This umbrella review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [30], and it is registered in the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42024500736).

The inclusion criteria and description of studies in this review followed the PICOS (Population, Intervention, Control, Outcomes, and Study design) framework for reviews [31].

We performed a literature search for scientific articles published until January 19, 2024, using the following databases: PubMed (Medline), PEDro, Embase, Web of Science, and Cochrane Library. Multimedia Appendix 1 provides the search strategies, tailored for each database. A total of 2 independent reviewers (JLS-G and JLS-R) conducted the search using the same methodology. Any discrepancies during this process were resolved through consensus including a third reviewer (VN-L). In addition, we manually examined the reference sections of the original studies, and, if needed, contacted the authors for additional information.

A total of 2 independent reviewers (JLS-G and JLS-R) initiated a screening process to evaluate relevance of the systematic reviews and meta-analyses. The initial screening was based on the title, abstract, and keywords for each review. When there was no consensus, or if the abstracts provided insufficient information, the full text was examined. During the second screening phase, the full text was evaluated to confirm inclusion criteria. Data presented in the results section were extracted using a protocol to ensure retrieval of the most relevant information for each study. The sample sizes, type of exercise, duration, means, and SDs of the telomere size data were extracted for each of the studies included in this review. During the analysis, we checked whether the results were repeated in different reviews or meta-analyses; if the effect sizes coincided, only one of the effect sizes was chosen; if the effect sizes did not match, the primary study was selected.

A total of 2 independent reviewers (JLSG and VNL) evaluated the methodological quality of the systematic reviews and meta-analyses, using the modified quality assessment scale for systematic reviews (AMSTAR-2 [A Measurement Tool to Assess Systematic Reviews]). AMSTAR-2 [33] is a questionnaire comprised of 16 domains, with simple categorical options: “yes,” when the result is positive; “no,” when the standard is not met or there is not enough information to answer it; and “partial yes,” when there was partial adherence to the standard. In addition, we calculated the kappa coefficient (κ) and percentage (%) agreement scores to assess reliability before reaching consensus. Interrater reliability was estimated using κ, where κ>0.7 indicates a high level of agreement between reviewers, κ of 0.5‐0.7 indicates a moderate level of agreement, and κ<0.5 suggests a low level of agreement [34].

We evaluated the risk of bias using the Risk Of Bias in Systematic reviews (ROBIS) assessment, which includes three domains: (1) relevance of assessment (optional); (2) identification of concerns with the review process through 4 domains related to study eligibility criteria, identification and selection of studies, data collection and study appraisal, and synthesis and findings; and (3) judgment on the risk of bias. The ROBIS tool includes signaling questions to assess specific domains and to guide the judgment of the systematic review’s risk of bias, with responses categorized as “yes,” “probably yes,” “probably no,” “no,” or “no information.” The risk of bias is then categorized as “low,” “high,” or “unclear” [35].

A total of 2 independent reviewers (JLS-G and JLS-R) assessed the risk of bias in the selected studies. In addition, we computed the kappa coefficient (κ) and percentage (%) agreement scores to evaluate reliability before reaching consensus.

The Physical Activity Guidelines Advisory Committee Grading Criteria (PAGAC) were used to evaluate grading of evidence. The criteria used for assessing the quality of evidence included (1) applicability of the study sample, exposures, and outcomes to the research question; (2) generalizability to the population of interest; (3) risk of bias/study limitations; (4) quantity and consistency of findings across studies; and (5) magnitude and precision of the effect. Based on this information, final evidence grades and conclusion statements for each research question were formulated [36].

The overlap analysis of primary studies among the systematic reviews was performed with the Graphical Representation of Overlap for OVErviews (GROOVE) tool [37]. Using a matrix of evidence, GROOVE establishes the number of primary studies and systematic reviews, the absolute number of overlapped and nonoverlapped primary studies, and an overall corrected covered area (CCA) assessment. GROOVE also offers a detailed CCA assessment for each possible pair of systematic reviews (or “nodes”), with a graphical and easy-to-read representation of these results. In addition, it provides an optional function that incorporates the structural missingness within the matrix.

Given the high variability of designs, patients and endpoints across studies, results were integrated following a random effects model. The Hedges unbiased standardized mean difference was used to determine effect sizes. Heterogeneity was measured using Q-tests, I² coefficients, and prediction intervals [38]. Metaregression and metapartition [39] were used to explain heterogeneity. These techniques involve partitioning sum of squares of the homogeneity test into 2 components, Q_B and Q_W, to account for quantitative or qualitative variables. If the Q_E value was large and statistically significant compared with the value of Q_W, it was considered that this variable explained part of the heterogeneity found in the integration of results across studies. If the variable was qualitative, subgroups of studies, which differed in the effect size and also showed less heterogeneity, were created. The procedure was repeated for subgroups that still presented large heterogeneity for other variables. Publication bias was assessed using the contour-enhanced funnel plot, Egger test, Doi plot, and LFK index procedures. 95% CIs were calculated for effect sizes. The significance level was set at 5%. The analyses were carried out with the meta v7.0, metafor v4.6 and metasens 1.5‐2 libraries of the R statistics software (version 4.4.0; R Foundation for Statistical Computing).

As this is an umbrella review, it is not necessary to have the approval of the ethics committee.

The initial literature search revealed 3628 records. In addition, 2 more were retrieved manually from the references. A total of 12 systematic reviews were eligible for qualitative synthesis, and 5 of them allowed a meta-analysis. Figure 1 shows the study screening strategy.

Table 1 lists the characteristics of the systematic reviews and meta-analyses (type of study, design, sample, intervention, lab technique to measure TL, risk of bias, evaluation of quality, and conclusions).

The overall confidence ranged from critically low to moderate quality scores. A total of 2 studies (16.6%) obtained an overall confidence of moderate, and 10 studies (83.4%) had critically low quality (Multimedia Appendix 2). The items with the highest scores were “did the review authors use a comprehensive literature search strategy?”, “did the review authors describe the included studies in adequate detail?”, “did the review authors provide a satisfactory explanation for, and discussion of, any heterogeneity observed in the results of the review?”, and “did the review authors report any potential sources of conflict of interest, including any funding they received for conducting the review?” The lowest scoring items were “did the review authors account for risk of bias in individual studies when interpreting/ discussing the results of the review?”, “did the review authors report on the sources of funding for the studies included in the review?”, and “if meta-analysis was performed, did the review authors assess the potential impact of risk of bias in individual studies on the results of the meta-analysis or other evidence synthesis?” The interrater reliability of the methodological quality assessment was high (κ≥0.8).

Multimedia Appendix 3 summarizes the risk of bias using ROBIS. 40% (5/12) of the studies had a low risk of bias, 20% (2/12) had an unclear risk of bias, and 40% (5/12) had a low risk of bias. The domains related to “data collection” and “synthesis and findings” had the highest risk of bias. On the other hand, the domain related to “eligibility criteria” had the lowest risk of bias. The interrater reliability of the risk of bias was high (k>0.8).

The grades of evidence were classified as not assignable, limited, moderate, or strong according to the PAGAC. The level of evidence was limited to moderate among the studies. Most reviews evaluated a heterogeneous population of individuals, which limited the applicability and generalizability of the results. The domain “quantity and consistency” was affected because some studies had inconsistency in the direction of or the effect size itself (Multimedia Appendix 4).

A total of 39 primary studies were identified across all systematic reviews and meta-analyses, 24 of which were distinct studies. The overall overlap in the matrix of evidence was moderate (CCA=32.5%) and it remained moderate (CCA=15.63%) after adjusting for the chronological structural missingness. Multimedia Appendix 5 shows a graphical representation of the GROOVE results.

The number of studies in the systemic reviews that were finally integrated in the quantitative (meta-analytic analysis) was 22. The estimated effect size was 0.28 (95% CI 0.118-0.439), with a heterogeneity test value (Q) of 43.08 (P=.003). The I² coefficient was 51% and the prediction interval was −0.26 to 0.81. The number of weeks of intervention explained part of this heterogeneity (Q_B=7.54; P=.006). This factor explained 20% of the overall heterogeneity. Figure 2 shows that studies where the number of weeks of intervention was greater than 30 had smaller effect sizes. In fact, if the intervention time was categorized into <30 and ≥30 weeks, Q_B was 11.64 (P<.001).

Figure 3A shows that the effect size was 0.39 for studies with an intervention length of less than 30 weeks, and −0.01 for those with interventions longer than 30 weeks. There was no significant difference in the average number of weekly sessions between these two groups (3.3 vs 3 sessions, respectively). Interestingly, the first group presented a statistically significant heterogeneity (I²=41%), whilst the second, despite having a large range of outcomes, did not. Since heterogeneity was high in the group of studies with shorter interventions (<30 wk), it was analyzed with respect to other variables. The factor that better explained this heterogeneity was the type of physical exercise defined by the 4 categories (subgroups) as shown in Figure 3 (Q_B=11.15, P=.01). One subgroup was comprised of just 1 study that included a combination of aerobic exercise with endurance training and achieved a large effect size. The other 3 subgroups did not have a statistically significant heterogeneity between them, although the subgroup of studies defined by an intervention based on HIIT, which alternates short periods of intense anaerobic exercise with brief recovery periods, had a marked I² value (49%). The HIIT subgroup only included 3 studies, one of which had a very high effect size compared to the other two. When we compared the effect sizes of the 3 types of noncombined physical exercises, the value was small for aerobic exercise and endurance training, and moderate for HIIT.

The contour-enhanced funnel plot (Multimedia Appendix 6) did not identify publication bias. Many of the effect sizes were not significant, and at the top of the funnel, there was some asymmetry, which could be due to the heterogeneity of results. The nonpresence of publication bias was confirmed with the Egger method (P=.10). The Doi plot and LFK index also found a weak asymmetry (Multimedia Appendix 7).

This umbrella review studied the effects of physical exercise on TL. The estimated overall effect size indicated a small-moderate positive impact of exercise on TL. However, we found some discrepancies in the results across studies that may be explained by several methodological and contextual reasons, including different methodologies for measuring TL, such as quantitative polymerase chain reaction and terminal restriction fragment assays, that can contribute to inconsistencies in TL calculations. In this regard, Smith et al [47] pointed out that intraindividual TL can vary significantly depending on the lab technique used to measure it. Differences in study populations could also be a factor responsible for disparities in results. Demographic factors, such as age, sex, ethnicity, and health status, can influence the relationship between exercise and TL. For instance, LaRocca et al [48] and Puterman et al [21] suggest that the benefits of exercise on TL may be more pronounced in older adults and in those with higher stress levels. In addition, the duration and intensity of the intervention could be another confounding factor. Our findings suggest that interventions longer than 30 weeks tend to show a lesser positive impact on TL. This could be due to a ceiling effect where additional benefits of exercise do not translate into further increase of TL. This is consistent with Werner et al [29], who observed that the beneficial effects of exercise on TL are more evident in the early phases of physical exercise programs.

The high heterogeneity in study designs, including interventions, participant adherence, and control measures, could have also contributed to different outcomes. In fact, Denham et al [23] and Puterman et al [21] have already highlighted the importance of these methodological issues when interpreting TL results. Studies such as Cherkas et al [20] had found that vigorous aerobic physical activity was associated with longer telomeres in adults, which is in line with our finding of a positive effect of aerobic exercise, especially when combined with endurance training, on TL. However, a systematic review by Du et al [49] did not identify a significant association between aerobic exercise and TL in younger populations, suggesting that age may be an important moderating factor. Denham et al [23] found mixed results related to endurance physical training. Our review also found significant variation in the effects of noncombined endurance training, which suggests that factors such as the intensity and duration of this type of physical exercise may influence the magnitude of the effect on TL. Finally, HIIT had previously shown a significant positive effect on TL [29]. Our review results are consistent with this; nonetheless, the low number of HIIT studies and variety of designs limit generalizability.

We present a comprehensive analysis of systematic reviews and meta-analyses of the effect of physical exercise on TL, providing a broad and robust synthesis of the available evidence. The methodology used to assess the quality of studies and the application of standardized tools for evaluating the risk of bias and heterogeneity strengthens the validity of our results. However, our work has some limitations. The methodological quality of the reviews included in this analysis varies, with many of them classified as critically low. Also, the heterogeneity in study designs, samples, and interventions makes difficult to generalize the results, even after the use of a random effects model to integrate outcomes. Furthermore, some meta-analyses reported results only using postintervention measures, whilst others used postintervention differences adjusted for preintervention measures; this may lead to variations in effect sizes.

The implementation of exercise programs into clinical practice could be an effective strategy to limit telomere shortening, promoting healthy aging and reducing the incidence of age-related diseases. HIIT and the combination of aerobic exercise with endurance training may be promising interventions in this regard but still require to be further tested in high-quality studies. In addition, physical exercise should be tailored to each individual’s physical capability and health condition to maximize cellular health benefits. Implementing physical exercise programs will also require ongoing health education and monitoring to ensure adherence to and adjustment of interventions based on observed outcomes. Future studies should investigate sex and ethnic TL differences in response to various types of physical exercise. This could help personalize recommendations and optimize benefits on cellular health and aging.

Our umbrella review and meta-analysis identified a small-moderate positive effect of physical exercise on TL, which seems to be influenced by the duration and type of physical exercise. In health care systems, the implementation of evidence-based physical exercise, training programs, and/or recommendations tailored to each individual might be a valuable preventive strategy for a healthy aging. More studies, with larger sample sizes, testing the impact of standardized, evidence-based physical exercise interventions on TL are warranted.