Effect of lifestyle interventions on cardiovascular risk factors among adults without impaired glucose tolerance or diabetes: A systematic review and meta-analysis

Structured lifestyle interventions can reduce diabetes incidence and cardiovascular disease (CVD) risk among persons with impaired glucose tolerance (IGT), but it is unclear whether they should be implemented among persons without IGT. We conducted a systematic review and meta-analyses to assess the effectiveness of lifestyle interventions on CVD risk among adults without IGT or diabetes. We systematically searched MEDLINE, EMBASE, CINAHL, Web of Science, the Cochrane Library, and PsychInfo databases, from inception to May 4, 2016. We selected randomized controlled trials of lifestyle interventions, involving physical activity (PA), dietary (D), or combined strategies (PA+D) with follow-up duration ≥12 months. We excluded all studies that included individuals with IGT, confirmed by 2-hours oral glucose tolerance test (75g), but included all other studies recruiting populations with different glycemic levels. We stratified studies by baseline glycemic levels: (1) low-range group with mean fasting plasma glucose (FPG) <5.5mmol/L or glycated hemoglobin (A1C) <5.5%, and (2) high-range group with FPG ≥5.5mmol/L or A1C ≥5.5%, and synthesized data using random-effects models. Primary outcomes in this review included systolic blood pressure (SBP), diastolic blood pressure (DBP), total cholesterol (TC), low density lipoprotein cholesterol (LDL-C), high density lipoprotein cholesterol (HDL-C), and triglycerides (TG). Totally 79 studies met inclusion criteria. Compared to usual care (UC), lifestyle interventions achieved significant improvements in SBP (-2.16mmHg[95%CI, -2.93, -1.39]), DBP (-1.83mmHg[-2.34, -1.31]), TC (-0.10mmol/L[-0.15, -0.05]), LDL-C (-0.09mmol/L[-0.13, -0.04]), HDL-C (0.03mmol/L[0.01, 0.04]), and TG (-0.08mmol/L[-0.14, -0.03]). Similar effects were observed among both low-and high-range study groups except for TC and TG. Similar effects also appeared in SBP and DBP categories regardless of follow-up duration. PA+D interventions had larger improvement effects on CVD risk factors than PA alone interventions. In adults without IGT or diabetes, lifestyle interventions resulted in significant improvements in SBP, DBP, TC, LDL-C, HDL-C, and TG, and might further reduce CVD risk.

Introduction Cardiovascular disease (CVD) is the number one killer globally. [1] CVD is also the major cause of morbidity and mortality among persons with diabetes, and the largest contributor to health care costs associated with diabetes. [2,3] On the other hand, CVD and diabetes share similar risk factors such as unhealthy diet, physical inactivity, and obesity. [2][3][4] Previous studies have demonstrated that structured lifestyle interventions incorporating physical activity, diet, and behavior change strategies could prevent or delay type 2 diabetes incidence and reduce CVD risk factors. [5][6][7] However, these major prevention trials focused on populations with impaired glucose tolerance (IGT). [5][6][7] Although individuals with IGT are the priority target population because they lie at the higher end of the diabetes risk spectrum, populations without IGT but with other CVD risk factors may outnumber those with high diabetes risk and have the same urgent needs for risk reduction, as many RCT studies have indicated. [8][9][10][11][12][13][14] According to the American Diabetes Association's (ADA) definitions of pre-diabetes (which includes impaired fasting glucose (IFG): 100-125mg/dL), about 60% of US individuals with pre-diabetes do not have IGT, [15] and according to the World Health Organization's (WHO) definition of intermediate hyperglycemia (measured by fasting plasma glucose (FPG): 110-139mg/dL), about 70% of individuals with this condition do not have IGT. [16] Whether lifestyle interventions should be applied more broadly to the population at lower risk (i.e. those below the IGT threshold) to reduce CVD risk needs to be examined.
According to an American Heart Association (AHA) Special Report,[17] cardiovascular health is defined by 7 metrics, including health behaviors and health indicators as follows: smoking status, body mass index (BMI), physical activity (PA) levels, healthy diet scores, total cholesterol (TC), blood pressure (BP) level, and fasting plasma glucose level. To achieve the AHA ideal cardiovascular health promotion goal, each indicator must fall into certain ranges (e.g., FPG<100 mg/dL). This definition of cardiovascular health addresses health behaviors and health indicators related to both CVD and diabetes, and thus offer guidance for how to achieve improvements in preventing both CVD and diabetes at the same time.
Evidence regarding the effects of lifestyle intervention on CVD risk reduction has previously been systematically synthesized by examining 6 of the 7 CVD health indicators mentioned above, especially by examining the different stratum of BMI (e.g., moderate weight loss will reduce both diabetes and CVD risk among overweight or obese populations [5][6][7]), as indicated by the 2013 AHA/ACC Guideline on Lifestyle Management to Reduce Cardiovascular Risk.
[18] However, how this evidence is aligned with the stratification of different glucose levels is still unclear. Lack of this information may prevent public health practitioners from fully understanding the role lifestyle interventions can play in reducing both diabetes and CVD risk among populations with varying risk levels. In contrast, a synthesis of evidence on the impact of lifestyle interventions among populations with different risk levels may help to inform decisions regarding the allocation of finite public health resources.
We conducted a systematic review to assess the aggregated impact of lifestyle interventions on glucose regulation and CVD risk factors among adults (age!18 years) without IGT or diabetes. By conducting this review, we intend to answer the following research question: can lifestyle interventions similar to those found efficacious among populations with IGT achieve the same magnitude of improvement in CVD risk reduction among populations with lower diabetes risk? We also aimed to examine whether lifestyle interventions focused on diet, PA or their combination have varying impact on CVD risk reduction. To understand how to reach the comprehensive goal of preventing both CVD and diabetes, we also examined how the lifestyle interventional effect on CVD risk reduction is related to the effect sizes of glucose improvement and weight loss.

Search strategy and selection criteria
We followed Cochrane Collaboration standards for a meta-analysis of randomized control trial (RCT) studies to develop our protocol. [19] We systemically searched MEDLINE, EMBASE, CINAHL, Web of Science, the Cochrane Library, and PsychInfo databases, from inception to May 4, 2016. Medical Subject Headings, text words, and search strategies are presented in our online-only supplements (S1 File). We examined reference lists of all included studies and relevant reviews for additional studies. We directly contacted authors to clarify data as needed.
We selected RCTs published in any language that examined lifestyle strategies involving PA and/or dietary (D) interventions, among adults (!18 years) and with glycemic indicators and CVD risk factors reported as intervention outcomes (e.g., systolic blood pressure (SBP), diastolic blood pressure (DBP), TC, low density lipoprotein cholesterol (LDL-C), high density lipoprotein cholesterol (HDL-C), or triglycerides (TG)). Included studies investigated persons without IGT or diabetes. We excluded all studies that included individuals with IGT, confirmed by 2-hours oral glucose tolerance test (75g), but included all other studies recruiting populations with different glycemic levels. However, to examine whether there was heterogeneity of effect by baseline glycemia, we grouped all studies as: (1) low range glycemia group with mean fasting plasma glucose (FPG)<5.5mmol/L or mean glycated hemoglobin (A1C)< 5.5% and (2) high range group with mean FPG!5.5mmol/L or mean A1C!5.5%. Data from the low and high range glycemic groups were analyzed separately. We only included interventions with a follow-up interval of at least 12 months.

Study selection and data extraction
Two reviewers independently reviewed each article title and abstract for inclusion. If any disagreement occurred between two reviewers, a third reviewed the item and consensus was reached through discussions.
We extracted data regarding demographic and intervention characteristics. Primary outcomes included SBP, DBP, TC, LDL-C, HDL-C, and TG. In our review, all interventions were classified as PA alone, D alone, or combined interventions (PA+D). PA interventions included any strategy used to promote physical activity levels using counseling, exercise prescription, and/or a supervised or unsupervised exercise program. D interventions included any strategy used to reduce or control calorie intake, e.g., very low-calorie diet (<800 kcal/d) or low-calorie diet (800 to 1500 kcal/d). Studies using combined PA and D strategies usually also employed behavioral modification strategies, including counseling, education, cognitive-behavioral therapy, or social support, as an intervention component.

Statistical analysis and quality assessment
We assessed study quality by examining potential selection, attrition, and detection bias. [19] We did not exclude any study that was considered poor quality (e.g., studies with attrition !30%). However, we conducted a sensitivity analysis to compare pooled effects between studies with potentially significant bias and those without. For example, for those studies with attrition !30%, their data were not used in our primary meta-analyses, but were used in our sensitivity analyses.
Among studies with similar intervention and comparison groups reporting a similar outcome of interest, we conducted meta-analyses to determine pooled effects. We calculated the mean difference between baseline and follow-up measures for the intervention (I) and comparison (C) groups (delta I and delta C) and the standard error of each difference. We used three strategies to estimate pooled effects: (1) stratified by baseline glucose levels (low range vs. high range); (2) stratified by the length of follow-up (12months vs. 13-23 months vs. !24 months); and (3) stratified by type of interventions (PA vs. D vs. PA+D).
We used DerSimonian and Laird random-effects models[20] to determine pooled effects. Effect size was defined by the mean difference between delta I and delta C divided by the standard deviation of the mean. We used meta-regression to determine whether various studylevel characteristics (mean age, follow-up interval, duration of the intervention, number of intervention contacts, attrition, and year of publication) affected the between-group differences in SBP, DBP, TC, LDL-C, HDL-C, and TG, and we examined interaction terms for all models. We also used meta-regression analyses to examine the relationship between interventional effects on CVD risk reduction and interventional effects on diabetes risk reduction measured by the effect sizes of glucose improvement and weight loss. The meta-regression was conducted using SPSS (version 20.0, Armonk, NY: IBM Corp.). We used the chi-squared test to examine heterogeneity, and we used Cochrane Review Manager software (version 5.1; Copenhagen, Denmark) to calculate pooled effects.
If a comparison group in a study used a similar approach as the intervention group did, but only differed in dose, intensity, or frequency (e.g., diet plan A vs. diet plan B; or swimming vs. walking), we analyzed the effects of treatment in a single arm model to determine withingroup changes (between post-intervention and pre-intervention in one arm) for both intervention and comparison group. These effects were also estimated by using the DerSimonian and Laird random-effect model. We did not, however, conduct any sensitivity analysis for these studies. Because this paper focused on the net lifestyle intervention effect (any lifestyle intervention vs. no intervention [e.g., usual care (UC)]), pooled effects from our single arm model are not reported in our results section, but are presented as an online supplementary table (Table C in S1 File).

Results
Seventy-nine studies [10,11,13,14, and 30 companion publications [9, encompassing 15618 participants (Table 1: range, 20 to 1089) fulfilled the inclusion criteria (Fig 1). Follow-up time ranged from 12 to 54 months. The mean age of the participants was 50.6 years (range, 30.2 to 70.4 year), and mean BMI was 30.5 kg/m 2 (range, 23.3 to 38.7 kg/m 2 ). Mean baseline SBP, DBP, TC, LDL-C, HDL-C, and TG were 127.5 mmHg, 79.2 mmHg, 5.4 mmol/L, 3.3 mmol/L. 1.3 mmol/L, and 1.5 mmol/L, respectively. More studies took place in community settings than in clinics (58 vs. 21). Sampling methods varied, but most participants were     <5.5% were classified as low range group, and 40 studies with mean FPG !5.5mmol/L or mean A1C !5.5% were classified as high range group. We observed considerable heterogeneity in the treatments provided to both intervention and comparison groups (Tables A&B in S1 File). In 29 studies, a similar approach was used in both intervention and control groups: data from these studies were synthesized by a single arm model, and are presented in Table C in S1 File as an online supplement. In the other 50 studies, UC was used in the control group. In the 50 studies that compared an intervention to UC, 38 had two arms, 5 studies[49,64,87,88,91] had 3 arms, and 7 studies[13,24,28,44,54,62,93] had 4 arms (e.g., PA, D, PA+D and control arm). The randomization procedure was described in 48 studies ( Table B in S1 File). In 29 studies, allocation concealment was adequately reported. Meta-regression analyses indicated that there was no significant interaction between the between-group change in FPG and all study-level characteristics, such as mean age, publication date, the length of F/U, number of contacts, attrition, and their interaction terms. An Egger's plot demonstrated a symmetrical shape distribution (except for two outliers) which is consistent with no publication bias.

Changes in CVD risk factors
In 57 studies or study arms comparing interventions to UC with attrition <30%, the pooled effect estimate from all studies demonstrated that compared to UC, all lifestyle interventions,  (Table 2). When including the 15 studies with attrition !30% in the sensitivity analysis, we observed similar effects. The remaining results are limited to studies with attrition <30%.

Comparison according to participant baseline glycemic level
In the 39 studies among persons with low range glycemic level, lifestyle interventions were associated with significantly improved SBP (-0.

Comparison according to intervention modality
Analyses stratified by intervention types showed that PA+D vs UC achieved the best incremental improvements in SBP (-2.

Correlation between interventional effects on CVD risk reduction and glucose change and weight loss effect sizes
Findings from meta-regression analyses demonstrated that except for LDL-C category, Pearson's correlation, r between CVD risk reduction effect sizes and glucose effect sizes ranged from 0.73 to 0.83 in SBP, DBP, TC, HDL-C, and TG, but r between CVD risk reduction effect sizes and baseline FPG were very low, only ranging from 0.26 to 0.44 in SBP, DBP, TC,   HDL-C, and TG. The r between CVD risk reduction effect sizes and weight followed the same patterns: except for LDL-C category, r between CVD risk reduction effect sizes and weight loss effect sizes ranged from 0.51 to 0.75 in SBP, DBP, TC, HDL-C, and TG, but r between CVD risk reduction effect sizes and baseline weight were very low, only ranging from 0.02 to 0.30 in SBP, DBP, TC, HDL-C, and TG. Compared to weight loss, glucose response is a better indicator of the CVD risk factor response because the glucose response has a stronger correlation with the CVD risk factor response as r ranges showed above (Table 3).

Discussion
In this review of the effectiveness of lifestyle interventions on the reduction of CVD risk factors among adults with low glycemic levels (below the IGT threshold), we found that lifestyle interventions, including physical activity, diet, and behavioral modification, can significantly improve CVD risk profiles, including SBP, DBP, TC, LDL-C, HDL-C, and TG. When stratified by glycemic levels, we found similar intervention effects between studies of participants with low vs high-range glycemic levels, except for TC and TG. Greater improvements were observed among studies with 12 months of follow-up than those with longer follow up, such that only SBP, DBP, and HDL-C improvements were sustained after 24 months. Studies that used a combined strategy of PA and D had the strongest effect on improving CVD profiles, followed by studies using D interventions only; studies only using a PA intervention strategy had the weakest effect. We have previously reported that multi-faceted interventions combining PA and D are effective in improving glucose regulation in populations with average low-range and high-range glucose levels. [125] The results of the present analyses suggest the effect of such interventions also applies to traditional biologic CVD risk factors.

Lifestyle interventions and cardiovascular risk reduction
Lifestyle interventional effects on CVD risk reduction observed in our studies among people without IGT or diabetes are consistent with those from the main trials of diabetes prevention among persons with IGT. For example, the US Diabetes Prevention Program (DPP) Study among people with IGT reported improvements in CVD profiles for all categories as measured by the mean differences between lifestyle intervention and placebo groups. The magnitude of improvements in CVD profiles in the DPP [126] in 1-year follow-up are consistent with those from our review (DPP vs this review: SBP, -2.50 vs -2.07 mmHg; DBP, -2.71 vs -1.62 mmol/L; TC, -0.06 vs -0.06 mmol/L; LDL-C, -0.02 vs -0.08 mmol/L; HDL-C, 0.01 vs 0.02 mmol/L; TG, -0.18 vs. -0.08 mmol/L, respectively). This comparison is also true for other major diabetes prevention trials (e.g., Finish Diabetes Prevention Study). [127] Our findings may have important implications for decision makers in the areas of both diabetes and CVD primary prevention. Our meta-regression analyses indicated that the magnitude of improvements in CVD risk profiles is less correlated with baseline glucose level, but highly correlated with the effect sizes of glucose improvement. Meanwhile, the meta-regression analyses also indicated that the magnitude of improvements in CVD risk profiles is less correlated with baseline body weight, but highly correlated with the effect sizes of weight loss. We thus conclude that lifestyle interventions may provide important benefits across the full distribution of glycemic levels and body weight, including populations with glycemic levels below the IGT threshold, for both the low and high ranges of baseline FPG, and for populations with normal weight but with CVD risk factors. However, economic factors as well as the effectiveness of interventions influence decisions regarding the types of interventions provided to individuals with glycemic levels below the IGT threshold. [128,129] The cost-effectiveness of lifestyle interventions that can simultaneously reduce diabetes and CVD risk among individuals with glycemic levels below the IGT threshold should be examined.
Our findings demonstrate that lifestyle interventions, compared to UC, achieved improvement in both diabetes prevention and CVD risk reduction, and these improvements were not only statistically significant, but also have clinical relevance. Previous studies indicated that each 0.03 mmol/L increase in HDL-C is associated with the reduction of coronary heart disease risk by 2-3%, [130] and each 5 mmHg reduction in SBP and 2 mmHg reduction in DBP reduce stroke risk by 13% and 11.5%, respectively. [131] According to an epidemiology study, a 1% decrease in total cholesterol leads to a decrease in the incidence of coronary events by 2%. [132] One study also indicated that weight loss improved CVD profiles because each kilogram change in body weight was related to the change in the risk of coronary heart disease by 3.1%. [133] Given that lifestyle intervention program participants in our reviewed studies usually achieved improvements in CVD across a full spectrum of outcomes simultaneously, the overall combined benefits brought by lifestyle interventions could be amplified. An estimation of overall effect on CVD risk would be helpful for our understanding the importance of interventional impact. Unfortunately, although there are several models available for CVD risk calculation (e.g., Framingham Risk Score, [134] and the ACC/AHA CVD risk calculator [135]), we are not aware of any available estimation model by which we can calculate the overall combined effect of changes of different individual risk factor. Further research and validation test, therefore, maybe needed for creating this model. If this kind model is available in the future, we can apply this model to our meta-analytic findings to estimate the overall combined effect of changes of different individual risk factor. For example, if a population, through lifestyle and behavior changes, achieved CVD risk reductions as much as showed in our meta-analyses, we can estimate the overall health benefits (e.g., how many CVD events can be prevented in the future). Despite this unavailability, the improvement in glucose regulation [125] coupled with our findings regarding the improvement in CVD risk reduction suggested that lifestyle interventions can achieve a comprehensive improvement goal as stated in AHA Special Report[17] of preventing CVD and diabetes simultaneously among persons with lower diabetes risk.
Strong evidence shows that PA programs have important independent effects on non-insulin-mediated glucose transport, markers of inflammation, insulin resistance, blood pressure, lipid profile, fitness, and improved lean-to-fat mass ratio. [136] Our findings suggest that these effects were more likely observed in studies using multi-component interventions, including PA, calorie restriction, and behavioral support but less so for PA-only interventions. This finding may be related to methodological shortcomings in exercise-only interventions such as low adherence, insufficient exercise volume or length of intervention. Previous studies suggest that it may take up to 2 years for a previously sedentary obese individual to attain enough volume of exercise to effectively reduce CVD risk factors, and individuals in unverified, out-patient interventions are less likely to engage in the prescribed amount of exercise. [137,138] However, we previously reported that exercise-only interventions in our included studies significantly reduced FPG and body weight [125] which in turn further prevented diabetes. Since PA-related improvements in glucose regulation and weight loss can lead to reductions in CVD risk profiles, potential indirect benefits should be taken into account when interpreting our findings. Lifestyle interventions and cardiovascular risk reduction Unhealthy lifestyle factors are related to the atherosclerotic process and these long-term exposures lead to the clinical manifestations of cardiovascular events. [139] A previous study also indicated that lifestyle changes, only in the long-term, are likely to lead to CVD risk factor reduction.
[30] Our findings demonstrate that the effects of lifestyle changes on the reduction in CVD risk factors reached their highest point at 12 months of follow-up, then gradually decreased over time. This may reflect the fact that the longer-term intervention may be more effective on reducing CVD risks only if participates remain highly adherent to the intended interventions, which is seldom observed. It could be also true that using CVD mortality, rather than CVD risk reduction alone, to measure the long-term effect of lifestyle changes on CVD is more appropriate as the extended legacy findings of the Chinese Da Qing Study indicated. [140] Because we used a comprehensive search strategy including all major medical databases, we found a large number of eligible studies. Pooled effects based on a large sample size provide more robust findings than those from any single study. Our review has some limitations as well. First, lifestyle interventions were used in heterogeneous settings, among different populations of varying ages, health status, and race/ethnicity background. While the main components of the lifestyle interventions were generally PA and D, each of the strategies had its own requirements in type, dose, intensity, and frequency. UC also had varying definitions among different comparison groups. Heterogeneity across studies was also reflected in the length of intervention, duration and follow-up, and number of sessions. However, our meta-regression analyses found no interactions between the between-group change in glycemic indicators and study-level characteristics. We also stratified our data syntheses by glycemic level, length of Lifestyle interventions and cardiovascular risk reduction follow-up, and type of interventions, taking the heterogeneity among included studies into account. Second, although we stratified by level of glycemic risk at the study level, there was considerable heterogeneity within studies, and the nature of aggregated data prevented individual level classification by glucose level. As a result, there was likely considerable overlap in participant characteristics between low range and high range glycemic groups in our study, which may introduce some misclassification bias. Misclassification bias could be also  Lifestyle interventions and cardiovascular risk reduction introduced by usage of both FPG and A1C in our review to identify population with low glycemic risks. Although a previous study indicated that the agreement between FPG and A1C is high, [141] they are not equal with each other. [142] Because of this misclassification bias, some individuals identified as with low glycemic risks could actually have glucose metabolism abnormalities. Audiences need to be cautious when interpreting our findings.

Conclusions
Our review is the first comprehensive examination of the impact of lifestyle interventions on risk for progression of dysglycemia and CVD risk reduction among persons below the IGT threshold. This systematic review suggests that lifestyle change is critical to both CVD risk reduction and diabetes prevention across the full spectrum of risk, complementing the major trials of diabetes prevention that focused on persons with IGT. This review also provides supportive evidences for designing strategies aimed at reducing CVD burden as delineated in the AHA Strategic Impact Goal through 2020 and Beyond.
[17] Our findings demonstrated that among adults without IGT or diabetes, PA and D interventions, especially combined can significantly improve SBP, DBP, TC, LHL-C, HDL-C, and TG, in addition to glucose regulation and weight loss, and that these risk reductions may further prevent CVD events.