Published evidence of lifestyle in T2D risk and prevention
The review highlighted several well-conducted RCTs demonstrating the effect of lifestyle interventions in reducing the risk of T2D. In the Diabetes Prevention Program (DPP; n = 3234), an intensive lifestyle intervention was superior to a pharmacological intervention of metformin in reducing the incidence of diabetes over 2.8 years of follow-up compared to placebo [
79]. The incidence of T2D was reduced by 58% (95% CI 48–66%) in the lifestyle group and by 31% (95% CI 17–43%) in the metformin group compared to the placebo group. Follow-up of the same cohort for a further 10 years showed persistent beneficial effects from lifestyle interventions, with T2D onset being delayed by approximately 4 and 2 years in the lifestyle and metformin groups, respectively, compared to placebo [
6]. The Finnish Diabetes Prevention Study (DPS; n = 522) implemented a similar lifestyle protocol, which also conveyed a 58% (95% CI 30–70%) reduction in diabetes incidence [
86]. Both protocols focused on overweight or obese (BMI > 25 kg/m
2) participants with impaired glucose regulation, and promoted a 5–7% weight reduction, a total and saturated fat intake reduction, an increase in fiber intake, and regular exercise. Numerous other trials focusing on lifestyle interventions in a range of ethnic groups and study settings followed [
20,
29,
58,
73], reporting reductions in diabetes incidence of comparable or lesser magnitude to the DPP and Finnish DPS trials. Long-term follow-up indicates that benefits of intensive lifestyle modification on diabetes incidence are sustained for up to 20 years [
68].
Large prospective cohort studies have also reported robust associations between lifestyle exposures and T2D in diverse populations. For example, in the Kailuan prospective study (n = 50,656) [
15], changes from the ideal cardiovascular health status score were inversely associated with risk of T2D over an average 3.8 years of follow-up. In the Finnish Twins Study (n = 20,487) [
55], leisure-time physical activity reduced the risk of incident T2D in both monozygotic and dizygotic twins who were physically active compared to their sedentary siblings (HR 0.6, 95% CI 0.43–0.84,
P = 0.003), even after factoring in familial risk and home environment. In the Strong Heart Study [
64], among American Indians (n = 1651) followed for 10 years, high physical activity was associated with a reduced risk of T2D (OR 0.71, 95% CI 0.51–0.99 in the highest quartile compared to those who reported no physical activity), although the estimates were attenuated and became non-significant after adjusting for adiposity. A similar beneficial impact of physical activity in T2D incidence was observed in a European case-cohort study of nearly 30,000 adults [
46], and in 68,000 female US health professionals [
78].
Yates et al. [
5] systematically reviewed the literature on RCTs testing the efficacy of diet and/or exercise interventions in the prevention of T2D. Of the eight trials reviewed, most involved a combined diet and exercise intervention. Compared with standard of care, the reduction in the risk of developing T2D attributed to the lifestyle interventions averaged 50% during the trial’s randomization phase. Long-term follow-up of the DPP Outcomes Study [
6], the Finnish DPS [
87], and the China Da Qing Diabetes Prevention Study [
68], all indicated that the reduced risk of diabetes attributable to lifestyle intervention is sustainable for 10–20 years post-randomization. Although attempts have been made to parse out the relative contributions of diet and exercise in diabetes prevention, most lifestyle trials have not been designed for this purpose, and generally assess diet and exercise using self-report methods prone to respondent bias. However, Slentz et al. [
88] recently reported that, within an exercise-only intervention trial, high volume moderate-intensity exercise (~18.2 km/week of walking) alone substantially reduced glucose tolerance in people at high risk of T2D, despite modest effects on body weight reduction (~2 kg). Diet-only interventions, such as that used in the PREDIMED trial focusing on Mediterranean-style diets [
89], have yielded reductions in diabetes risk of approximately 50% compared with a control intervention.
RCTs are often considered the gold-standard in the hierarchy of causal evidence, as double-blind, placebo controlled trials are generally robust to confounding and reverse causality. However, in lifestyle intervention trials, masking treatment allocation from the participant and investigators is extremely challenging, which may result in compensatory behaviors that introduce bias and confounding – a rarely discussed caveat that affects the validity of data from lifestyle RCTs. Nevertheless, abundant epidemiological studies and clinical trials have implicated multiple lifestyle factors in the development of T2D.
Poorer social environments, within which fewer resources and opportunities exist to maintain healthy lifestyles, convey an increased risk for many diseases, including obesity and T2D [
90]. Studies in twins suggest that the relationship between socioeconomic status and obesity may be modified by genetic variation [
91]. Using data from the UK Biobank, Tyrrell et al. [
92] studied the interaction of 66 established BMI-associated variants and 12 obesogenic lifestyle exposures in obesity; the authors extended previous discoveries of genetic interactions with physical activity [
93,
94] and TV viewing [
95], and identified a novel interaction with the Townsend Deprivation Index [
96].
Evidence on the association between environmental exposures (e.g., to particulate matter and persistent organic pollutants) and T2D has yielded mixed results: some studies showed a statistically significant relationship between long-term exposure and risk of T2D [
18], with higher risk attributed to traffic-specific pollution [
33,
47], whereas others found no association [
21]. Furthermore, one study reported an association of traffic-specific pollution exposure with T2D risk in women [
51]. Exposure to arsenic [
24,
81] and persistent organic pollutants was also significantly associated with T2D [
19,
61], but no such association was seen for cadmium exposure [
35,
44].
Coffee consumption has been associated with lower risk of dysglycemia in observational studies [
48], yet recent Mendelian randomization analyses do not support a causal relationship [
97,
98]. In the Adventist Health Studies [
65] and the Women’s Health Study [
75], consumption of red and processed meats was significantly associated with increased T2D risk. Intake of dairy products was not consistently related to T2D, cardiovascular disease, or all-cause mortality [
38]. In the EPIC InterAct Study [
99], dietary fiber consumption was associated with lower T2D incidence, though this was partially explained by body weight. While psychosocial health is an important risk factor, especially in the elderly [
13], the relationship between depressive symptoms and dysglycemia may be mediated by other factors such as lifestyle [
14]. Working overtime was significantly associated with risk of T2D among nurses [
70] and among Japanese men working more than 50-hours overtime per month [
83].
Although the literature on specific lifestyle exposures and T2D risk is extensive, all studies used to inform guidelines are based on the average estimated effects in the studied population, which would be acceptable if susceptibility to risk factors and response to preventive interventions were homogeneous. However, there is tremendous between-person variability in susceptibility and response to lifestyle exposures, which undermines the value of uniform recommendations. Indeed, it is estimated that the majority of people undergoing exercise interventions do not show an adequate response [
8]. There are various reasons why there may be a lack of response to lifestyle interventions, many of which are irrelevant to individual biology; these factors are listed in Table
1. Ignoring these factors when estimating the likely impact of lifestyle precision medicine in T2D would substantially overestimate its value [
100]. Nevertheless, harnessing genotypes and other omic variants to optimize lifestyle interventions for population subgroups may significantly impact individual and population-scale diabetes trajectories. Such approaches are especially appealing in prediabetes, where measures of glycemia alone are inadequate, and additional biomarkers are likely needed to predict or prevent progression to full-blown disease.
Table 1
Factors influencing response to lifestyle interventions
Behavioral compensation | In most cases, assignment to lifestyle interventions in clinical trials cannot be masked from the participants or investigators. This may prompt changes in behavior that are not the main objective of the trial and which differ by treatment arm, or may cause investigators to treat participants in the lifestyle and control arms differently. These sources of bias may underlie what appears to be variability in treatment response. |
Regression to the mean | Trials that include only one outcome assessment, and which assess change in the outcome as the difference between the baseline and follow-up measure, are likely to be prone to regression dilution bias (or regression to the mean). This phenomenon occurs because most assessments are made with some degree of error, meaning that, in some participants, the change in the outcome will be underestimated and in others it will be overestimated. Where the outcome is assessed using a physical stress test (such as on a treadmill or bicycle ergometer), differences in effort at the beginning and end of the trial will also contribute to the apparent variability in treatment response. This problem could in principle be overcome in a randomized controlled trial by conditioning treatment response on response to the control intervention, although this is not conventionally done in studies of responders and non-responders, which generally focus only on intervention groups. |
Adherence | Variability in the extent to which participants follow protocols in clinical trials (adherence) is likely to play a significant role in determining the extent to which an intervention appears to work. Although adherence is usually monitored in trials, monitoring adherence to lifestyle interventions is challenging, as the accurate and precise assessment of diet and exercise is notoriously difficult. The use of self-reported diet and/or exercise instruments to monitor adherence is likely to be insufficient in lifestyle trials, as participants in the active intervention arm may feel pressured to provide confirmatory responses to lifestyle questions. |
Background heterogeneity in behaviors | Lifestyle interventions are often comprised of around 150 mins/week contact time, accounting for approximately 2% of all waking time. During the 98% non-contact time, participants’ behaviors are likely to vary considerably, influencing the extent to which the trial’s outcomes change. |
Most contemporary studies of the interplay between genetic and lifestyle factors have focused on gene variants and lifestyle exposures previously associated with the disease of interest. Few of the exposures mentioned above have been studied in the context of gene–lifestyle interactions. A systematic search of the PubMed database (conducted on July 19, 2017; see Additional file
1: S1 for the search string) identified 30 original research articles and 13 review articles or commentaries. Of these, only seven publications focused on T2D as the outcome; additional ten relevant papers were identified through ancestral searches some of which are RCTs [
101‐
106]. Others reported on prospective observational studies examining gene variant interactions with different diet components [
107‐
110] and with physical activity [
107,
111].