 (−0.23 to 0.43)	4	Very low	Current evidence is too uncertain for firm conclusions; HbA1c may require longer duration and/or higher baseline HbA1c to detect interaction effects.
Fasting insulin	↔ No clear difference	0.69 (−0.72 to 2.10)	5	Low	Fasting indices may miss interactions; future work should prioritise dynamic insulin sensitivity outcomes (OGTT/clamp/CGM-derived metrics when feasible).
HOMA-IR	↔ No clear difference	0.04 (−0.14 to 0.22)	4	Low	No consistent interaction detected on HOMA-IR; interpret cautiously due to low certainty and limitations of fasting proxies.
Total cholesterol	↔ No clear difference	8.41 (−12.04 to 28.86)	3	Low	Do not expect potentiating lipid improvements; manage lipids per standard care and focus on global risk reduction.
HDL-c	↔ No clear difference	0.32 (−1.38 to 2.02)	6	Moderate	No consistent potentiating effect of metformin on exercise-induced HDL changes.
LDL-c	↔ No clear difference	4.07 (−15.89 to 24.02)	3	Low	No consistent potentiating effect; low certainty.
Triglycerides	↔ No clear difference	9.47 (−15.53 to 34.47)	4	Low	No statistically significant between-group difference; interpret cautiously due to limited number of trials and imprecision.
SBP	↓ Attenuated reduction	3.76 (0.63–6.89)	5	Moderate	Suggests metformin may blunt BP-lowering adaptations (SBP). If BP lowering is a key target, consider closer BP monitoring and further trials on timing/sequence.
DBP	↓ Attenuated reduction	1.98 (0.42–3.55)	5	Moderate	Suggests metformin may blunt BP-lowering adaptations (DBP). If BP lowering is a key target, consider closer BP monitoring and further trials on timing/sequence.

(Effect metric: MD; positive values generally indicate a less favourable training response for outcomes expected to decrease, and negative values a smaller favourable response for outcomes expected to increase, depending on coding of change scores.).

Abbreviations: BMI, body mass index; BP, blood pressure; DBP, diastolic blood pressure; HbA1c, glycated haemoglobin; HDL-c, high-density lipoprotein cholesterol; HOMA-IR, homeostasis model assessment of insulin resistance; LDL-c, low-density lipoprotein cholesterol; SBP, systolic blood pressure; MD, mean difference; VO₂peak, peak oxygen uptake.

^aDirection of effect refers to whether metformin modifies the exercise-induced change compared with exercise alone/placebo: ↓ indicates smaller improvement (for beneficial increases, e.g., VO₂peak) or attenuated reduction (for beneficial decreases, e.g., blood pressure); ↔ indicates no clear difference.

^bGRADE certainty rated as High/Moderate/Low/Very low considering risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Across nine studies contributing to the meta-analyses, metformin co-administration consistently attenuated training-induced improvements in VO₂peak, SBP, and DBP, with effect sizes corresponding to ∼1 mL kg⁻¹·min⁻¹ less gain in VO₂peak, ∼4 mmHg less reduction in SBP, and ∼2 mmHg less reduction in DBP. In contrast, pooled effects on body weight, BMI, waist circumference, fasting glycaemia, HbA1c, fasting insulin, HOMA-IR, and lipids were small and imprecise, with confidence intervals spanning no effect. GRADE ratings were generally moderate for VO₂peak, anthropometry, and blood pressure, and low to very low for most metabolic and lipid outcomes.

Discussion

In this systematic review and meta-analysis of adults with conditions across the spectrum of glucose dysregulation, metformin co-administered with structured exercise was associated with smaller training-induced improvements in VO₂peak and attenuated reductions in SBP and DBP compared with exercise alone. The magnitude of these effects—approximately 1 mL kg⁻¹·min⁻¹ less gain in VO₂peak, 4 mmHg less reduction in SBP, and 2 mmHg less reduction in DBP—corresponds to a 40–60% attenuation of the training response in some studies. By contrast, we found no clear evidence that metformin meaningfully modified exercise-induced changes in body weight, BMI, waist circumference, fat-free mass, fasting glycaemia, HbA1c, fasting insulin, HOMA-IR, or lipid profiles over the 10–24 week interventions studied. Taken together, these findings suggest that combining two established therapies does not guarantee potentiating benefit across all cardiometabolic domains and may selectively blunt some of the most prognostically relevant exercise adaptations.

A pharmaco–physiological interaction between metformin and exercise is clinically important because both interventions are cornerstone of type 2 diabetes prevention and treatment.⁴⁸ In the trials included, metformin attenuated improvements in cardiorespiratory capacity by approximately 40%, a magnitude that may be clinically meaningful given the robust associations between cardiorespiratory fitness, glycaemic control and long-term mortality.^45,46 For individuals with type 2 diabetes—who already face elevated cardiovascular risk—smaller training-related gains in VO₂peak and blood pressure may translate into less favourable overall risk modification. In practice, this could mean that patients taking metformin need to exercise at a higher relative intensity to achieve the same absolute workload, potentially increasing perceived exertion and undermining long-term adherence, with downstream consequences for cardiometabolic health.^12,26,42,49

The heterogeneity in the responses to metformin observed across studies may reflect interindividual differences in baseline phenotypes and metformin exposure history. Inclusion of participants spanning prediabetes to established type 2 diabetes likely contributed to variability, and Konopka et al.³⁴ reported particularly high variability in exercise responses in the metformin group. Retrospective analyses have suggested that participants exhibiting attenuated training adaptations while receiving metformin had higher baseline mitochondrial Complex I respiration and insulin sensitivity, and lower HOMA-IR,⁵⁰ raising the possibility that metformin interference may be more pronounced in those with a relatively favourable metabolic profile. However, this hypothesis was not consistently supported in the present meta-analysis: studies enrolling participants with type 2 diabetes showed comparable or even a more pronounced attenuation in HbA1c than studies in prediabetes.^37,38 Previous metformin exposure may be another determinant. It has been proposed that metformin-naïve individuals could experience greater interference with exercise adaptation due to lack of habituation,²⁶ which might explain why metformin-naïve participants showed a significantly attenuated reduction in fasting glucose in one trial,³⁷ whereas this effect was not observed in chronic metformin users.^25,39 Nevertheless, lack of synergy was not limited to metformin-naïve individuals, as Moreno-Cabañas et al.,²⁵ reported significantly blunted VO₂peak improvements among metformin-habituated participants, underscoring that habituation does not necessarily mitigate the interaction across all outcomes.

Differences across studies may also be driven by the specific metabolic endpoints selected and the methods used to assess them. In Konopka et al.³⁴ whole-body insulin sensitivity derived from an oral glucose tolerance test showed a blunted response with metformin, whereas other glucose regulation markers (HbA1c, fasting insulin, and HOMA-IR) did not demonstrate significant interactions. This pattern aligns with Malin et al.,³⁵ in which insulin sensitivity, assessed by the euglycemic-hyperinsulinemic clamp tended to be attenuated by metformin without significant changes in fasting insulin or HOMA-IR. Furthermore, Pilmark et al.³⁷ observed a blunting effect of metformin on exercise-induced improvements in postprandial glucose, as assessed using the mixed meal tolerance test (MMTT). Additionally, differences in insulin-stimulated carbohydrate oxidation observed in clamp studies⁵¹ suggest that consideration should also be given to analyse postprandial insulin to understand the overall metabolic response. Together, these findings suggest that potential effects may be more readily detected with dynamic or gold-standard assessments, including oral glucose tolerance test (OGTT), MMTT, continuous glucose monitoring, and postprandial insulin, than with fasting indices.^34,35 Accordingly, it remains important to determine whether metformin modifies insulin sensitivity across insulin concentrations that are more representative of daily life and typical clinical conditions.

Intervention characteristics likely contribute further to heterogeneity. Ramos et al.⁴¹ compared HIIT and continuous aerobic training and observed that metformin attenuated the positive impact of high-volume HIIT on metabolic syndrome severity to a greater extent than continuous training. Other reports have suggested that lack of synergy may be primarily limited to aerobic exercise and that resistance training combined with metformin could be superior to resistance exercise alone.²⁷ However, in the present meta-analysis, studies implementing resistance training or combined training demonstrated a similar or even greater impact of metformin than aerobic interventions.^35,38,39 Moreover, Walton et al.⁵² reported an attenuated muscle hypertrophy response to progressive resistance exercise in healthy older adults, with a trend suggesting a similar effect on muscle strength and power. These observations highlight the need for trials explicitly designed to test interaction effects by exercise modality and intensity, rather than inferring them from underpowered subgroup comparisons. Finally, no differences were detected based on metformin dosage, with all studies reporting doses between 1000 and 2000 mg/day. Although these doses are commonly considered clinically effective for most patients, the optimal dose in physically active individuals might differ from sedentary individuals, warranting focused study.

Study design and unmeasured lifestyle factors may also have influenced outcomes. Although the insufficient number of studies prevented subgroup analyses according to study design, the forest plots did not suggest differences depending on whether the study was an RCT or non-randomised trial. None of the included studies comprehensively recorded dietary intake, and dietary patterns could plausibly modify both glycaemic endpoints and training adaptations. For example, Baptista et al.³⁸ reported a decrease in blood glucose in the exercise-only group but an increase in the exercise-plus-metformin group; the authors suggested that expectations of pharmacological glycaemic control might lead to compensatory lifestyle behaviours. Although speculative, this underscores the importance of concurrently assessing diet and other behaviours to interpret interaction effects between exercise and pharmacotherapy.

A plausible integrative model is that metformin modifies training adaptation through converging effects on (i) mitochondrial signalling and Complex I flux, (ii) redox–sensitive pathways that normally translate contractile stress into transcriptional remodelling, and (iii) vascular responses that influence muscle perfusion and oxygen delivery.^15,18,34,42 Within this framework, the attenuation of VO₂peak and blood pressure adaptations may share a common upstream constraint—reduced haemodynamic stimulus and/or altered mitochondrial–redox signalling—whereas fasting glycaemic markers remain largely unchanged over 10–16 weeks because their responsiveness is dominated by baseline glycaemia, dietary intake, and the limited sensitivity of fasting indices to detect changes in insulin sensitivity.^{19,25,34,35,37} Figure S38 presents a conceptual model of the speculative interaction between metformin and exercise with respect to training adaptations.

The present meta-analysis also found that metformin significantly attenuated exercise-induced reductions in blood pressure. Mechanistically, it has been suggested that metformin may alter AMPK signalling, thereby reducing eNOS activation and nitric oxide (NO) production, thereby limiting vasodilation and blood pressure lowering.^24,25 Malin et al.²⁴ further proposed that metformin may attenuate training-related reductions in vascular inflammation (hs-CRP) and improvements in exercise-induced endothelial reactivity.²⁴ Reduced muscle perfusion, via lower NO bioavailability or interference with exercise-dependent vasodilation, could decrease the haemodynamic stimulus for adaptation, thereby limiting oxygen delivery and contributing to smaller improvement in VO₂peak.^25,36

These findings should not be interpreted as evidence against the use of metformin when clinically indicated for glycaemia. Rather, they challenge the assumption that co-initiation of metformin and structured exercise will uniformly produce potentiating effects across fitness, blood pressure, and glycaemic domains. Thus, it may be reasonable to consider “exercise-first” or “exercise-intensified” strategies, or to vary the timing and dose of metformin relative to training sessions,⁵³ while awaiting high-quality evidence from adequately powered interaction trials. At a minimum, clinicians should recognise that the combination of metformin and exercise may yield heterogeneous responses and monitor fitness and blood pressure rather than assuming maximal benefit from co-prescription.

This systematic review with meta-analysis has several limitations. First, it included RCTs and non-randomised controlled trials, thus some studies presented serious risks of bias. Second, considerable heterogeneity existed across study populations and exercise interventions. Specifically, participants in the various studies exhibited a range of metabolic phenotypes, varying in the degree of insulin resistance and beta-cell dysfunction, making it difficult to draw condition-specific conclusions. Additionally, most studies enrolled relatively small samples, limiting statistical power. Although the outcomes analysed in this meta-analysis are relevant to clinical practice and public health, it would be interesting to include other outcomes, such as mortality and quality of life. Moreover, most trials were relatively short (10–24 weeks) and enrolled participants willing to undergo supervised training, which may limit generalisability to routine primary care settings and to longer-term real-world adherence. Finally, the absence of dietary assessment and control represents a potential confounder, particularly for outcomes closely tied to energy balance and macronutrient intake.

Despite these limitations, the review has several strengths. To our knowledge, this is the first meta-analysis examining the effects of metformin prescription on adaptations to exercise training. Furthermore, it provides a comprehensive approach by considering both pharmacological and exercise prescriptions, offering valuable insights for both research and clinical practice. By synthesising controlled trial data across a broad set of physiological outcomes, this work addresses a key evidence gap regarding whether metformin modifies exercise training adaptations and provides a framework for designing future mechanistic and clinical trials.

Overall, the present evidence indicates that combining two beneficial treatments may yield complex effects that cannot be inferred from their independent efficacy. Although certainty of evidence was low or moderate for most outcomes, the findings of the current review underscore the importance of systematically evaluating exercise–drug interactions to optimise public health recommendations and individualised care pathways.

Future research should focus on patients with type 2 diabetes mellitus in the community, examining interventions that combine metformin and exercise compared with exercise interventions. Outcomes of interest could include exercise-induced adaptations, such as cardiorespiratory fitness, strength, body composition, cardiovascular parameters, and metabolic biomarkers. Such research could guide future meta-analyses by standardising the PICOTS elements (Participants, Intervention, Comparison, Outcome, Time, Setting), thereby enabling a more robust synthesis of results.

In conclusion, metformin co-administered with structured exercise was associated with smaller improvements in VO₂peak and attenuated reductions in blood pressure, with moderate certainty of evidence for both outcomes. Furthermore, no clear potentiating effects were observed for body composition, fasting glycaemic indices, or lipid profiles over the intervention periods studied. These findings support moving beyond a “one-size-fits-all” co-initiation strategy toward trials and clinical pathways that define optimal sequencing, dosing/timing, and exercise modalities in the individuals most likely to benefit. More broadly, they highlight the need to systematically evaluate exercise–drug interactions rather than assume that effective pharmacotherapies and exercise training will always combine in a simple, potentiating manner.

Contributors

PEU, MLSA, and MI conceptualised the study. PEU and MLSA wrote the original draft, performed the systematic search and extracted the data. PEU performed the formal analysis. MI supervised the study and edited the final version. All authors contributed to the writing and revision of the manuscript, had full access to all the data in the study and were responsible for the decision to submit for publication. MI is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity and accuracy of the data analysis.

All authors accessed and verified the underlying data, as well as read and approved the final version of the manuscript.

Data sharing statement

This study used data drawn from previously published studies. All tables and figures generated are included in the manuscript and supplementary materials.

Declaration of interests

We declare no competing interests.