The study protocol was registered with PROSPERO (CRD42023417622) (56). This systematic review and NMA were conducted in accordance with the guidelines provided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement (PRISMA-NMA).

The MEDLINE (PubMed), Embase, Cochrane Library, EBSCO, and Web of Science databases were systematically searched for relevant articles using a comprehensive electronic search strategy, as shown in Supplementary Appendix 2. The search strategy was based on key phrases related to the PICOS tool: (P) Population: “Hypertension” OR “obesity” OR “Type 2 diabetes” OR “T2D” OR “metabolic syndrome” OR “Older persons over 60 years”; (I) Intervention: “physical activity” OR “training” OR “aerobic exercise” OR “moderate intensity continuous training” OR “moderate interval training” OR “high interval training” OR “resistance training” OR “strength training” OR “combined training” or “sprint interval training” or “high intensity interval training”; (C) Comparator: “control group” or “no exercise” or “usual care”; (O) Outcomes: “arterial stiffness” or “pulse wave velocity” or “PWV”; and (S) Study type: “randomized controlled trial”, “randomized”, “placebo”, “RCT”. The search was limited to English-language articles published between the inception of the databases up to February 2024. We included RCTs comparing different types of exercise on AS in high-risk populations of CVD. In addition, we retrieved similar review articles and screened their references to maximize the completeness of the sources.

Duplications were initially eliminated using EndNote X9 software (Clarivate Analytics, Philadelphia, PA, USA). Subsequently, two researchers (R-SW and X-LF) independently screened titles and abstracts to identify all potentially relevant studies. The same two reviewers independently identified and assessed studies that met the inclusion criteria. Disagreements were resolved through discussion or consultation with a third expert (XY), if necessary. The detailed inclusion criteria were as follows: (1) all studies had to be RCTs; (2) subjects had known risk factors associated with CVD according to the American College of Sports Medicine guidelines; (3) interventions included INT, AE, RT, and CT; (4) the comparator received no intervention, usual care, or health education; (5) outcomes included PWV, SBP, and DBP; and (6) the studies had to be written in English.

The exclusion criteria were as follows: (1) the subjects of the study were healthy adults; (2) the type of experiment was animal experiments or randomized crossover experiments; (3) the effect of exercise intervention was acute (<3 weeks); (4) incomplete data or no control group; (5) returned reviews, duplicate publications, letters to the editor, and meta-analysis articles.

In the included randomized controlled trials, exercise interventions comprised INT, AE, RT, and CT. As depicted in Table 1, we operationally defined these four modalities of exercise intervention as follows.

The primary outcome assessed in this study included PWV, with secondary outcomes being SBP and DBP. There are several types of PWV: carotid–femoral pulse wave velocity (cfPWV), brachial-ankle pulse wave velocity (baPWV), central pulse wave velocity (cPWV), central-brachial pulse wave velocity (cbPWV), etc. In general, cfPWV is considered the first gold-standard measure of AS. The cfPWV is defined as the ratio of the distance between the measurement sites to the time difference for the distal pulse wave to reach the measuring sites. The cutoff point for AS was defined as a PWV value exceeding 10 m/s (58).

In this study, PWV primarily refers to the measurement of cfPWV. Since we also included baPWV indicators, we implemented a data transformation method to reconcile the measurement discrepancies between these two metrics. Specifically, we utilized the regression equation cfPWV=0.65×baPWV+2.1 to achieve accurate data correction (59). This method effectively addresses the variations in measurements, thereby improving the reliability of our results. The term SBP refers to the maximum arterial pressure during ventricular contraction, while DBP represents the arterial pressure when the heart is in a relaxed state. Pulse pressure (PP) is defined as the difference between SBP and DBP (60). The criteria for Stage I hypertension were defined as SBP of 130–139 mmHg and/or DBP of 80–89 mmHg. These parameters were incorporated into the outcomes and served as the indicators for assessing AS.

The data were extracted independently by two investigators (R-SW and XY), with any disagreements resolved through consensus or the input of a third author (X-LF) if necessary. Information collected included the first author, publication year, country, subject characteristics (sample size, gender, age, PWV, SBP, DBP, and concomitant diseases), intervention details (exercise type, intensity, duration, frequency, and supervision status), and outcome measures reported in each eligible study, as shown in Supplementary Appendix 4. In cases where information was insufficiently provided by studies included in this review article, email communication was initiated to obtain missing values.

The risk of bias (ROB) of the included studies was evaluated by two researchers based on the Cochrane Risk of Bias assessment tool (61), which includes the following seven domains, encompassing seven distinct domains: (a) allocation generation, (b) allocation concealment, (c) blinding of participants and personnel, (d) blinding of outcome assessment, (e) handling of incomplete outcome data, (f) freedom from selective reporting bias, and (g) other forms of bias. These key domains were considered key domains for assessing the risk of bias. However, due to the inherent difficulty in blinding participants to an exercise intervention, this component was not included in the overall risk of bias score. Subsequently, we categorized each study's overall risk of bias into three levels: low risk if none of the above domains were rated as high risk and ≤3 domains were rated as unclear; moderate risk if one study was rated as high risk or no study was rated as high risk but ≥4 studies were rated as unclear; and high risk for all other cases (62).

The Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework was utilized to evaluate the certainty of the evidence contributing to the network estimates for both primary and secondary outcomes (63).

The experimental effect was estimated by combining the pre-to-post changes of both the experimental and CON. The standard deviation (SD) of the change value was calculated using the formula provided in the Cochrane Handbook (version 6.3) (64) [the formula is SDchange= SDbaseline2+SDfinal2D−(2×Corr×SDbaseline×SDfinal)].

Heterogeneity among studies was assessed using network estimates and pairwise meta-analytic techniques with Review Manager 5.3 (Nordic Cochrane, Denmark). A sensitivity analysis was also conducted to explore this heterogeneity (65). Pooled effect estimates were calculated using a random-effects model, and mean difference (MD) values were determined for PWV, SBP, and DBP. Heterogeneity was quantified using the I² statistic and Cochran's Q-test, with significant heterogeneity defined as an I² >50% or a p-value ≤0.10 (66). Publication bias was evaluated with a funnel plot and Begg's test.

Statistical analysis was performed using STATA 16.0 (STATA Corp, College Station, TX, USA) through a frequentist framework and random-effects multivariate NMA (67). Weighted mean differences were reported for continuous variables, with pooled effect estimates accompanied by 95% confidence intervals (CIs) and 95% prediction intervals.

Exercise interventions were compared using network geometry, with lines connecting nodes representing direct relationships. The size of each node and the thickness of connecting lines reflected the number of studies (68).

Inconsistencies were estimated using the loop-specific approach, node-splitting technique, and global inconsistency method to assess ring, local, and global inconsistencies (69). A p-value <0.05 indicated that the inconsistency requirement was met, permitting further NMA analysis (70). The transitivity assumption, which assumes valid indirect comparisons and uniform effect modifiers, was evaluated using a consistency model to ensure random allocation of interventions (70, 71). The network contribution diagram calculated the impact of each direct comparison on both individual and overall NMA results.

The cumulative ranking curve (SUCRA) was used to rank the effects of different exercise modalities (64). SUCRA values range from 0 to 100, with higher values indicating better outcomes (72). Publication bias in the network meta-analysis was assessed using a funnel plot and symmetry criterion.

A total of 1,433 articles were retrieved from the database, while an additional 7 articles were obtained through alternative sources, resulting in a combined pool of 1,440 articles for screening. After eliminating duplicate entries, a total of 845 unique articles remained. Subsequently, after reviewing the titles and abstracts, 568 articles were excluded, leaving us with 277 potential candidates. Further evaluation of the full-text content led to the exclusion of an additional 234 articles, resulting in only 43 remaining articles that met our criteria. In conclusion, 43 eligible articles were ultimately identified as per Figure 1.

Supplementary Appendix 4 provides details on the studies included, conducted between 2003 and 2023 across various regions: Asia (12), North America (13), South America (2), Europe (11), Oceania (1), Latin America (2), and Africa (2). The studies involved 2,034 high-risk CVD participants. The experimental group included 1,056 subjects, divided into the following groups: INT (365), AE (327), RT (149), and CT (215). The CON group included 978 subjects.

Participants were about 51% women, aged 10–75 years. Six studies focused solely on men, while 37 studies included both genders. All studies followed the ACSM guidelines for high-risk cardiovascular populations, considering obesity, diabetes, hypertension, and advanced age as risk factors.

Interventions included INT, AE, RT, and CT (AE + RT) at moderate to high intensity. Exercise duration ranged from 8 to 96 weeks, with activities such as running, cycling, rowing, swimming, stepping exercises, and resistance training. Sessions were held three to seven times per week, lasting 30–90 min each. Control groups received usual care or no exercise.

Of the studies, 33 used supervised exercise, 3 combined supervision with home-based interventions, and 7 had no supervision. Carotid–femoral PWV was measured using applanation tonometry (SphygmoCorCPV, ATCor Medical) or SphygmoCor software (version 9.0). Measurements were taken using a high-fidelity micromanometer (Millar Instruments), calculating PWV by dividing the distance between the carotid and femoral recording sites by the time delay between their pulse waves (71).

The ROB assessments for each study are presented in Supplementary Appendix 5. Two studies exhibited a high risk of bias in randomization serialization, while the concealment of allocation methods was reported in 15 studies. Blinding of outcome assessment had a low risk of bias in 30 studies, whereas 7 studies showed a high risk of bias regarding missing outcome values. Selective reporting demonstrated a low risk of bias in 10 studies. In addition, sample sizes smaller than n < 10 or significant errors in the measurement process were considered to have a high risk of other biases, with only 33 studies showing a low risk of bias. In summary, 24 articles were evaluated as having low ROB, while moderate and high ROB were observed in 14 and 5 articles, respectively.

This study examined the impact of four exercise interventions (INT, AE, RT, and CT) on PWV, SBP, and DBP in individuals at high risk of CVD. The network meta-analysis, including 43 RCTs with 2,034 participants, found that CT and INT significantly reduced PWV, whereas RT and AE did not. As shown in Figure 4, SUCRA analysis ranked CT as the most effective, followed by INT. AE and RT were less effective. Subgroup analysis showed that moderate-intensity INT was most effective for AS, while high-intensity CT outperformed AE in improving AS.

In studies involving individuals at high risk of CVD, Montero et al. (10) examined the effects of AE on arterial AS in subjects with prehypertension and hypertension. Their results showed that AE, particularly when combined with a reduction in SBP below the median or extended duration, positively affected AS in patients with prehypertension or hypertension.

Montero et al. (9) also studied the impact of AE on AS in obese individuals, finding no significant improvement in AS among middle-aged and older obese adults undergoing AE. However, subgroup and meta-regression analyses suggested that low-intensity AE combined with a reduction in DBP could reduce AS in obese individuals.

Marzolini et al. (7) compared the effects of CT (AE + RT) to AE alone in patients with coronary heart disease. They found that AE + RT led to superior improvements in AS and was more effective than AE alone in enhancing body composition, strength, and certain cardiovascular health indicators.

Evans et al. (11) investigated the effects of RT on AS in high-risk CVD populations. Their study showed that RT did not worsen AS or the blood system, despite increasing sympathetic norepinephrine levels, which may lead to vasoconstriction and higher blood pressure. In fact, RT could be more effective than AE in improving cardiovascular health-related conditions.

Consistent with previous research, this study found that both a CT program (combining AE and RT) (SMD = −0.69, p < 0.001) and an INT program (SMD = −0.50, p < 0.001) effectively reduced PWV in individuals at high risk for CVD. Given that an effect size of 0.5 or greater suggests practical relevance (62), both CT and INT are recommended as effective interventions for improving AS in high-risk CVD populations.

In contrast to previous findings, AE (SMD = −0.33, p > 0.01) did not significantly reduce PWV in high-risk populations despite its known benefits for AS in healthy individuals. AE's positive effects in healthy people are attributed to mechanisms like vascular remodeling, enhanced endothelial function, and reduced oxidative stress (9). However, AE did not significantly lower AS in individuals with hypertension, obesity, or elderly individuals who are at high risk for CVD. Effective AE for high-risk populations typically involves lowering systolic blood pressure below the median, extending the intervention duration, or adjusting exercise intensity. Therefore, AE is not recommended as the primary intervention for improving AS in high-risk CVD populations.

The impact of RT on AS remains debated. Some studies suggest that high-intensity RT may increase blood pressure and reduce AS, while others find no effect (6, 10). Our study also showed that RT (SMD = −0.11, p > 0.01) did not significantly reduce PWV. However, the subgroup analysis revealed that high-intensity RT effectively decreased PWV (SMD = −0.40, p < 0.001), suggesting its potential benefit for improving AS in high-risk individuals. Safety concerns related to high-intensity RT, including potential increases in blood pressure and acute cardiac strain, limit its clinical recommendation for improving AS.

The biological mechanisms behind RT's effects on AS vary by health status. In young, healthy individuals, high-intensity RT can stimulate the sympathetic nervous increased AS (73). In contrast, individuals at high risk for CVD, such as those with metabolic syndrome or diabetes, may experience decreased hormone secretion that mitigates these adverse effects. In addition, RT's impact on arterial structure and load-bearing characteristics can positively influence AS.

Finally, high-intensity CT (SMD = −0.71) was more effective than moderate-intensity CT (SMD = −0.67) in improving AS in high-risk CVD populations. Previous studies have suggested CT is the best intervention for improving AS in general populations. However, the sequence of AE and RT in CT is important; CT with AE following RT but not before RT can reduce AS (8). This study included CT with both AE and RT, but the sequence was not consistently specified. Further research is needed to determine the optimal order of exercises in CT programs.

The secondary findings of this study evaluated the impact of four exercise interventions (INT, AE, RT, and CT) on blood pressure, including SBP and DBP. Analysis of SBP showed that INT (SMD = −0.382, p < 0.001) and AE (SMD = −0.369, p < 0.001) significantly reduced SBP in high-risk CVD populations, while CT (SMD = −2.22, p > 0.01) and RT (SMD = −1.05, p > 0.01) did not. SUCRA ranking indicated that INT had the highest effect (SUCRA = 81.3), followed by AE (SUCRA = 80.4) and CT (SUCRA = 52.5), with RT scoring the lowest (SUCRA = 29.7).

For DBP, none of the interventions showed significant reductions. SUCRA analysis revealed that CT had the best effect (SUCRA = 78.8), followed by AE (SUCRA = 69.2) and INT (SUCRA = 54.9), while RT had the lowest effect (SUCRA = 27.4).

These results align with previous research confirming AE and INT's benefits on SBP. Some studies suggest that high-intensity INT might be more effective than AE for individuals with cardiopulmonary issues or advanced age due to shorter exercise and recovery times. The impact of RT on blood pressure remains debated, with mixed results in prior studies (74). This study found that neither RT nor CT significantly improved blood pressure.

The lack of significant effects on DBP across all interventions may be due to individual variability in high-risk CVD populations. Normal blood pressure values are 130/80 mmHg, and DBP typically should not exceed 90 mmHg (75). Although AE, RT, and CT have been shown to improve various cardiovascular metrics, age-related changes in blood vessels, such as elastin fiber degeneration and increased collagen, may obscure the immediate benefits on DBP for high-risk groups like the elderly, obese, or those with heart disease.

This study examined how exercise intensity affects PWV. The results showed that in the moderate-intensity group, CT significantly improved PWV [SMD = −0.67, 95% CI (0.07 to 1.28), p < 0.05; see Supplementary Appendix 15.10]. In the high-intensity group, both RT [SMD = 0.40, 95% CI (0.12 to 0.68), p < 0.05] and HIIT [SMD = 0.40, 95% CI (0.16 to 0.64), p < 0.05] were effective in improving PWV (see Supplementary Appendix 16.10).

SUCRA rankings showed that for moderate-intensity exercise, INT (SUCRA = 83.3) was the most effective, while RT (SUCRA = 21.6) was the least effective (see Supplementary Appendix 15.9). For high-intensity exercise, CT (SUCRA = 85.7) was the most effective, and AE (SUCRA = 29.5) was the least effective (see Supplementary Appendix 16.9).

These findings underscore the importance of exercise intensity in determining the effectiveness of interventions on PWV. Previous research shows varied effects of INT, AR, RT, and CT based on intensity. For instance, Way et al. (12) found that high-intensity aerobic exercise was more effective for improving AS in healthy adults than moderate-intensity exercise. On the other hand, Montero et al. (9) reported that low-intensity aerobic exercise improved AS in obese individuals, while moderate to high-intensity aerobic exercise did not.

In summary, higher-intensity CT and RT may be more effective for improving PWV, offering valuable guidance for choosing suitable exercise intensities for individuals at high cardiovascular risk.

This study utilized a network meta-analysis, allowing us to compare multiple intervention modalities simultaneously. This approach provides a broader research perspective and significantly enhances the study's value. We focused on high-risk CVD populations, an area that has been underexplored in previous research. Since AS is a common complication of CVD, examining this group is crucial for validating the preventive and therapeutic benefits of exercise on AS.

Our findings are notable and diverge from previous research. We recommend CT and INT as preferred exercise methods for improving arterial stiffness in high-risk CVD populations. In addition, moderate-intensity AE and supervised high-intensity RT are also recommended for these individuals.

However, this study has certain limitations. First, we only included INT, AE, RT, and CT as common interventions, excluding other forms like mental and physical exercises, stretching, and pharmacological treatments, which have limited data. Second, while we focused on PWV, SBP, and DBP as key indicators of AS, other measures such as augmentation index, blood lipids, and blood sugar were not included due to validity concerns. Third, insufficient sample size is also a significant limitation when conducting subgroup analyses on exercise intensity. In addition, despite various analyses, the heterogeneity in CT remains unresolved and is acknowledged as a limitation in this study. Future research with larger sample sizes should address these limitations and explore additional interventions. Potential limitations in future studies may include variability in study populations, differences in study designs, and the influence of unmeasured confounders.