Physical activity, sedentary behavior, and long-term cardiovascular risk in young people: A review and discussion of methodology in prospective studies.

The long-term effects of physical activity (PA) or sedentary behavior on cardiovascular health in young people are not well understood. In this study, we use a narrative format to review the evidence for a prospective association with adiposity and other well-established biological cardiovascular risk factors in healthy young people, considering only studies with at least 2 years of follow-up. PA appears to elicit a long-term beneficial effect on adiposity and particularly markers of cardiovascular health. With adiposity, however, a few studies also reported that higher levels of PA were associated with higher levels of adiposity. Time spent sedentary does not appear to be related to adiposity or markers of cardiovascular health independent of PA. We then discuss the uncertainties in the underlying causal chain and consider a number of alternative modeling strategies, which could improve our understanding of the relationship in future studies. Finally, we consider the current methodology for assessing PA and sedentary time.

Introduction

Cardiovascular disease (CVD) continues to be the number one cause of death and disability in the world.1, 2 While CVDs such as coronary heart disease and stroke manifest clinically in middle age or older adulthood, their origins begin much earlier. Exposures acting during growth and maturation may exert long-term effects on cardiovascular physiology and have a major impact on the development of CVD. For example, evidence shows that alterations in fetal environment such as exposure to maternal smoking and nutritional status, childhood socioeconomic position and overweight independently predispose to development of CVD in adulthood.4, 5, 6 Similarly, lack of physical activity (PA) during childhood and youth may lead to cardiovascular disturbances and progression of atherosclerosis that could contribute to development of CVD in adulthood. Indeed, organizations and governments have identified a population-wide promotion of healthy lifestyle including PA in young people as a key priority for primordial prevention of CVD. A major advance in population-based studies of the influence of childhood or youth PA on long-term cardiovascular health outcomes has been the application of objective methods to assess PA that prevents bias related to recall and social desirability. This has also facilitated the possibility to obtain detailed information on frequency, duration, and intensity of PA.

Two earlier reviews, published in 2010 and 2011, respectively, systematically addressed studies on the observational associations between objectively assessed PA and adiposity outcomes.9, 10 These reviews highlighted the importance of distinguishing between cross-sectional and prospective studies as Jimenez-Pavon and colleagues found a negative associationin 32 of 41 cross-sectional studies (78%), while Wilks and colleagues, who only considered prospective studies, found a negative association in 4 of 10 studies (40%). Because cross-sectional studies have a number of limitations, including inability to infer causality, this inconsistency is important to address by considering recently published experimental and prospective studies. In 2014, Tanaka and colleagues published a review on the prospective association between changes in objectively measured sedentary behavior (SED) and measures of adiposity. However, by imposing this restriction the authors were only able to include 3 studies. The review found evidence for a positive association in 1 of the 3 studies. In summary, the previous evidence synthesis did not clearly support a prospective association between PA/SED and adiposity outcomes. In contrast to these reviews on adiposity outcomes, we are aware of no published reviews on the prospective association between PA and biological risk factors in young people.

This narrative review aims to give an overview of the evidence from population-based studies relating objectively assessed PA and SED in childhood or youth with long-term (≥2 years) cardiovascular risk factor progression. In addition, we will discuss the current methodological challenges and future direction in the objective assessment of PA in large-scale studies following young people over time.

Scope of the review

In our discussion we will consider PA as a behavior separately from SED. This is because a body of literature has identified behaviors such as TV-viewing12, 13 and total sitting time to predict all-cause and cardiovascular mortality in adults, independent of other domains of PA. Our discussion will consider observational studies which

Used an objective measure to quantify whole-day PA (i.e., total activity, light PA (LPA), moderate PA (MPA), moderate-to-vigorous PA (MVPA), or vigorous PA (VPA)) or SED at baseline.

Included healthy, population-based samples of children and adolescents aged ≤18 years at baseline and followed the same individuals for a period of ≥2 years.

Related an exposure to any form of the conventional biological risk factors (recommended for use in CVD risk stratification among asymptomatic adults) with the addition of indices of insulin-resistance. We considered all forms of adiposity outcomes.

Were published after October 29, 2009, as this was the final search date in the review by Wilks and colleagues (applies for PA–adiposity investigations only).

We identified relevant studies from our records by conducting a search on PubMed using combinations of the relevant exposures and outcomes and by going through the reference lists of the identified studies. As there are a greater number of studies on adiposity outcomes than studies on biological risk factors, we will address these in separate paragraphs. Furthermore, we will briefly discuss findings from controlled or randomized controlled intervention studies conducted in a general (young) population.

How do PA and SED associate with adiposity and biological risk factors?

PA and adiposity

We identified a total of 13 published studies which fulfilled our criteria.15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27 The duration of follow-up ranged from 2 to 7 years with 9 studies covering ≤3 years.15, 16, 17, 18, 22, 23, 24, 25, 27 The studies included samples aged 4–18 years at baseline. The median (25th–75th quartiles) study size was 554 (315–984). Twelve studies used accelerometry as their exposure variable, while 1 used a combination of accelerometry and heart rate. The outcome variables were more heterogeneous. Nine studies (69%)17, 18, 19, 20, 22, 23, 24, 26, 28 reported a significant association between PA and a measure of adiposity such that a higher level of PA was associated with a lower level of adiposity (negative association). Surprisingly, 3 studies reported significant associations in the opposite direction.16, 20, 23 When examining the relative importance of specific PA intensities, significant negative associations were apparent with LPA in 0 of 3 studies (0%), with MPA in 1 of 4 studies (25%) and with VPA in 2 of 4 studies (50%).17, 24 Finally, significant negative associations with adiposity were found for MVPA in 7 of 8 studies (87%).18, 19, 20, 23, 24, 25, 26 Four studies included SED as a covariate, which generally had a small effect on the estimates and only affected the conclusion in 1 study. The choice of outcome measure in adiposity studies appeared of importance as all studies (4 in total) that considered MVPA and used a reference method for assessing adiposity such as DXA or densitometry reported a negative association with adiposity.18, 20, 23, 26 In contrast, 6 of the MVPA studies used a body mass index (BMI)-based measure with significant negative associations to adiposity found in 3 studies, while 2 studies found significant positive associations. The latter appears to be explained by an increase in muscle or bone tissue as both of these studies also reported significant associations between PA and fat free mass. If PA associates with fat free mass, which is suggested in the literature,30, 31 future studies should use methods which are accurate enough to separate fat mass from lean mass.

SED and adiposity

We identified a total of 9 published studies which fulfilled our criteria.15, 20, 21, 24, 25, 26, 27, 32, 33 The duration of follow-up ranged from 2 to 7 years with 5 studies having ≤2.5 years of follow-up.15, 24, 25, 27, 32 The studies included youth 4–18 years at baseline. The median (25th–75th quartiles) study size was 554 (403–984). Eight studies used accelerometry, while 1 used a combination of accelerometry and heart rate. The measures used for outcomes were more heterogeneous. Two studies (22%)25, 33 reported a significant association between SED and adiposity; more time spent on SED was associated with a higher level of adiposity. One study found an association in the opposite direction, i.e., higher SED associated with lower adiposity. Of the 2 studies reporting a positive association, 1 used a relatively high cut-point to define SED (<1100 counts/min (CPM)), which is very likely also to include time spent on LPA. This finding is surprising as none of the 3 studies examining LPA separately from SED found an association and 1 of these even found a negative association.16, 17, 27 Six studies used a cut-point of <100 CPM to define SED,15, 21, 24, 26, 27, 33 which is comparable to the cut-point of 1.5 METs used in the combination study. In order to determine the independent association between SED and adiposity, it is important to adjust for other dimensions of PA. This was performed in 7 studies by adjusting for MVPA. The isolated effect of this adjustment was, however, not possible to infer, as either the studies did not report the estimates that were not adjusted for MVPA or the studies adjusted for other covariates when including the adjustment for activity.

PA, SED, and biological risk factors

We identified a total of 8 published studies which fulfilled our criteria for PA,17, 20, 35, 36, 37, 38, 39, 40 while 1 study used SED. The study using SED found no significant associations with any of the biological risk factors before or after adjustment for MVPA. The duration of follow-up in the PA studies ranged from 3 to 12 years with 4 studies having ≤3 years of follow-up. Included participants were 5–16 years at baseline. The median (25th–75th quartiles) study size was 293 (209–466). Seven studies used accelerometry and 1 used pedometers. Five studies used a composite risk score20, 35, 37, 38, 39 and 6 studies provided results for individual risk factors. Four of 5 studies (80%)20, 37, 38, 39 reported a significant negative association between PA and the composite score, meaning that a higher level of PA was associated with a more favorable cardiovascular profile. Results for the individual risk factors were generally coherent with 4 of 5 studies reporting a significant negative association with insulin resistance,20, 36, 38, 40 0 of 3 with glucose, 2 of 3 with triglyceride37, 38 and 3 of 4 with blood-pressure.17, 37, 38 Two of 3 studies reported significant positive associations with high-density lipoprotein cholesterol (HDL-C).20, 38 In contrast to studies on adiposity outcomes, studies on biological risk factors included little reporting of specific PA intensities. Only 3 studies gave results for different intensities.17, 36, 39 Two of these studies demonstrated that VPA, but not lower intensities, was significantly associated with the risk factors17, 39 while 1 study showed similar associations for total activity and MVPA. MVPA was used in 5 studies with 4 reporting significant associations.20, 36, 37, 38

Intervention studies

Dobbins et al. published a Cochrane-review in 2013 which included a total of 44 randomized-controlled trials with interventions ranging from 12 weeks to 6 years. They found little evidence for an effect on blood-pressure or mean cholesterol. However, another review found that studies in general showed a significant, but small, effect of PA on HDL-C and triglycerides, but no effect on total cholesterol or low-density lipoprotein cholesterol (LDL-C). A review synthesizing effects on PA, fitness, and motor skills concluded that strong interventions carried out by educated physical education (PE) teachers and including around 5 PE lessons per week are needed to improve these parameters. In the Sogndal school-intervention study, the implementation of the intervention was carefully controlled and included 60 min PA on all school days, and part of the activity was of high intensity. They found an improvement in cardiovascular fitness of 15% in the least fit quartile compared with the similar quartile in controls. This suggests that the intervention was intensive and in that study, nearly all the biological risk factors also improved. The intervention in Sogndal was designed based on findings from the Copenhagen School Child Intervention Study wherein an increase in PE lessons from 2 to 4 lessons a week revealed minimal or no changes in risk factors. Later, the Childhood Health, Activity, and Motor Performance School Study Denmark (the CHAMPS-study DK) tested a PE dose of 6 h per week and found improvements in a composite cardiovascular risk score. In Iceland, Hrafnkelsson et al. conducted an intervention wherein PA was gradually increased from 30 to 60 min per school day over 2 years. They did not find improvements in any risk factors, but the study was quite small with only 3 intervention schools included in the cluster randomization. Moreover, accelerometer measurements showed only 4 min more MVPA in the intervention group in the first year and no difference in the second year when compared to the control group.

Perspectives

An important question when addressing the relationship between PA and biological risk factors is whether the associations are independent or mediated by adiposity.48, 49 This issue is important for understanding the etiology of cardiometabolic risk and is crucial for public health efforts as an independent effect would mean that an increase in PA would reduce CVD risk, even though no change in adiposity was achieved. Cross-sectional literature suggests that adjusting for adiposity attenuates the association slightly, indicating that the effect of PA on biological risk factors is not entirely explained by adiposity.50, 51, 52, 53 Our review included 3 studies which adjusted their models for changes in adiposity,36, 54 with the effect of this adjustment only reported in 1. Metcalf and colleagues reported that adjusting for change in BMI or waist-circumference had virtually no influence on their estimates. Further, 3 studies included an adiposity measure in their composite score,35, 38, 39 which highlights the complexity of this issue. By doing so, adiposity is treated equal to the biological risk factors, but if adiposity is actually the cause of these, this could drive the associations. A promising methodological advancement is the emergence of mediation analysis and structured equation models. These use 2-stage regression to decompose associations into, e.g., total, direct, and indirect effects. With these more formal quantifications of mediation, it may be possible to assess the relative importance of a putative indirect effect in relation to the direct effect. If the indirect effect is small, from a public health perspective, it is of little importance whether adiposity is a mediator or a confounder. However, more work is needed to improve the methods and address issues such as varying degrees of measurement error in exposure/mediator variables and reverse causation, which are not resolved by using these methods. Hence, these limitations should be considered when interpreting the models.

The reviewed studies considered numerous different models to investigate the prospective associations. A frequently used model was the “determinant”-model. Here, a follow-up outcome or a change in outcome (the two are equivalent when adjusted for baseline outcome) was regressed on a baseline exposure. However, not all studies adjusted for baseline outcome, which is important as the size and direction of change from baseline to follow-up are expected to be associated with the baseline level of the outcome because of regression to the mean and flooring/ceiling effects. The potential impact of not adjusting for baseline outcome is demonstrated in 1 study where the direction of the association was inverted following this adjustment. Another frequent model was the “change”-model in which the absolute change in outcome over time was modeled on the absolute change in exposure. In this model, bias may arise by not adjusting for baseline values of both exposure and outcome. In the reviewed studies, only one considered both in their models of absolute change. As some studies adjusted for baseline exposure, some adjusted for baseline outcome and some adjusted for neither, this may be the cause of some of the heterogeneity in the study results. The change-model, however, has been criticized for being a “masked” cross-sectional study as it could be the outcome which changed first, again making inference of the causal pathway difficult. To better establish the temporal relationship between PA and cardiovascular risk factors, future studies could attempt to collect (and analyze) repeated measurements of exposures and outcome during follow-up, which enables the possibility to specify a time lag between PA and the change in risk factor. Further, with repeated measurements, a structural equation modeling approach could be used to assess the importance of early life exposure versus change in that exposure over time, while taking into account the correlations between time-points. Interpretation of results was further challenged in several studies as an isotemporal substitution model (ISM) was used, but only one mentioned this interpretation in their reporting. An ISM may arise when multiple intensity domains and total wear time is included in the same model. The interpretation of an ISM would be the effect of changing a unit of time of, e.g., LPA with the same unit of time of MVPA. This is not the same as that of the more frequently used partition model. We will not favor one model over the other but will recommend that future studies clearly state their models and their interpretation of these. Finally, even though 4 of our reviewed studies considered some form of adjustment for dietary factors, finer control of measures of diet quality and quantity would also help improve future observational studies.

The use of accelerometry in the assessment of PA and SED in prospective studies is encouraged by its feasibility and higher validity than self-report. Nevertheless several shortcomings with accelerometers have been identified and include both technical and reactivity issues and controversies regarding how best to define non-wear and various intensity cutoff-points. The technical shortcoming of the accelerometer is the misrepresentation of important health-related activities like cycling, strength training, static exercise, aquatic activities and the intensity leveling-off with increased running speed.62, 63 Misrepresentation of beneficial PAs with moderate and vigorous intensity could explain some of the diversity of the prospective PA and adiposity associations reported. The assessment of SED using the <100 CPM cutoff point has been shown to have an acceptable sensitivity (71.7%) and specificity (67.8%) to estimate sitting time, with no optimal cutoff points for sitting plus standing. Thus, the SED estimation in the study by Basterfield and colleagues included activities with higher intensity than standing and suggests a measure that is far from the concept of SED defined by the posture allocation of sitting. Most studies investigating the association of SED with adiposity did not show any significant association, and this could partly be explained by the sensitivity and specificity of the cutoff point. The definition of the non-wear is also problematic. The definition used across the selected studies is quite different, and it has been shown that various non-wear definitions generate substantially different results with a significant impact on the assessment of SED. The comparison across different studies seems to require a consensus and harmonization on what definitions of cutoff points should be used and non-wear handled. Additionally, more research is needed to investigate the shortcomings of the accelerometer and how new solutions can be feasibly applied in large-scale studies.

In summary, a beneficial effect of PA on long-term change in adiposity in young people was generally observed, but some studies found the opposite effect. This may be explained by the marker of adiposity used. In observational as well as intervention studies among children or adolescents, PA had a protective association with established biological risk factors. These findings suggest that providing the right opportunities for children and adolescents to engage in PA is important for primordial prevention of CVD. Sedentary behavior did not appear to increase long-term adiposity independent of other activity domains and almost no evidence was available with biological risk factors for this exposure. Future studies should attempt to better decompose the causal chain between PA and CVD risk factors, and there is a need for improved reporting and interpretation of statistical analyses, harmonization of data-reduction and development of methods to distinguish between different SED behaviors.

Authors' contributions

JT identified relevant studies, extracted the information, and drafted the manuscript; JCB, LBA, NCM, KF, and AG identified relevant studies and drafted the manuscript. All authors have read and approved the final version of the manuscript and agree with the order of presentation of the authors.

Competing interests

None of the authors declare competing financial interests.