A re-analysis of data from the Look AHEAD study with the addition of Hp phenotype measurement was undertaken to determine the relationship between the weight loss ILI (compared to DSE) and CAD events among each of the Hp phenotype groups separately. The design, methods and major findings of the Look AHEAD study (ClinicalTrials.gov identifier: NCT00017953) have been reported previously [4, 5]. Briefly, 5145 patients with type 2 diabetes were recruited between August 2001 and April 2004. Participants were aged 45–75 and had a body-mass index (BMI) of 25 kg/m2 or more (27 or more in participants taking insulin), a glycated hemoglobin (HbA1c) of 11% or less, a systolic blood pressure of < 160 mmHg, a diastolic blood pressure < 100 mmHg, a triglyceride level of less than 600 mg/dL, the ability to complete a valid maximal exercise test and an established relationship with a primary care provider. Participants with and without a history of cardiovascular disease were included [4, 5]. Participants were randomized to receive either ILI (aimed at achieving and maintaining weight loss of at least 7% by focusing on reduced caloric intake and increased physical activity) or to receive DSE over a median follow-up of 9.6 years. The ILI included group and individual counseling sessions, which occurred weekly during the first 6 months with decreasing frequency over the course of the trial. Specific intervention strategies included a calorie goal of 1200 to 1800 kcal per day (with < 30% of calories from fat and > 15% from protein), the use of meal-replacement products, and at least 175 min of moderate-intensity physical activity per week. A toolbox of strategies was available for participants having difficulty achieving the weight-loss goals. DSE included three group sessions per year focused on diet, exercise, and social support during years 1 through 4, and annually thereafter. Each participating center obtained ethical approval, and all participants provided written informed consent [4, 5].

Hp phenotyping was performed using a validated high throughput enzyme linked immunosorbent assay (ELISA) that can distinguish the Hp2-2 protein from the non-Hp2-2 proteins with a sensitivity and specificity of 99% and 98.1% respectively [17]. The ELISA identifies Hp phenotypes based on the differences in Hp protein size/structure [17]. There is a 1:1 correspondence between Hp genotype and Hp phenotype [18]. Hp phenotype does not change over time; therefore, a blood sample from any follow-up visit was used. Of the 5145 Look AHEAD participants, a serum sample was available for Hp phenotyping for 4542 (88.3%). The remaining 603 participants were excluded because serum samples from these participants were not available due to the depletion of samples from other studies or due to consent limitations.

Our primary outcome of major CAD events was defined as a composite of the following pre-specified Look AHEAD outcomes [4, 5]: fatal and non-fatal MI, hospitalization for angina, and fatal CAD (definite and probable). An independent adjudication committee validated all outcome events [4]. Although the mechanism is not well understood, stroke is an endpoint that has been associated with the Hp1-1 phenotype rather than the Hp2-2 phenotype [19, 20]. Stroke is a composite of different stroke subtypes with different etiologies that are not always related to atherosclerosis, suggesting that CAD and stroke should be separated from a composite CVD outcome for analyses by Hp phenotype. Therefore, the present analysis studied the primary outcome of CAD events rather than the original Look AHEAD study primary outcome of a composite of death from cardiovascular causes, nonfatal myocardial infarction, non-fatal stroke, or hospitalization for angina.

In a sensitivity analysis, we also investigated the relationship between the ILI and other Look AHEAD outcomes including the study primary composite outcome of CVD (death from cardiovascular causes, nonfatal myocardial infarction, nonfatal stroke, or hospitalization for angina), total mortality, and severe hypoglycemia events (loss of consciousness, seizure, or a glucose < 70 mg/dL that prevented self-treatment and required assistance of another person) (Additional file 1: Table S1).

All analyses were conducted using STATA/SE software version 18 (StataCorp, College Station, TX). With the exception of when testing for Hardy Weinberg equilibrium (HWE), the common approach of dichotomizing the Hp2-2 phenotype variable to represent Hp2-2 phenotype (yes/no) was used because of the low frequency of the Hp1-1 phenotype and the similar structure and function of Hp1-1 and Hp2-1 relative to Hp2-2 [8, 9, 16, 21, 22].

Participants were grouped based on a combination of their treatment assignment and Hp phenotype, and baseline characteristics were summarized using t tests, or Kruskal-Wallis tests for continuous variables and x2 tests for categorical variables.

The goal of the present analysis was to replicate the original Look AHEAD study [4] analysis as closely as possible within each phenotype group. As such, cause-specific Cox proportional hazards regression models were used to assess the effect of the intervention on incident CAD, as is recommended for etiological regression analyses even with competing risks [23, 24]. We assessed for effect modification of this association by haptoglobin phenotype by including an interaction term between intervention group and Hp phenotype and stratifying our results by Hp phenotype. Multivariable models were adjusted for traditional risk factors as well as any other variables that were different between treatment groups at baseline. The frequency of the Hp2-2 phenotype differs among race-based and geographic populations [10] and so race/ethnicity was also identified as an important variable to be included in the model. As such, models were adjusted for age, sex, study site, previous CVD, race/ethnicity (for model in all participants), triglycerides, systolic blood pressure, diastolic blood pressure, income (with category for missing), education, diabetes medication use, anti-hypertensive medication use, lipid medication use, and antidepressant medication use.

We identified a priori sub-groups for stratification of our primary analyses for this study. Current reporting guidelines recommend disaggregation of results by sex [25], and the distribution of the Hp phenotype frequencies differ among race-based and geographic populations [10]. Current diabetes care guidelines suggest that diabetes duration and established CVD are important factors in glucose management [26], Thus, stratified analyses by race/ethnicity, sex, previous CVD at baseline, diabetes duration (> 10 years), and diabetes medication use at baseline were performed in each phenotype group separately. For the race/ethnicity stratification, we were only able to run the adjusted model in the following race/ethnic groups with sufficient numbers for valid estimates and to ensure anonymity: White (67% of total sample), Black (17%) and Hispanic (13%). Interactions were tested between ILI and race/ethnicity, sex, CVD history at baseline, diabetes duration, and diabetes medication use by adding an interaction term to the model for each phenotype group.

On September 14, 2012, on the basis of a futility analysis and recommendation from the data and safety monitoring board, the Look AHEAD intervention was stopped and all data were censored at that date [4]. Follow-up time for the current analysis was defined as the time from date of randomization to date of documented outcome, or until a participant was censored if no event occurred. To account for multiple testing, we applied a Bonferroni corrected significance level of P < 0.002 (0.05 divided by 24). In a sensitivity analysis, inverse probability weighting was used to assess the impact for potential selection bias of excluding individuals missing Hp phenotype data from all enrolled participants (11.7%) [27]. In another sensitivity analysis, we restricted follow-up to years 1, 3 and 5.