Heterogeneity of Treatment Effects From an Intensive Lifestyle Weight Loss Intervention on Cardiovascular Events in Patients With Type 2 Diabetes: Data From the Look AHEAD Trial

In patients with type 2 diabetes, it has been demonstrated that weight loss has beneficial effects on metabolic control and cardiovascular risk factors (1,2), and bariatric surgery may decrease the risk of cardiovascular events in obese patients with type 2 diabetes (3). However, it has not been demonstrated that weight loss through lifestyle intervention also has a positive effect on cardiovascular outcomes. The Action for Health in Diabetes (Look AHEAD) trial randomized overweight and obese patients with type 2 diabetes to an intensive lifestyle intervention program that promoted weight loss through decreased caloric intake and increased physical activity or to a control group with regular diabetes support and education (4). It was ended prematurely on the basis of a futility analysis after a median follow-up of 9.6 years. Although the intervention led to greater weight loss and greater reductions in glycated hemoglobin (HbA_1c) and cardiovascular risk factors, there was no reduced risk of cardiovascular events or mortality in the intervention group compared with the control group (1).

Subgroup analyses in the Look AHEAD trial have shown that in several subgroups, the lifestyle program led to an important reduction of cardiovascular risk factors. A post hoc analysis has shown that individuals who lost more than 10% of their body weight had a significantly lower risk of major cardiovascular events (MACE) (5). These findings suggest that lifestyle modification may have a beneficial effect in reducing cardiovascular outcomes in individual patients or patient subgroups, but no subgroups based on single baseline characteristics were identified with a significant treatment effect on cardiovascular outcomes. However, treatment decisions based on group level are suboptimal, as they are based on single patient characteristics and it may be possible that a combination of patient characteristics influences the treatment effect from an intervention. Furthermore, most trials do not have enough power to study treatment effects in subgroups (6). The question therefore remains whether intensive lifestyle interventions can be beneficial in reducing cardiovascular events in individual patients, when no treatment effect has been found on average.

The aim of this post hoc analysis of the Look AHEAD trial is to explore the possible presence of heterogeneity of treatment effect (HTE) of an intensive lifestyle intervention on the occurrence of MACE in overweight or obese patients with type 2 diabetes and to identify patient characteristics associated with HTE.

We submitted a research proposal for the current study to the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Central Repository. The proposal was reviewed by an independent review panel, and access to the anonymized individual patient data from the Look AHEAD trial was provided.

The Look AHEAD trial (registration no. NCT00017953, ClinicalTrials.gov) included 5,145 overweight and obese patients with type 2 diabetes, of whom the data for 4,901 are available in the public access data sets, as those participants from Native American sites are excluded, per consent limitations. The design and methods of Look AHEAD have been reported elsewhere (4). In short, participants were recruited from 16 clinical centers in the U.S. and were then randomized to either diabetes support and education (control) or intensive lifestyle intervention. Trial enrollment began in 2001, with follow-up continuing through 2012. The Look AHEAD trial complied with the Declaration of Helsinki, ethical approval was obtained from local institutional review boards, and all study participants provided written informed consent.

As defined in the trial, the primary end point was the first occurrence of a four-point MACE, a composite end point of nonfatal myocardial infarction, nonfatal stroke, hospitalization for angina, and death from cardiovascular causes.

Covariate data were missing in <1% of study participants and were imputed using single imputation using predictive mean matching, based on other patient characteristics and outcomes. All analyses were conducted with R statistical software version 3.4.1 (www.r-project.org), using add-on packages Hmisc, survival, and penalized (7). For all analyses, a P value < 0.05 was considered statistically significant, unless noted otherwise.

To model treatment effect, first, a Cox proportional hazards model for the prediction of MACE was derived in the Look AHEAD trial population. The model contained the following prespecified predictors: age, sex, current smoking, history of cardiovascular disease (CVD), use of insulin, the duration of diabetes, systolic blood pressure (SBP, in mmHg), non-HDL cholesterol (in mmol/L), HbA_1c (in mmol/mol), estimated glomerular filtration rate (measured according to the Chronic Kidney Disease Epidemiology Collaboration formula, in mL/min/1.73 m²), BMI (in kg/m²), and the presence of micro- or macroalbuminuria. The choice for these predictors was based on a recently developed lifetime risk model in patients with type 2 diabetes (8). To model treatment effect directly, treatment-by-covariate interaction terms for all included predictors were added to the model (6,9). To avoid chance findings and overfitting, no statistical selection was applied. Thus, all preselected predictors and interactions were included in the model. Continuous predictors were truncated at the 1st and 99th percentiles to limit the effects of outliers. To improve the robustness of the model, transformation was applied for continuous variables when this improved model fit based on the Akaike information criterion (10). The proportional hazards assumption was checked by visually assessing the scaled Schoenfeld residuals. The final model coefficients were estimated using penalized estimation methods using an L2 quadratic (i.e., “ridge”) penalty to further prevent overfitting (6,7). The predictive value of the model was assessed based on calibration plots of predicted versus observed 10-year risks of MACE and the C-statistic as a measure for discrimination.

Subsequently, the newly derived prediction model was used to estimate the 10-year risk of MACE for each study participant using two scenarios: as if they had been treated 1) with diabetes support and education or 2) with intensive lifestyle intervention. The treatment effect was defined as the 10-year risk with diabetes support and education minus the 10-year risk with the intensive lifestyle intervention.

Next, to assess HTE for intensive lifestyle intervention, the study population was divided into quartiles based on predicted treatment effect. The hazard ratios (HRs) and associated 95% CIs for the effect of treatment per quartile were obtained with Cox proportional hazards models corrected for the prognostic variables used in the risk models to ensure that no confounding had been induced by the division of the study population in quartiles.

Baseline characteristics were described for the four quartiles to identify characteristics that are associated with a high versus low predicted benefit from an intensive lifestyle intervention for the risk of MACE. The differences between these patient characteristics over quartiles of baseline risk were compared using Kruskal-Wallis tests for continuous variables and χ² tests for categorical variables, with Bonferroni correction for multiple testing. Furthermore, in order to generate hypotheses about the causes of possible HTE, we graphically displayed changes in body weight, waist circumference, HbA_1c, SBP, and LDL cholesterol during the study period stratified for quartiles 1 and 4 and for treatment allocation.

Of the patients included in the trial, 2,448 (50%) were allocated to the intensive lifestyle intervention arm. Patients included in the trial were on average 59 years old, 41% were male, and 14% had a history of CVD. The mean BMI was 36 kg/m². The median duration of type 2 diabetes before inclusion in the trial was 5 years. Detailed baseline characteristics of the study population are shown in Supplementary Table 1 for the entire study population and in Table 1 stratified for quartiles of estimated treatment effect. The median follow-up in the trial was 9.4 years (interquartile range [IQR] 8.5–10.2), during which MACE occurred 799 times.

Supplementary Table 2 shows the formula for the estimation of the risk of MACE that was used for the predictions. Supplementary Fig. 1 shows good agreement between the predicted and observed risk of MACE in the study population (internal validation); the C-statistic for discrimination was 0.73 (95% CI 0.71–0.75).

The median estimated baseline 10-year risk for MACE when treated with the control treatment was 15% (range 0.3–96%) (Fig. 1). The median estimated absolute treatment effect on 10-year risk for MACE with lifestyle intervention was −1.3% and varied substantially, ranging from −39% to +43% (Fig. 1).

Figure 2 shows the event rates in both treatment arms, the associated HRs, and the median absolute treatment effect stratified for quartiles of estimated treatment effect. Cox proportional hazard models adjusted for all prognostic factors included in the risk model showed an observed benefit of intervention versus control in quartile 1 (HR 0.64 [95% CI 0.49–0.83]), no statistically significant treatment effect from treatment in quartiles 2 and 3 (HR 0.81 [95% CI 0.58–1.14] and 1.13 [95% CI 0.80–1.60], respectively), and a detrimental effect of intervention in quartile 4 (HR 1.37 [95% CI 1.09–1.73]). Table 1 shows the baseline patient characteristics stratified for the quartiles of predicted treatment effect. Figure 3 shows the percentage change from baseline in body weight, waist circumference, HbA_1c, SBP, and LDL cholesterol during 10 years of follow-up in quartiles 1 and 4 stratified for trial allocation.

This exploratory analysis of the Look AHEAD trial demonstrated HTE from an intensive lifestyle intervention on the occurrence of MACE in overweight and obese patients with type 2 diabetes. Furthermore, patient characteristics were identified that are associated with possible HTE, including patient demographics, medical history, measures of socioeconomic status, and laboratory values.

Currently, most international guidelines for type 2 diabetes include recommendations of lifestyle interventions (11–13). The European Association of Preventive Cardiology recently published a position paper stressing the importance of exercise training in patients with type 2 diabetes and CVD, based on the potential of exercise to improve cardiovascular and metabolic functions, despite the lack of evidence of a positive effect on cardiovascular risk (14). In the current study, it is demonstrated that there is a subgroup of patients in the Look AHEAD trial who did benefit from a lifestyle intervention aimed at weight loss in terms of a reduction in the risk of MACE, confirming the importance of lifestyle interventions in at least part of the population with type 2 diabetes.

Based on the results from the current study, however, there may also be a group of patients in whom an intensive lifestyle intervention has a detrimental effect on CVD-free survival. It is however important to realize that the intervention in the Look AHEAD trial is a specific and intensive lifestyle intervention, and these results may be different in other lifestyle programs. The Look AHEAD intervention aims at a low caloric intake (1,200–1,800 kcal per day) and increased physical activity (at least 175 min of moderate-intensity physical activity per week). It is unclear how different weight loss lifestyle interventions would influence the findings of the current study. Furthermore, the lifestyle intervention in Look AHEAD has been found to improve, e.g., quality of life (15), mobility (16), sleep apnea (17), sexual dysfunction (18), and depression (19), and improved glycemic control will also benefit the risk of microvascular complications (12).

Nonetheless, based on the results from the current study, it may be wise to be cautious with regard to very intensive lifestyle interventions, such as the intervention from the Look AHEAD trial, in certain patient categories. Future research should be aimed at investigating which types of (intensive) lifestyle interventions are effective and safe to use in these subgroups of overweight and obese patients with type 2 diabetes.

The current study identifies several patient characteristics that differ between quartiles of predicted treatment effect (as shown in Table 1), whereas in (prespecified) simple subgroup analyses, no baseline characteristics were identified that modified the treatment effect of the intensive lifestyle intervention. In a prespecified subgroup analysis, a nonsignificantly higher event rate for the primary outcome was found in the intervention arm compared with the control arm in the subgroup with patients with a history of CVD, versus a nonsignificantly lower event rate in patients without CVD (1), which is in line with the findings in the current study. A post hoc, machine learning–based analysis identified higher HbA_1c levels as a characteristic associated with treatment benefit (20), which is in contrast with the findings in the current study. This difference may be explained by the multivariable approach of the current study, compared with the subgroup-based approach of the machine learning–based study. Subgroup analyses are univariable analyses, whereas HTE likely cannot be explained by single patient characteristics only (6). Importantly, although certain characteristics may have been found to be associated with the treatment effect of intensive lifestyle intervention in Look AHEAD, this does not necessarily imply a causal relation between this characteristic and the modified treatment response. It may prove to be difficult to disentangle the relation between these risk factors and HTE. For example, although the proportion of current smokers is higher in quartile 4 compared with the other quartiles, smoking status is also associated with having a history of CVD and with socioeconomic status. Using the current methodology, it is not possible to prove which of these risk factors, if any, are independent and causal treatment effect modifiers.

However, although the methodology of the current study is not suitable for investigating causal relationships, it can be used to generate hypotheses about possible mechanisms underlying the HTE found in the study.

In the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, which randomized patients with a history of or a high risk for CVD to either strict or standard glycemic control, an increased risk of mortality was seen in the intensive glycemic control arm (21). This increase was most prominent in patients with a high baseline HbA_1c (>8.5% or 69 mmol/mol) (22). Subsequent analyses found that this was partly, although not fully, explained by the increased risk of hypoglycemia. In ACCORD, the risk of hypoglycemia was higher in African Americans, those with lower education levels, those with higher baseline HbA_1c levels, those with signs of nephropathy, and users of insulins (23,24), characteristics that are also associated with a detrimental effect of treatment in the current study. It is possible that the risk of hypoglycemia may partly explain the increased risk in part of the study population in Look AHEAD. In line with the findings in the ACCORD trial, the current study found that in the quartile with the largest risk increase, the mean baseline HbA_1c levels were highest, and the decline in HbA_1c during the study period was markedly steeper than in those patients with a predicted risk reduction (those who would benefit from the intensive lifestyle intervention), as shown in Fig. 2. In ACCORD, however, not all of the increased risk of mortality could be explained by hypoglycemia, as is the case in the current study, and it is still uncertain what other underlying mechanisms are present.

The current study has several strengths. First of all, a multivariable risk prediction-based approach was used to deal with several limitations from subgroup analyses (6). Second, using a risk-based approach to define quartiles of treatment effect, randomization remains intact within these quartiles. To further ensure that no confounding has been induced by chance during stratification, the HRs of intervention per quartile were corrected for prognostic factors.

Several limitations should also be acknowledged. First, it is important to note that these analyses are exploratory and not prespecified in the trial. Therefore, the conclusions should be interpreted with caution. Additionally, differences of baseline characteristics between quartiles of treatment effect were at least partially created by the choice of predictors used in the prediction model. Third, the results from this study cannot be used to make causal inferences but are merely hypotheses generating. Fourth, we did not have data available to analyze the presence of hypoglycemia, variability in glucose control, or hemoglobin glycation index as potential mechanisms underlying the potential treatment effects. Finally, as the treatment effect–based model was derived within the Look AHEAD data, it is possible that there is overfitting of this model to the data, which gives the risk of false discoveries (6). However, to limit this risk, we used prespecified predictors and estimated the final model using penalized regression.

In conclusion, this exploratory study of the Look AHEAD trial shows HTE from an intensive lifestyle intervention aimed at weight loss on the occurrence of MACE in overweight and obese patients with type 2 diabetes. Using an approach based on treatment effect modeling, it is possible to identify subgroups of patients with a possible beneficial or potentially even detrimental treatment effect of the intensive lifestyle intervention used in the Look AHEAD trial on cardiovascular outcomes. The patient characteristics associated with a potential treatment benefit are, among others, no history of CVD, good control of type 2 diabetes, no use of insulin, higher socioeconomic status, and the absence of macroalbuminuria. Future research into intensive lifestyle weight loss interventions for CVD risk reduction should be specifically aimed on the one hand at subgroups of patients with a higher likelihood of treatment benefit and on the other hand at finding safe lifestyle interventions for subgroups of patients with a potential treatment harm.

Funding. Look AHEAD was conducted by the Look AHEAD Research Group and supported by the NIDDK, the National Heart, Lung, and Blood Institute, the National Institute of Nursing Research, the National Institute on Minority Health and Health Disparities, the Office of Research on Women’s Health, and the Centers for Disease Control and Prevention. The data from Look AHEAD were supplied by the NIDDK Central Repository.

This manuscript was not prepared under the auspices of Look AHEAD and does not represent analyses or conclusions of the Look AHEAD Research Group, the NIDDK Central Repository, or the National Institutes of Health.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. T.I.d.V. wrote the manuscript and researched data. J.A.N.D. reviewed and edited the manuscript. Y.v.d.G., F.L.J.V., and J.W. contributed to discussion and reviewed and edited the manuscript. J.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented in abstract form at the European Society of Cardiology Congress 2019, Paris, France, 31 August–4 September 2019.