Introduction
Over the past decade, chronic conditions such as type 2 diabetes mellitus (T2DM) have received increasing attention because of their substantial and growing global impact [1, 2]. The prevalence of T2DM has quadrupled over the past three decades [3]. Current estimates from the International Diabetes Federation indicate that 589 million people are living with diabetes, the vast majority of whom have T2DM, with this figure projected to increase to 853 million by 2050 [4].
Characterized by insulin resistance, impaired insulin secretion, and chronic hyperglycemia [5, 6], T2DM substantially increases the risk of cardiovascular disease, which remains the leading cause of death in this population [7, 8]. In high‐income countries, disease prevalence peaks in older populations [9], paralleling demographic shifts toward increased longevity. These trends underscore the urgent need for accessible, evidence‐based strategies that complement pharmacotherapy and reduce the growing cardiometabolic burden.
Lifestyle modification, including exercise, remains the cornerstone of T2DM standard care. Both endurance and resistance exercise reduce visceral adiposity, improve glycemic control, and attenuate global cardiovascular risk [5, 10]. Current guidelines recommend combining these modalities to maximize their benefits, yet adherence to conventional exercise programs is often poor because they are time‐consuming, monotonous, or require specialized facilities.
High‐intensity interval training (HIIT) is another exercise option; a meta‐analysis of randomized trials in adults with T2DM found reductions in HbA1c following HIIT [11], while comprehensive lifestyle programs that integrate exercise with targeted nutrition can more than double the likelihood of partial remission compared with standard support [12, 13, 14, 15]. Notably, a meta‐analysis of 81 RCTs suggested that programs combining endurance, high‐intensity and resistance training may improve cardiorespiratory fitness and cardiometabolic outcomes more than individual training modalities [16]. In this context, small‐sided soccer offers a feasible multicomponent training option, combining continuous aerobic work, repeated high‐intensity bouts, and functional strength movements into an enjoyable and socially engaging format [17, 18].
Randomized trials of recreational football performed twice weekly for 12–16 weeks have reported improvements in fasting glucose, body composition, and cardiorespiratory fitness in adults with prediabetes or T2DM [19, 20]. Yet, these exercise programs are almost always undertaken alongside glucose‐lowering medication. In fact, although lifestyle interventions constitute the foundation of T2DM care, pharmacological treatment is often necessary to protect target organs and reduce long‐term complications [21].
Consequently, most people living with T2DM are treated with agents such as metformin, GLP‐1RA, SGLT2 inhibitors, or insulin to achieve glycemic targets and facilitate weight loss. Emerging data suggest that a considerable fraction of weight reduction induced by pharmacotherapy involves loss of lean body mass rather than fat mass [22]. Because skeletal muscle is the principal reservoir for insulin‐mediated glucose disposal, any erosion of muscle mass could blunt the metabolic dividends of exercise [23].
Whether GLP‐1RA attenuates, preserves, or enhances adaptations in body composition and cardiorespiratory fitness to multicomponent training remains unclear and warrants further investigation.
To address this evidence gap, we conducted a 14‐week randomized controlled trial testing whether small‐sided soccer training three times per week would (i) enhance body composition, metabolic profile, and cardiorespiratory fitness in adults with T2DM and (ii) yield comparable or greater benefits in participants receiving concomitant GLP‐1RA therapy. Clarifying the interaction between multicomponent exercise and modern antidiabetic pharmacotherapy could inform integrated, individualized treatment strategies for the burgeoning T2DM population.
We hypothesized that 14 weeks of small‐sided soccer training, comprising three 60‐min sessions per week, would improve cardiorespiratory fitness and body composition in adults with T2DM compared with usual care. Secondary analyses examined lipid and glycemic outcomes. Exploratory analyses assessed whether baseline GLP‐1RA use modified responses to training.
Materials and Methods
Study Design and Participants
This randomized controlled trial enrolled men and women aged 40–70 years with T2DM diagnosed within the preceding 10 years. The study encompassed the entire Faroe Islands, as delineated by the Steno Diabetes Center Faroe Islands, with invitations to participate distributed through Vangin, the national digital health communication platform. Exclusion criteria were severe microvascular or macrovascular complications and pregnancy.
Use of exogenous insulin or insulin secretagogues was also an exclusion criterion. Eligible participants provided written informed consent and were randomly allocated in a 60:40 ratio to either an exercise group (EX) that performed a multicomponent soccer‐based exercise intervention or a usual‐care control group (CON). The unequal allocation reflected an anticipated higher dropout rate in EX.
Participants and coaches were aware of allocation; clinical investigators conducting outcome assessments were blinded. All randomized participants were invited to post‐intervention assessments regardless of exercise attendance. The trial was registered before the intervention began in September 2024 (first published 26 June 2024).
Randomization
Stratified randomization was employed to ensure balance in three key variables: sex (male/female), age (≥ 55 years vs. < 55 years), and GLP‐1RA use (yes/no). Within each stratum, participants were randomly assigned to EX or CON in a 60:40 ratio using a fixed random seed (random_state = 42) to guarantee reproducibility. The allocation sequence was generated in Python 3.12.4 using scikit‐learn's train_test_split.
Clinical Assessments
Body Composition
Total body fat and lean body mass were measured via dual‐energy X‐ray absorptiometry (DXA) (Norland XR‐800, Norland Corporation). Participants wore light clothing and removed any metallic items before the scan. DXA scans were performed at baseline and after 14 weeks.
All analyses were performed using Illuminatus DXA software (Norland Corporation, Oslo, Norway).
Blood Samples
Blood samples were obtained after an overnight fast; participants took their morning medications as usual. Venous blood was drawn with a 21‐gauge needle and minimal stasis into appropriate BD Vacutainer tubes (Plymouth, United Kingdom). Samples requiring centrifugation were centrifuged at 2000 g for 10 min.
Samples for C‐peptide analysis were frozen and analyzed within 1 week.
According to the routine laboratory method documentation, hemoglobin, leukocyte count, and platelet count were analyzed on a Sysmex XN using Sysmex Cellpack and the corresponding Sulfolyser, Lysercell, and Fluorocell reagents. HbA1c was analyzed on a Tosoh G8. Glucose, total cholesterol, HDL cholesterol, triglycerides, and creatinine were analyzed on a Cobas pro c503 using Roche GLUC3, CHOL2, HDLC4, TRIGL, and CREP2 reagents, respectively.
C‐peptide, vitamin B12, and thyroid‐stimulating hormone were analyzed on a Cobas pro e801 using Roche C‐Peptide, Vitamin B12 II, and TSH assays, respectively. The documented method for 25‐hydroxyvitamin D was the LIAISON 25 OH Vitamin D TOTAL assay on a DiaSorin LIAISON XL. LDL cholesterol and estimated glomerular filtration rate (eGFR) were calculated rather than measured by separate assays.
The laboratory's maximum allowable quality‐control coefficients of variation (CVmax) were 1.5% for hemoglobin; 2.8% for leukocytes and HbA1c; 3%–12%, depending on control level, for platelets; 4% for glucose, total cholesterol, HDL cholesterol, thyroid‐stimulating hormone, and creatinine; 5% for triglycerides; 6% for C‐peptide; 7% for vitamin B12; and 10% for 25‐hydroxyvitamin D. These are quality‐control limits, not observed intra‐assay or inter‐assay CVs.
HOMA2‐%B, HOMA2‐%S, and HOMA2‐IR were calculated using the HOMA2 Calculator v2.2.3 (Diabetes Trials Unit, University of Oxford) from fasting plasma glucose (mmol/L) and fasting C‐peptide (nmol/L). All input pairs used for these calculations were within the calculator's valid ranges (glucose 3.0–25.0 mmol/L; observed 4.5–11.9 mmol/L; C‐peptide 0.2–3.5 nmol/L; observed 0.21–2.84 nmol/L); no values were truncated or excluded for being outside those ranges. Participants using exogenous insulin or insulin secretagogues were excluded from the trial.
Baseline GLP‐1RA use was identified from medication records or baseline self‐report. Concomitant glucose‐lowering medications were identified from medication records.
Blood Pressure
Systolic and diastolic blood pressure (BP) and resting heart rate (RHR) were measured in a seated position according to the scientific statement from the American Heart Association [24]. After resting for 5 min, an initial BP measurement was obtained in both arms, followed by two additional measurements in the arm with the highest initial reading, with at least 1 min between measurements. The average of the final two readings was recorded as the test results for BP and RHR.
Maximal Aerobic Capacity
Participants were instructed to abstain from vigorous physical activity for 24 h and to avoid alcohol, tobacco, and caffeine on the testing day. All participants arrived following a 4‐h fast. Peak oxygen uptake was assessed using an incremental cycling test performed on an electronically braked cycle ergometer (Excalibur Sport, Lode, Groningen, Netherlands), with breath‐by‐breath gas exchange measured using an online metabolic system (Cosmed Quark b2, Milan, Italy).
The protocol began with a 4‐min warm‐up at 30 W for females and 50 W for males, followed by a continuous ramp increase in workload every 30 s (7.5 W for females and 10 W for males) until volitional exhaustion. Heart rate was monitored continuously throughout the test (HRM‐Dual, Garmin, Olathe, KS). VO2peak was defined as the highest 30‐s average oxygen uptake achieved during the test.
Yo–Yo Intermittent Endurance Level 1 Test
Participants completed the Yo–Yo Intermittent Endurance Level 1 Test (Yo–Yo IE1) indoors at 18°C–20°C. Testing was preceded by a standardized warm‐up. All participants underwent a familiarization run before baseline testing to ensure proper understanding of the procedure.
Exercise Training Intervention: Small‐Sided Soccer
Participants in EX performed small‐sided soccer training three times per week for 14 weeks, with each session lasting 60 min. Each session began with a 10–15 min standardized warm‐up (based on the SDU recommended protocol [25] or modified FIFA 11+ protocols [26, 27]) including dynamic mobility and progressive running drills. This was followed by 4 × 12‐min small‐sided soccer games (3v3 to 5v5) as previously described [19].
Training progressed from walking soccer during the first four weeks [28] to regular small‐sided games allowing running [29] during the next 10 weeks. This differed from the registered intervention description, which specified 2 weeks of walking soccer followed by games ranging from 3v3 to 7v7. All sessions were led by trained coaches to ensure appropriate progression and adherence.
Adherence was defined as the number of completed sessions out of 42 prescribed sessions. Reasons for missed sessions were recorded throughout the intervention and categorized as musculoskeletal, training‐related injury, medical (non–exercise‐related), non–exercise‐related injury, or non‐medical causes. In a nested acute assessment, 45 exercise‐group participants completed one walking‐soccer and one running‐permitted small‐sided soccer session.
Heart rate and movement were recorded during both sessions using Polar Team Pro units, and perceived exertion was assessed after each session. These measurements were collected during the two assessed sessions, not throughout the 14‐week intervention. Reasons for missed exercise sessions, including reported injuries, were recorded prospectively from reports by participants and coaches.
Reports from participants and coaches were used to document reasons for missed exercise sessions, including injuries. These records did not constitute systematic adverse‐event surveillance in either randomized group.
Sample Size Calculations
The sample size was based on the between‐group difference in change in lean body mass, the primary outcome. In a previous randomized trial involving older adults with prediabetes, mean changes were +0.7 ± 1.5 kg in the exercise group and −0.3 ± 1.6 kg in the control group, corresponding to a between‐group difference of 1.0 kg [19].
We used this difference as the expected effect for sample‐size planning; it is not an established minimum clinically important difference for adults with T2DM. Assuming a two‐sided α of 0.05 and a 1.5:1 exercise‐to‐control allocation, 49 and 33 evaluable participants, respectively, provided approximately 80% power to detect a 1.0‐kg difference using the prior study's change‐score variability. For 85% power, the corresponding numbers were 55 and 37.
The recruitment target was 126 participants, providing a margin above the 92 evaluable participants required for approximately 85% power to allow for withdrawals and missing follow‐up measurements.
Use of Generative AI
OpenAI's ChatGPT was used to assist with drafting and language editing of the manuscript. The authors independently reviewed and verified the AI‐assisted text, including scientific statements and references, and take full responsibility for the final manuscript.
Statistical Analysis
Baseline characteristics were summarized as means (standard deviations) for continuous variables and counts (percentages) for categorical variables. Between‐group comparisons at baseline were performed using independent‐samples t‐tests for continuous variables and χ2 tests or Fisher's exact tests for categorical variables, as appropriate. Changes in medication use during the intervention were compared between groups using χ2 or Fisher's exact tests.
Primary and secondary outcomes were analyzed using constrained linear mixed models with inherent baseline adjustment. In this framework, all participants were constrained to a common mean at baseline, thereby providing adjustment for baseline differences without including baseline as an explicit covariate. The model included fixed effects for group (exercise vs. control), time (pre, post; categorical), and the group × time interaction.
Stratification variables (sex [male/female], age category [< 55 years/≥ 55 years], and GLP‐1RA treatment [yes/no]) were included as additional fixed effects. An unstructured covariance matrix was specified to account for within‐participant correlations between repeated measurements. Model parameters were estimated using restricted maximum likelihood (REML), and statistical inference was based on the Satterthwaite approximation for degrees of freedom.
The primary estimand was the between‐group difference in change from pre‐ to post‐intervention, obtained from the group × time interaction term. Within‐group changes were derived from linear combinations of model coefficients. Estimates are presented with 95% confidence intervals (CI).
The between‐group difference in change in lean body mass was the sole confirmatory analysis. No multiplicity adjustment was applied to secondary outcomes, within‐group contrasts, or GLP‐1RA interaction analyses. These analyses were exploratory, and their p values and 95% confidence intervals are unadjusted and interpreted as hypothesis‐generating.
Exploratory analyses assessed whether baseline GLP‐1RA use modified the intervention effect on lean body mass, fat mass, or absolute VO2peak. For each outcome, the constrained mixed model included time × GLP‐1RA and time × group × GLP‐1RA terms. None of these outcome‐specific interaction tests was specified in the trial registration.
The fat‐mass test was selected post hoc. No separate statistical analysis plan was prepared. Estimates are reported with 95% confidence intervals and unadjusted p values; any stratum‐specific contrasts are descriptive.
Although the trial registration also raised an exploratory question concerning SGLT2‐inhibitor use, no SGLT2‐inhibitor interaction analyses were performed. We assessed sensitivity to missing post‐intervention lean body mass and relative VO2peak values using delta‐adjusted multiple imputations. Missing values were imputed within randomized groups using baseline outcome value, sex, age category, and GLP‐1RA use as predictors (50 imputations).
We then shifted imputed exercise‐group post‐intervention values by specified amounts and analyzed each completed dataset using baseline‐adjusted ANCOVA, combining estimates with Rubin's rules. We identified the shift at which the between‐group p value crossed 0.05. This was a sensitivity analysis using a different analysis model from the primary mixed model.
Model assumptions were evaluated by inspection of residual and fitted value plots. Fasting glucose, C‐peptide, HOMA2‐IR, HOMA2‐%B, HOMA2‐%S, and triglycerides were analyzed on the natural‐log scale. Back‐transformed effects are reported as percentage changes in geometric means.
Between‐group effects represent ratios of those changes, expressed as percentages. Other outcomes were analyzed on their original scales. These secondary analyses are exploratory, and their p values were not adjusted for multiple testing.
The primary analysis followed the intention‐to‐treat principle and included all randomized participants with available outcome data. The mixed‐model framework accommodates incomplete follow‐up under a missing‐at‐random assumption. To explore whether adherence influenced the results, we repeated the constrained mixed‐model analyses comparing exercise participants who attended at least 28 of 42 prescribed sessions (two‐thirds of sessions) with all control participants.
This threshold was selected for the sensitivity analysis. Exercise participants without an attendance record were not classified as highly adherent. Because the exercise subgroup was defined after randomization, these comparisons were exploratory.
Statistical analyses were performed using R (version 4.4.3) with the LMMstar package (version 1.1.0). Statistical significance was defined as two‐sided p < 0.05.
Results
Baseline Characteristics and Participant Flow
Of 829 individuals invited, 152 responded. Twenty‐six were excluded for logistical reasons, leaving 126 eligible individuals; 15 withdrew before allocation. The remaining 111 participants were randomized to exercise (EX; n = 66) or usual care (CON; n = 45).
All randomized participants were invited to follow‐up testing regardless of exercise attendance. Nine EX and six CON participants had no follow‐up data because they did not respond to the invitation or declined it; underlying reasons were not systematically recorded (Figure 1). Follow‐up assessments were attended by 57 EX and 39 CON participants.
Participants without follow‐up were more often women and younger than 55 years, and had lower baseline cardiorespiratory fitness and higher HbA1c than those who attended. These comparisons are descriptive and involve only 15 participants without follow‐up (Table S1). Among those who attended, a further two EX and five CON participants lacked a valid post‐intervention lean‐body‐mass measurement.
Thus, 55 EX and 34 CON participants had lean‐body‐mass measurements at both visits. Baseline characteristics by randomized group and medication changes are presented in Tables 1 and 2, respectively. At baseline, 24 of 66 exercise participants (36.4%) and 16 of 45 usual‐care participants (35.6%) were identified as GLP‐1RA users (Table 1).
Ozempic was the most frequently recorded preparation. Among GLP‐1RA users, metformin was recorded in 15/24 exercise participants (62.5%) and 10/16 usual‐care participants (62.5%); the corresponding numbers for SGLT2 inhibitors were 2/24 (8.3%) and 0/16 (0%).
Training Attendance, Intensity, and Reported Injuries
Attendance records were available for 59 of 66 participants allocated to exercise. They attended a mean of 21.7 ± 14.5 of 42 prescribed sessions (51.7%). Seven exercise participants had no attendance record.
In the nested acute assessment, 45 exercise participants completed both a walking‐soccer and a running‐permitted soccer session. Mean heart rate was 125 ± 17 and 138 ± 12 beats/min, respectively. Among the 59 with records, 11 had a musculoskeletal complaint or non‐training injury recorded as a reason for missed sessions, five had another medical reason, three had a non‐medical reason, and one reason was unclear.
Two records described training‐related injuries, and one additional injury was recorded during the testing phase. No specific reason for missed sessions was recorded for the remaining 36 participants. These records describe reported barriers to exercise attendance, rather than systematic adverse‐event surveillance.
They cannot be used to estimate adverse‐event incidence or compare it between groups, and they do not explain attendance at follow‐up.
Body Composition and Cardiorespiratory Fitness
No between‐group difference in change was detected for the primary outcome, lean body mass (0.04 kg, 95% CI −0.97 to 1.05; p = 0.94; Figure 2A). Nor were between‐group differences detected for fat mass (−0.78 kg, 95% CI −1.88 to 0.31; p = 0.16), body fat percentage (−0.55 percentage points, 95% CI −1.38 to 0.29; p = 0.20; Figure 2B,C), or left or right leg lean mass (p = 0.62 and p = 0.49, respectively).
Compared with usual care, exercise produced a greater increase in absolute VO2peak (between‐group difference in change: 237 mL/min, 95% CI 143 to 331; p < 0.001) and relative VO2peak (2.80 mL/kg/min, 95% CI 1.69 to 3.92; p < 0.001; Figure 3A,B). Yo–Yo IE1 performance also favored exercise (122 m, 95% CI 12 to 232; p = 0.030; Figure 3C). Observed values, outcome‐specific sample sizes, and model‐estimated changes are provided in Table S2.
Among 89 participants with paired DXA and scale‐weight measurements, the difference between DXA total mass and scale weight increased by a mean of 1.33 kg in EX (95% CI 0.72 to 1.94) and 1.50 kg in CON (95% CI 0.83 to 2.18). The between‐group difference in this change was −0.17 kg (95% CI −1.07 to 0.72; p = 0.70; Table S3).
Blood Pressure and Resting Heart Rate
There were no between‐group differences in change in systolic BP (−2.43 mmHg, 95% CI −7.04 to 2.17; p = 0.30), diastolic BP (−1.32 mmHg, 95% CI −4.03 to 1.38; p = 0.33), or RHR (0.98 beats/min, 95% CI −2.15 to 4.12; p = 0.53; Table 3).
Blood Lipid Profile
No between‐group differences were observed for LDL cholesterol (−0.01 mmol/L, 95% CI −0.22 to 0.20; p = 0.92), HDL cholesterol (0.01 mmol/L, 95% CI −0.04 to 0.06; p = 0.72), or triglycerides (−4.0%, 95% CI −17.7 to 12.0; p = 0.60; Table 3). The triglyceride estimate is the back‐transformed between‐group effect from the log‐scale model.
Glycemic Markers
There was no between‐group difference in HbA1c change (0.48 mmol/mol, 95% CI −0.80 to 1.77; p = 0.46) or fasting glucose change (−2.8%, 95% CI −7.5 to 2.1; p = 0.25; Table 3; Figure 4A). Exploratory analyses showed between‐group differences in relative change for C‐peptide (−10.3%, 95% CI −19.0 to −0.6; p = 0.038), HOMA2‐IR (−10.8%, 95% CI −20.1 to −0.6; p = 0.039), and HOMA2‐%S (+12.2%, 95% CI 0.6 to 25.1; p = 0.039). HOMA2‐%B did not differ between groups (−2.6%, 95% CI −10.8 to 6.3; p = 0.55; Table 3; Figure 4B–D).
These secondary analyses were not adjusted for multiple testing.
Other Biomarkers
No between‐group differences were observed for leukocyte count, platelet count, vitamin B12, thyroid‐stimulating hormone, or vitamin D (all p > 0.18; Table 3).
Effect Modification by GLP‐1RA Treatment
In exploratory full interaction models, there was no clear evidence that GLP‐1RA use modified the intervention effect on lean body mass (interaction estimate for non‐users minus users: 1.55 kg, 95% CI −0.58 to 3.67; p = 0.15), absolute VO2peak (90.9 mL/min, 95% CI −103.4 to 285.2; p = 0.36), or fat mass (1.07 kg, 95% CI −1.17 to 3.30; p = 0.35). These were the three outcome‐specific interaction tests performed. The confidence intervals do not establish equivalent effects in GLP‐1RA users and non‐users.
Sensitivity to Missing Data
Under missing‐at‐random imputation, the between‐group estimates were 0.06 kg for lean body mass (95% CI −1.07 to 1.19) and 2.94 mL/kg/min for relative VO2peak (95% CI 1.86 to 4.02). Shifting imputed exercise‐group lean‐body‐mass values by −3 to +3 kg did not produce a statistically significant between‐group difference; the estimated tipping points were −9.4 and +9.0 kg. The relative VO2peak comparison remained nominally significant unless imputed exercise‐group values were shifted downward by approximately 8.2 mL/kg/min (Figure S1).
Sensitivity Analysis by Training Adherence
Twenty‐five of 66 exercise participants attended at least 28 of the 42 prescribed sessions. Compared with all controls, this subgroup had estimated between‐group differences in change of 3.64 mL/kg/min for relative VO2peak (95% CI 2.36 to 4.93), 297 mL/min for absolute VO2peak (95% CI 191 to 403), and 168 m for Yo–Yo IE1 performance (95% CI 28 to 307). The lean‐body‐mass difference was 0.71 kg (95% CI −0.35 to 1.76; p = 0.19).
Body fat percentage also favored this subgroup (−1.19 percentage points, 95% CI −2.14 to −0.23; nominal p = 0.02; Table S4). Because attendance was measured after randomization, these comparisons are exploratory and do not establish that greater attendance caused the larger estimated effects.
Discussion
The primary analysis did not show a between‐group difference in lean body mass after 14 weeks of multicomponent small‐sided soccer training. Exploratory analyses showed greater improvements in relative and absolute VO2peak and Yo–Yo IE1 performance with exercise, whereas between‐group differences in adiposity and HbA1c were not detected. C‐peptide and HOMA2‐IR were lower, and HOMA2‐%S was higher, relative to usual care.
These secondary findings were not adjusted for multiple testing and require confirmation. Exploratory interaction analyses did not provide clear evidence that GLP‐1RA use modified the effects on lean body mass, absolute VO2peak, or fat mass.
Body Composition
We did not detect a between‐group difference in the primary outcome, lean body mass (exercise +1.63 kg; control +1.59 kg; difference 0.04 kg, p = 0.94). Between‐group differences were also not detected for body fat percentage (−0.55 percentage points, p = 0.20), fat mass (−0.78 kg, p = 0.16), or regional leg lean mass. Within‐group changes do not establish an intervention effect when the corresponding between‐group difference is uncertain.
Prior soccer‐based RCTs in prediabetes or T2DM have shown mixed lean mass findings [19, 30], and meta‐analytic estimates suggest trivial–small effects over 12–16 weeks [31]. In this context, the similar lean mass increases in both groups preclude firm causal interpretation and underscore the importance of larger trials with tighter control of co‐interventions to clarify whether multicomponent soccer training yields clinically meaningful hypertrophy in T2DM.
Cardiorespiratory Fitness
Regular exercise consistently improves cardiorespiratory capacity in T2DM [32], and our exploratory fitness findings are consistent with that evidence. Small‐sided soccer integrates continuous aerobic work with high‐intensity and resistance‐like actions, a combination that often matches or exceeds conventional endurance programs [25]. A meta‐analysis of 12–16‐week soccer interventions (two 60‐min sessions per week) reported VO2peak gains of ≈3.5 mL/kg/min [33].
In our trial, relative VO2peak increased by 1.86 mL/kg/min in the exercise group and declined by 0.94 mL/kg/min in controls, yielding a between‐group difference of 2.80 mL/kg/min. Absolute VO2peak increased by 134 mL/min in the exercise group and declined by 103 mL/min in controls, yielding a difference of 237 mL/min. Yo–Yo IE1 distance increased by 140 m in the exercise group and by 18 m in controls, with a between‐group difference of 122 m.
Higher aerobic capacity is associated with better cardiovascular prognosis, but this trial did not assess clinical events. The fitness gains were achieved through a supervised, game‐based intervention, although attendance averaged 21.7 of 42 prescribed sessions among participants with records. Whether this format improves long‐term adherence or clinical outcomes requires further study [17].
Blood Pressure and Lipid Profile
There were no differences in systolic or diastolic BP or RHR. Within‐group changes were generally small, with diastolic BP decreasing modestly in the intervention group only. Similarly, there were no between‐group differences in LDL cholesterol, HDL cholesterol, or triglycerides.
LDL decreased within both groups, HDL increased within the intervention group, and triglycerides remained unchanged.
Although between‐group contrasts were not statistically significant, these patterns are directionally consistent with cardiometabolic adaptations to regular training. Meta‐analyses of randomized controlled trials of small‐sided soccer report modest but clinically relevant reductions in BP and improvements in lipid profiles across diverse populations [31]. The absence of clear between‐group effects in the present trial may reflect relatively well‐controlled baseline risk factors, concurrent pharmacotherapy, and the limited duration of the intervention, which together may have attenuated the detectable incremental benefit of exercise compared with usual care.
Glycemic Control
HbA1c showed no between‐group difference in change but increased modestly in both groups, likely reflecting the low baseline HbA1c and limited scope for improvement in a relatively well‐controlled cohort. Meta‐analytic evidence demonstrates that structured exercise training is generally associated with reductions in HbA1c in individuals with T2DM, with greater effects observed at higher exercise volumes [34]. The absence of an HbA1c response in the present study may therefore reflect both the relatively well‐controlled baseline glycemic status and the specific characteristics of the intervention.
Exploratory analyses showed lower C‐peptide and HOMA2‐IR and higher HOMA2‐%S in the exercise group relative to usual care. No between‐group difference was detected for fasting glucose, HbA1c, or HOMA2‐%B. HOMA2‐IR and HOMA2‐%S are related indices derived from the same fasting measurements, rather than independent evidence of two separate effects.
These findings do not establish improved insulin sensitivity or reduced insulin demand. They require cautious interpretation because multiple secondary outcomes were tested and participants took their usual morning medications before blood sampling.
Influence of GLP‐1RA Therapy on Changes in Body Composition and Cardiorespiratory Fitness
The exploratory interaction analyses did not provide clear evidence that GLP‐1RA use modified the intervention effects on lean body mass, absolute VO2peak, or fat mass. The interaction estimates were imprecise, so these results do not establish that exercise responses are equivalent among users and non‐users. GLP‐1RA treatment was not randomized, and larger studies designed to test medication‐by‐exercise interactions are needed.
Strengths and Limitations
Strengths include randomized allocation, a supervised multicomponent intervention, and stratification by age, sex, and GLP‐1RA use. The primary outcome was assessed by DXA, and repeated measurements were analyzed using constrained linear mixed models.
The sample size limited precision, particularly for medication interaction analyses. Fifteen of 111 randomized participants had no follow‐up data. Post‐intervention lean body mass was unavailable for a further seven who attended follow‐up.
Participants without follow‐up had higher baseline HbA1c and lower cardiorespiratory fitness than those who attended (Table S1). Thus, participants with follow‐up data may not fully represent the randomized cohort. The mixed‐model estimates rely on a missing‐at‐random assumption.
If missing outcomes differed in ways not accounted for by observed baseline data, the treatment estimates could be biased. Our sensitivity analyses examined alternative assumptions but cannot eliminate this concern. No multiplicity adjustment was applied to secondary outcomes or within‐group contrasts, so those findings are exploratory.
Medication changes during follow‐up may have contributed to residual confounding. Self‐reported medication data showed limited sensitivity compared with registry‐derived data (Table S5). Dietary intake and physical activity outside the intervention were not objectively monitored, and attendance at prescribed sessions was modest.
Injury records were collected in connection with missed exercise sessions and were insufficient to estimate or compare adverse‐event incidence between groups.
Participants took their usual morning medications before fasting blood sampling. Medications affecting glucose or endogenous insulin secretion may therefore influence C‐peptide and the derived HOMA2 indices. We interpret these secondary findings cautiously.
GLP‐1RA treatment was not assigned by the trial. Prescription records could include overlapping preparations and did not reliably establish dose, treatment duration, or changes during follow‐up. These limitations may affect interpretation of the exploratory medication‐interaction analyses.
Within‐group increases in DXA‐derived lean mass should be interpreted cautiously. In participants with paired DXA and scale‐weight measurements, changes in the DXA compartments did not reconcile with changes in scale weight, and the difference between DXA total mass and scale weight increased at follow‐up in both groups. We could not determine the source of this discrepancy from the available measurements.
The 14‐week duration and relatively well‐controlled mean baseline HbA1c may have limited detectable changes in some outcomes. The absence of a detected between‐group difference in lean body mass should not be interpreted as evidence of equivalent effects.
Conclusion
Fourteen weeks of small‐sided soccer training did not produce a detectable between‐group difference in the primary outcome, lean body mass. Relative VO2peak improved by 2.80 mL·kg−1·min−1 compared with usual care, with corresponding improvements in absolute VO2peak and exercise capacity. No clear between‐group differences were detected in adiposity or HbA1c.
The differences in C‐peptide and HOMA2 indices arose from exploratory analyses without adjustment for multiple testing and require confirmation. Exploratory interaction tests provided no clear evidence that GLP‐1RA use modified the exercise effects on lean body mass, fat mass, or absolute VO2peak.
Perspective
Future trials should prospectively define medication‐by‐exercise interaction tests and recruit enough participants to estimate those interactions precisely. Documenting medication type, dose, duration, and changes during follow‐up would help interpret responses to combined treatment. Objective measures of training exposure and studies spanning a wider range of baseline glycaemic control could clarify whether exercise responses vary with treatment or disease status.
Funding
This study was supported by the Research Council Faroe Islands (grant no. 0358) and Betri Stuðul. The funders had no role in study design, data collection, data analysis, data interpretation, or manuscript preparation.
Ethics Statement
The study was approved by the Faroe Islands Ethical Committee (approval no. 2024‐12).
Consent
All participants provided written informed consent prior to participation.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information