Long-term Effect of Lifestyle Interventions on the Cardiovascular and All-Cause Mortality of Subjects With Prediabetes and Type 2 Diabetes: A Systematic Review and Meta-analysis

Type 2 diabetes is a major public health problem due to costs related to its management, reduced productivity, early incapacity for work (1,2), propensity for cardiovascular complications, and the consequences associated with increase in premature death. Prediabetes is normally considered as a risk factor for diabetes, but it also has a direct association with the development of cardiovascular complications. Prediabetes, as well as diabetes itself, is closely tied to the obesity pandemics and the overall weight gain in the general population in recent decades (3–5). Several studies have shown that lifestyle interventions, with intensive medical nutrition therapy and recommendations on a healthy diet and increased physical activity, improve cardiovascular disease risk factors, such as hypertension and dyslipidemia, in those with prediabetes and type 2 diabetes (6,7), and lifestyle interventions have been shown to be cost-effective (8).

However, even though the posttrial analysis of the Da Qing Diabetes Prevention Outcome Study (Da Qing DPOS) in subjects with prediabetes demonstrated a reduction in cardiovascular and all-cause mortality (9), the more recent follow-up data of the Diabetes Prevention Program Outcomes Study (DPPOS) (10) and Look AHEAD (Action for Health in Diabetes) did not show protection of lifestyle interventions against these outcomes (11). Considering that in clinical trials lifestyle intervention with medical nutrition therapy and physical activity was demonstrated to be effective in preventing type 2 diabetes and reducing risk factors for cardiovascular disease but without clear benefit in reducing mortality, we conducted a broad systematic review and meta-analysis of randomized clinical trials (RCTs) to evaluate the effect of long-term lifestyle interventions compared with standard care on the incidence of cardiovascular and all-cause mortality in subjects with prediabetes and type 2 diabetes.

This systematic review with meta-analysis was conducted according to the Cochrane Handbook for Systematic Reviews of Interventions (12) and reported according to the updated Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Checklist. Based on the published protocol (PROSPERO, CRD42021207431 [www.crd.york.ac.uk/prospero/display_record.php?RecordID=207431]), the search was conducted by K.P.Z. in MEDLINE, Cochrane Central Register of Controlled Trials (CENTRAL), Embase, and Web of Science until 15 May 2022 (Supplementary Table 1), with no language or date restriction. We also searched through trial protocol records (e.g., ClinicalTrials.gov and ISRCTN registry) and unpublished trials, reference lists of relevant publications, and unpublished data by contacting the authors of published and unpublished trials.

We selected RCTs including adults with criteria for prediabetes or type 2 diabetes (13,14) with comparison of lifestyle interventions (dietary and physical exercise recommendation) with active control (usual care or standard advice). If studies had more than one treatment group, we included in the meta-analysis the most intensive lifestyle intervention and as the control the group with minimum intervention (usual care, standard advice, or placebo). For a study that had pharmacological treatment as one arm, we considered as control the group with a placebo and standard advice (10). RCTs had to include provision of a dietary prescription and/or group-based structured program recommendations for lifestyle intervention with diet as the main treatment intervention to be included in our analysis. Moreover, data had to be reported from RCTs regarding cardiovascular death or all-cause mortality with at least 24 months of active interventions. There was no minimum for posttrial follow-up.

After removal of all duplicates, a pilot training was conducted to standardize the eligibility criteria among the reviewers, and five pairs of trained reviewers independently screened the titles and abstracts and evaluated the full texts of the selected studies using EndNote reference manager (version X7.17; Thomas Reuters, New York, NY). Disagreements were resolved by a third reviewer (F.G.).

The same pairs of independent reviewers extracted the data of all the included studies in duplicate using a prepiloted form. Information retrieved included first author, publication year, country, study design (RCT with or without posttrial), population, sample size, and a description of the intervention and control groups as well as primary and secondary outcomes and main results (participant characteristics, dropouts, mean follow-up, number of events). Cases of discrepancies in the data extraction were resolved by discussion and consensus, involving a third review author (F.G.) if necessary. When there was uncertainty about data, we contacted the authors of the studies.

As we included large trials with long-term follow-up periods, for some more than one article was published within the time point of interest (e.g., reporting of outcomes at 3 years and 10 years postintervention). In those cases, for the primary analysis, we selected the most recent article with relevant outcomes reported within the time point of interest.

Two independent reviewers (K.P.Z. and P.P.T.) critically appraised the included studies using the Cochrane tool to evaluate risk of bias (RoB 2.0) (15). The risk of bias was assessed according to the following domains: 1) risk of bias arising from the randomization process, 2) risk of bias due to deviations from the intended interventions, 3) risk of bias due to missing outcome data, 4) risk of bias in the measurement of the outcome, and 5) risk of bias in the selection of the reported result. Each of the domains was classified as “high risk of bias,” “low risk of bias,” or “some concerns.” The overall risk of bias of each study was judged as follows: “low risk of bias” if all domains had low risk of bias, “some concerns” if at least one domain raised some concerns but there was no domain with high risk of bias, and “high risk of bias” if at least one domain had high risk of bias or if multiple domains raised some concerns (15).

The overall quality of evidence was assessed with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework (16) by the same two reviewers, wherein RCTs received an initial grade of high by default and were downgraded based on prespecified criteria—risk of bias (assessed with the RoB 2.0), inconsistency (substantial unexplained interstudy heterogeneity), indirectness (presence of factors that limit the generalizability of the results), imprecision (consider the total number of events or patients and whether the 95% CI for effect estimates includes appreciable benefit or harm), and publication bias (significant evidence of a small study effect)—or upgraded (16).

We performed a descriptive synthesis of all included studies. A frequentist and arm-based pairwise meta-analysis was conducted using random-effect models and the DerSimonian and Laird method as a variance estimator (17). We calculated pooled effect sizes using the Mantel-Haenszel method for binary outcomes. The effects of randomized treatment on the cumulative number of events were estimated as relative risks (RRs) during the in-trial and overall follow-up periods in accordance with the intention-to-treat principle. RRs and 95% CIs were estimated from an approximate normal model. Considering that 5 of the 11 included studies had a small number of or zero events in either the intervention or control arms and that the number of events was >1%, we conducted additional analyses via a GLMM (binomial normal model), estimating RRs as effect measure to confirm our findings when necessary (12,18). We planned for assessment of the potential for publication bias with Egger test and graphically reporting with the funnel plot (if at least 10 studies were included in the quantitative synthesis). A P value <0.05 was considered statistically significant, and meta-analyses were conducted in R statistical software, version 1.4.17, with package meta v.4.18-2. The variability across trials attributable to heterogeneity was quantified with the I² statistic, with 25%, 50%, and 75% as cutoff points for low, moderate, and high heterogeneity among trial results (12).

We planned to perform the subgroup analysis by glycemic status (prediabetes or type 2 diabetes), sex (male or female), age (adult or elderly, based on the cutoff point of 60 years old), geographical region (Europe, Asia, or North America), mean duration of follow-up period (more or less than 15.8 and 11 years for cardiovascular and all-cause mortality, respectively), type of dietary (advice or prescription) and physical exercise (advice or prescription) intervention, and dietary and physical exercise advice or recommendations applied to the control groups (no advice, standard advice, or usual care according to each center). To consider the possible impact of intervention dilution over time and the difficulty in sustaining an intervention effect, we also performed a subgroup analysis stratified according to the follow-up period: 2–5, 6–15 years, and 16–30 years. The Da Qing DPOS analysis with data on 6 years of follow-up (19) was included in this subgroup analysis, and it was included as an additional article in the PRISMA flowchart.

Sensitivity analyses excluding studies without reporting of deaths as main outcomes were conducted for both outcomes. For cardiovascular death, we performed a sensitivity analysis excluding a study with “some concerns” in overall risk of bias. A meta-regression analysis was performed to determine the relationship between mean weight changes (in kilograms) from baseline (intervention compared with control) and the RR of mortality. For this analysis, the greatest mean difference in body weight over the entire study period was used for each study. The Japan Diabetes Complications Study (JDCS) was excluded from the analysis as its results were presented as BMI (20). Significant correlation was considered if the P value was <0.05. Given the nature of secondary data capture and analysis, the patients and the public were not involved in the design or interpretation of this study. The results of this review will be disseminated to appropriate audiences with proper caution.

From 23,335 records after the removal of duplicates, we assessed 490 records for full-text analysis. Finally, we included in this review a total of 11 studies (as shown in Fig. 1). The list of the reasons for exclusion in each phase is available in Supplementary Table 2.

As described in Table 1, 11 parallel RCTs were included resulting in a total of 16,574 subjects. Seven studies (63.6%) included subjects with prediabetes (9,10,21–25) and four (36.3%) with diabetes (11,20,26,27)—all but one (20) with a mean BMI of overweight and/or obesity. The mean duration of intervention was 4.25 years, and the total follow-up ranged from 2 to 30 years. Five studies had posttrial follow-up analyses (9–11,21,26), and the majority of them (45.4%) were from Europe (21–23,25–27).

In all trials, lifestyle interventions combined diet and physical exercise recommendations. Three (27%) studies consisted entirely of group sessions with educational activities and materials about a healthy diet, glycemic control, the importance of exercise, and behavioral approaches (24–26). The other eight studies (73%) consisted of a dietary prescription, calculated as a caloric deficit to reach at least 5% weight loss (9–11,21–23,27) or individual advice (20). Only one study recommended a low-fat diet instead of a traditional macronutrient distribution (10), three studies had partial or total diet replacement in a period of intervention (11,21,27), and one study also had a very-low-calorie diet as an alternative for weight loss (21). Five studies included recommendations to increase the time, intensity, and frequency of physical activity (20,21,23,24,26), and six studies included a specific exercise prescription, which varied, from 100 to 210 min of moderate exercise per week, increasing number of steps, or 20–30 min of aerobic activity (9–11,22,25,27). Three studies offered supervised exercise sessions (10,21,25), and in one study participants were given a CiTY CARD to exercise (22).

Regarding pharmacological treatment, only one study had metformin as one treatment arm, with other medications that could affect study outcomes discouraged (10). Two studies offered weight loss medication in the intervention group if target weight loss was not achieved (11,27). In one study oral hypoglycemic agents and antihypertensive and diuretic drugs were withdrawn at the beginning of intervention and reintroduced when necessary (27). One study excluded participants receiving long-term oral corticosteroids or weight loss medications (23). All studies reported adjustments in dosages of medications.

The control groups had exposure to a lower amount of lifestyle interventions combining less intensive diet and physical exercise recommendations or no specific intervention. In three studies general information was provided about the management of type 2 diabetes, with no specific advice about diet and physical exercise (9,22,25). In five studies subjects were offered general recommendations about diet and physical exercise according to guidelines (10,11,21,23,24); however, some of them encouraged weight loss through a combination of diet and physical exercise, with some degree of lifestyle intervention (10,23,24). The other three studies had usual diabetes and obesity care according to the recommendations applicable in each center (20,26,27).

Regarding the posttrial follow-up period, three studies (27.3%) included an observational phase with periodic face-to-face evaluations and telephone contacts to collect data on clinical outcomes (9,21,26) and two studies continued with lifestyle interventions (10,11). However, both included a lower-intensity intervention phase, with lifestyle reinforcement through refresher group sessions semiannually (11) or telephone contact (10). Four trials reported evaluations of adherence to dietary prescription through 3-day food records (9,10,22,27), and three of them reported the percentage of adherent individuals (9,10,22). The other articles did not include data on adherence to the prescribed diet (11,20,21,23–27). Moreover, for all studies adherence to the lifestyle interventions was categorized as concluding the study follow-up period or attending all visits, and in one study the number of predefined intervention goals achieved was calculated, with a score ranging from 0 to 5 (21).

The analysis of the risk of bias for all-cause mortality is shown in Supplementary Fig. 1. Overall, one study presented risk of bias on the randomization process (21) and four presented deviations from intended interventions (21–25). For one study, data were not shown on mortality, which was one of its main outcomes (25). No studies were rated as having a high risk of bias. For cardiovascular death, the risk of bias is shown in Supplementary Fig. 2. Only one study presented some concerns due to deviations from the intended interventions (22).

Four studies assessed all-cause mortality as the primary outcome, and data were collected through national databases, medical and hospital records, telephone calls, death certificates, and postmortem reports (9–11,26). Seven studies showed the number of deaths as losses to follow-up (20–25,27), and of these, just one reported inspection of electronic records for data abstention about losses by mortality (27).

For this outcome, a total of 16,554 subjects were included in the analysis (55.5% women, mean age 55.5 years, mean BMI 30.5 kg/m²). The total incidence of all-cause mortality was 13.83%, with 1,205 deaths in the intervention group and 1,085 in the control group. As shown in Fig. 2, lifestyle interventions were not superior to usual care in reducing all-cause mortality (RR 0.93; 95% CI 0.85–1.03; I² = 15%), with an absolute and nonsignificant risk reduction in all-cause mortality of 10 fewer events per 1,000 (95% CI −21 to 4) in the lifestyle intervention group. This result was confirmed by general linear mixed-effects model analysis (RR 0.95; 95% CI 0.87–1.03). The quality of evidence according to the GRADE system was moderate due to imprecision as the CI includes important benefits and harms (Supplementary Table 3).

In a sensitivity analysis for all-cause mortality excluding studies where this outcome was reported as losses to follow-up, there was still no significant difference between lifestyle interventions and usual care (RR 0.92; 95% CI 0.81–1.05) (Supplementary Fig. 3). We also performed a subgroup analysis by glycemic status, geographic location, age, type of dietary and physical exercise intervention, a period follow-up of studies, and the risk of bias of included studies, and it did not change the results (Supplementary Figs. 3–10). The subgroup analysis according to dietary and physical exercise recommendations applied to the control groups showed superiority of lifestyle intervention in reducing all-cause mortality when compared with “no advice about diet and exercise,” with results heavily driven by the Da Qing DPOS (Supplementary Fig. 11). The results of the subgroup analyses obtained with the generalized linear mixed-effects model are shown in Supplementary Table 4, and there was no change in the direction of the results. The duration of trials and the posttrial follow-up did not appear to favor longer or shorter studies (Supplementary Fig. 12). Meta-regression analysis showed no relationship between mean weight change and the RR of all-cause mortality (P = 0.513) (Supplementary Fig. 13).

As estimates were generally symmetrically distributed in funnel plots, publication bias is unlikely (Egger test intercept = −0.0198; P = 0.0959) (Supplementary Fig. 14).

In the analysis of cardiovascular mortality, 11,017 subjects were included (51.9% women, mean age 54.5 years, mean BMI 31.8 kg/m²), and the total incidence of outcome was 5.69%. The studies’ characteristics are described in Table 1. As shown in Fig. 3, lifestyle interventions were not superior to usual care in protection against cardiovascular mortality (RR 0.99; 95% CI 0.79–1.23; I² = 38%), with an absolute and no significant risk reduction of one less cardiovascular mortality event per 1,000 (95% CI −11 to 12) and moderate certainty of evidence according to the GRADE system (Supplementary Table 3). This result was confirmed using the general linear mixed-effects model analysis (RR 1.01; 95% CI 0.86–1.18). Publication bias could not be quantitatively evaluated due to the number of studies included in meta-analyses (n < 10).

A sensitivity analysis, excluding studies reporting cardiovascular death as losses to follow-up, showed similar results (RR 0.98; 95% CI 0.78–1.24, I² = 53%) (Supplementary Fig. 15). The sensitivity analysis by risk of bias, excluding those categorized as “some concerns,” also showed no changes in the result (Supplementary Fig. 16). The subgroup analyses showed that in grouping studies according to glycemic status, age, and type of dietary and physical exercise recommendations, and a period follow-up of studies, the results did not change (Supplementary Figs. 17–21). Finally, the Da Qing DPOS was the only study showing that lifestyle intervention had a protective effect for cardiovascular death in a geographic location (RR 0.70; 95% CI 0.51–0.97) and in the control group (RR 0.70; 95% CI 0.52–0.98) analysis, available in Supplementary Figs. 22 and 23, respectively. Meta-regression analysis showed no relationship between the mean weight change and the RR of cardiovascular mortality (P = 0.4491) (Supplementary Fig. 24). All subgroup analyses with use of the GLMM are shown in Supplementary Table 4, and there was no change in the direction of the results.

The results of this systematic review and meta-analysis of RCTs demonstrated that lifestyle interventions implemented thus far focused on long-term dietary and physical exercise recommendations were not superior to standard care in reducing cardiovascular and all-cause mortality in populations with prediabetes and type 2 diabetes.

Different protocols of lifestyle intervention in subjects with prediabetes and type 2 diabetes have been performed. Observational data suggest significant reductions in all-cause mortality (28); however, RCTs, in general, failed to confirm such benefits (10,11,26). A possible explanation is that this inconsistency might be driven by the different answers derived from observational studies in comparison with clinical trials on the same clinical questions, the first overestimating treatment effects and the second not having sufficient events and sample size to identify potential effects (29). Moreover, trial populations tend to be healthier than the general population. The healthy adherer effect is also a known bias in observational studies as individuals who engage in physical exercise and healthy eating patterns will generally be healthier and more likely to adhere to several other recommendations (30), restricting the generalizability of the results. In the present review, all but one (9) trial demonstrated no superiority of lifestyle interventions in reducing cardiovascular and all-cause mortality, showing a consistency in results and advancing the findings of past work (31) including the reports of the available trials with longer (≥24 months) intervention periods, larger samples, and longer follow-ups as well as, for the first time, with studies originally designed to analyze mortality outcomes.

Interestingly, the only positive trial was the Da Qing DPOS (9), which demonstrated a reduction in cardiovascular outcomes after more than two decades following the end of the intervention in a younger sample with a lower BMI and a different sociodemographic background in comparison with European and American populations (9,32). It cannot be discounted that the effects of lifestyle interventions are only apparent after several years; however, findings from the DPPOS, with the largest sample and ∼18 years of follow-up, did not show this benefit. At the same time, as adherence generally reduces significantly in long-term lifestyle intervention trials, it is extremely difficult to study those interventions for several years (10).

Moreover, the magnitude of weight loss of different interventions should be looked at as one of the important drivers of protection for hard outcomes such as mortality. In a meta-analysis including >170,000 subjects, bariatric surgery was compared with nonsurgical management for obesity with use of patient-level survival data from controlled trials and high-quality cohort studies. In a subgroup analysis with a population with type 2 diabetes, bariatric surgery showed a reduction of 60% in the HR for all-cause mortality, with a median life expectancy 9.3 years longer compared with the control group (33). These results may be related to the potential for weight loss with bariatric surgery, which is not observed in most lifestyle interventions studies, which usually lead to modest weight losses in the mid- to long-term (34). It is well-known that there is wide interindividual heterogeneity in weight loss response (35). A subanalysis of the Look AHEAD trial suggested that weight loss should be looked at as one of the main drivers of reductions in cardiovascular events. It was demonstrated that those who lost >10% of their weight in the first year of intervention independent of group allocation were able to reduce cardiovascular events and total mortality (11). However, when we performed meta-regression models, the magnitude of the weight change along the included studies was not a predictor of cardiovascular and all-cause mortality even when we considered the greatest weight loss and not the weight measurements of the longest time point.

In the case of cardiovascular outcomes, some dietary patterns, such as the Mediterranean and the Dietary Approaches to Stop Hypertension (DASH) diets, are well established as cardioprotective diets (36,37) The Mediterranean diet reduced by 30% the hazards for major cardiovascular outcomes; however, there was no superiority with regard to reducing all-cause mortality (38). A systematic review and meta-analysis of observational studies showed that adherence to the DASH diet had an inverse association with all-cause and cardiovascular mortality. Also, the dose-response analysis showed that each 5-point increment in adherence to the DASH diet could significantly lower the risk of all-cause and cardiovascular mortality in 5% (6–4%) and 4% (5–2%), respectively (36). In our review, these specific dietary patterns were not used in the included studies, so we still do not know whether including these dietary patterns as part of lifestyle interventions would have a protective effect on mortality.

Finally, due to the recognized importance of health care and the monitoring of people with prediabetes or type 2 diabetes, the majority of studies had active control groups (e.g., standard or usual care), and subjects in control groups also received some advice about diet and exercise. These small differences between groups may have been insufficient to cause significant reductions in crude outcomes such as mortality, and it seems that subjects would have benefits from both lifestyle interventions and standard care for this outcome. The Da Qing DPOS (9) was also the only study with comparison of an intensive lifestyle intervention with no advice about diet and physical exercise that had a significant number of cardiovascular and all-cause mortality events. Because of that, its results drove the significant reduction in the outcomes, favoring intensive lifestyle intervention when compared with no type of advice about diet and physical exercise in the subgroup analysis according to characteristics of the control groups. It is possible that its distinct population and the absence of advice about lifestyle changes in the control group may explain its findings.

This review has some limitations. First, the RCTs had distinct lifestyle programs, with different methods of intervention (e.g., educational groups, consultations with dietitians, individualized or group counseling) and distinct control groups, which should be considered in the interpretation of our results. To deal with this limitation, we conducted subgroup analyses to identify the possible impact of the type of dietary and physical exercise recommendation on the outcomes, and no change in the result was found. We also addressed whether the differences in weight changes between intervention and control groups and the potential influence of other factors, such as different responses according to glycemic status, mean age of the trial populations, and time of follow-up, would have a possible impact on all-cause and cardiovascular mortality, and, again, we did not identify any influence of these factors. Secondly, in studies where mortality was reported as losses to follow-up, the risk of bias was difficult to judge, especially for missing outcome data and the selection of the reported results. Notwithstanding, the sensitivity analysis including only studies that considered mortality as primary or secondary outcomes showed that the results did not change. We had no response from the authors of three publications with insufficient data on mortality for our analyses (38–40). However, we did not detect publication bias for all-cause mortality, and we were able to meta-analyze a considerable number of events, strengthening our findings. Finally, it would be interesting to compare the effect of lifestyle interventions with other kind of treatment modalities, such as pharmacological treatments. However, the only study that adhered to our inclusion criteria and had a medication arm (metformin) was the Diabetes Prevention Program (DPP) (10). As a result, we were not able to compare different studies with lifestyle interventions against those with pharmacological interventions in the meta-analysis.

The evidence from this meta-analysis of RCTs demonstrates that intensive lifestyle interventions implemented thus far with subjects with prediabetes and type 2 diabetes were not superior to standard care in reducing cardiovascular and all-cause mortality, with moderate certainty of evidence. New long-term and sufficiently powered RCTs with lifestyle interventions are needed with medical nutrition therapy modalities based on the most recent evidence of their protection against cardiovascular disease and mortality.

Acknowledgments. The authors thank Steven E. Kahn (University of Washington, Seattle WA) for his discussion on their findings and for giving critical and constructive thoughts and feedback.

Funding. This study was financed in part by the Research and Events Incentive Fund of Hospital de Clínicas de Porto Alegre (FIPE-HCPA 2020-0507). K.P.Z. and P.P.T. received a scholarship from Coordination for the Improvement of Higher Education from Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) (Brazil), and P.E.C. received a scholarship from Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) (Brazil). The research group received a grant from CNPq (CNPq/MCTI/FNDCT no. 18/2021; process: 420065/2021-0). E.B. received a Research Establishment Grant from the Queen’s University School of Medicine, consulting fees from Daiichi Sankyo, and payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Janssen. B.H. received payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Novo Nordisk, Eli Lilly, AstraZeneca, Boehringer Ingelheim, and Merck.

Duality of Interest. All authors completed the International Committee of Medical Journal Editors (ICMJE) uniform disclosure form at www.icmje.org/coi_disclosure.pdf. No other potential conflicts of interest relevant to this article were reported.

None of the fees, honoraria, or payments described above for E.B. and B.H. are related to this study.

Author Contributions. K.P.Z., P.P.T., and F.G. were responsible for study conceptualization and design. K.P.Z. and P.P.T. were responsible for data acquisition and analysis. All authors contributed with the selection of titles and abstracts, full text analysis, data interpretation, manuscript writing, and critical review for important intellectual content. K.P.Z. and V.C. performed the statistical analyses. K.P.Z., P.P.T., P.E.C., and F.G. obtained the financing. V.C. and F.G. were the supervisors.

Prior Presentation. Parts of this study were presented in abstract form at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3–7 June 2022.