Participants
The Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) study (ClinicalTrials.gov registration no. NCT00145925) recruited 11,140 participants with type 2 diabetes between June 2001 and March 2003 [18]. Primary outcomes of the trial have been published [19, 20]. Participants were ≥55 years of age and had been diagnosed with type 2 diabetes after the age of 30 years. In addition, they were required to have a history of cardiovascular disease (CVD) or one or more additional cardiovascular risk factors. The trial included two randomised interventions: (1) a double-blind assessment of the efficacy of perindopril/indapamide (2 mg/0.625 mg for 3 months, increasing to 4 mg/1.25 mg if tolerated) vs placebo and (2) an open-label evaluation of an intensive glucose-lowering regimen using modified-release gliclazide (with a target HbA1c of ≤48 mmol/mol [6.5%]) vs standard care. Participants had their serum creatinine levels measured as part of the study protocol at baseline, 4 months and 1 year and annually thereafter until completion of the study, with further tests at the discretion of clinicians. Urinary albumin/creatinine ratio (ACR) was measured as part of the study protocol at baseline, 2 years, 4 years and completion of the study. GFR was estimated using the Modification of Diet in Renal Disease formula. Participants underwent formal eye examination and visual acuity testing at baseline, 2 years, 4 years and completion of the study. Each participating centre obtained ethical approval, and all participants provided written informed consent.
The primary trial outcomes were composites of major macrovascular and microvascular events that occurred during a median of 5 years of follow-up. An independent adjudication committee validated all outcomes. Major macrovascular events were cardiovascular death, non-fatal myocardial infarction or non-fatal stroke. Major microvascular events were defined as a composite of new or worsening nephropathy or retinopathy, in turn defined as any of the following:
-
(1)
development of macroalbuminuria (urinary ACR >33.9 mg/mmol, confirmed by two results);
-
(2)
doubling of serum creatinine level to ≥200 μmol/l (with non-qualifying exceptions of terminal illness or acute illness and subsequent recovery of renal function);
-
(3)
the need for renal replacement therapy due to kidney disease (in the absence of other medical causes requiring transient dialysis), or death due to renal disease;
-
(4)
development of proliferative retinopathy (identified by the incidence of new blood vessels on the disc or elsewhere, vitreous haemorrhage, pre-retinal haemorrhage and fibrous proliferations on the disc or elsewhere in a participant found not to have this condition at entry);
-
(5)
development of macular oedema (characterised by a retinal thickening within one disc diameter of the macular centre in a participant not found to have this condition at entry);
-
(6)
occurrence of diabetes-related blindness (corrected visual acuity 3/60 or worse, persisting for ≥3 months and known to not be due to non-diabetes-related causes in a participant found not to have this condition at entry);
-
(7)
use of retinal photocoagulation therapy.
Blood samples were available from 17 out of 20 countries participating in the ADVANCE study (the exceptions were China, India and the Philippines), giving a total potential source cohort size for the study of 7376 individuals (66.2% of the overall study cohort).
To improve efficiency of the biomarker studies in the ADVANCE trial a case-cohort study has been established [21, 22]. In case-cohort studies, a random sample (called the ‘subcohort’) is drawn and phenotyped from the full cohort; this is very likely to contain both individuals who are ‘cases’ and ‘non-cases’. Cases (generally for multiple case definitions, such as microvascular disease and macrovascular disease) who were not included in the subcohort are then identified from the remainder of the cohort and were also phenotyped. The case-cohort study has several advantages over the nested case–control design, including the ability to investigate multiple endpoints simultaneously. For this case-cohort study, a random subcohort (n = 3500) was selected from the base population, which was enriched by the addition of individuals who had a cardiovascular event, a microvascular event or died during follow-up, giving a total study size of 4197 (Fig. 1) [21, 22].
Proton NMR analysis
Plasma samples were obtained at baseline from all study participants when they were in an unfasted state, given that these were people with type 2 diabetes at risk of hypoglycaemic episodes. Samples were collected across sites in a pragmatic fashion (commensurate with a multinational RCT) according to local facilities. Plasma samples were separated and stored centrally at −80°C until measurement. The present study used a previously unthawed aliquot of plasma for 1H-NMR analysis. 1H-NMR spectroscopy was performed on all available EDTA plasma samples from the ADVANCE case-cohort study at baseline using a low-volume (100 μl) variation of the quantitative 1H-NMR method (Nightingale Health, Helsinki, Finland) described previously [23, 24] and reviewed [25]. Sample spectra were analysed on a Bruker AVANCE III HD spectrometer to quantify a targeted list of metabolites, lipids and lipoproteins, as described previously [25]. This list included eight amino acids (alanine, glutamine, histidine, isoleucine, leucine, valine, phenylalanine and tyrosine), which are detectable using the method, and are not in ‘congested’ regions of the NMR spectrum where multiple metabolites overlap. Metabolomic analyses of plasma samples tend to yield lower analyte concentrations than serum, both by NMR spectroscopy and other methods, although plasma demonstrates better stability and reproducibility [26]. Samples with a low glutamine/glutamate ratio were excluded from analyses of glutamine associations. Levels of all other amino acids were consistent with published data.
Statistical analysis
Continuous data with approximately normal distributions (including all amino acids) are presented as mean ± SD; those with skewed distributions are presented as median (with interquartile range). Categorical data are presented as n (%). Pearson correlations were used to explore associations of the amino acids with each other. Associations of amino acids with classical risk factors were investigated across quarters of the distribution of each amino acid.
Cox regression models were fitted using the STSELPRE procedure for case-cohort analyses (StataCorp, College Station, TX, USA). Models estimated HRs for a 1 SD increase in each amino acid with each of the endpoints. Two models, with different potential confounding variables, were fitted for each amino acid/outcome combination: model 1 with age, sex, region and randomised treatment; model 2 with, additionally, a prior macrovascular complication of diabetes (myocardial infarction, stroke, hospital admission for a transient ischaemic attack or for unstable angina, coronary or peripheral revascularisation, or amputation secondary to peripheral vascular disease), duration of diabetes, current smoking, systolic blood pressure, BMI, urinary ACR, eGFR, HbA1c, plasma glucose, total cholesterol, HDL-cholesterol, triacylglycerols, aspirin or other antiplatelet agent, statin or other lipid-lowering agent, β-blocker, ACE inhibitor or angiotensin receptor blocker, metformin use, history of heart failure, participation in moderate and/or vigorous exercise for >15 min at least once weekly, and high-sensitivity C-reactive protein (CRP). A third adjustment model, attempting to include all amino acids in the same model, resulted in collinearity and estimates were thus not available. Non-linearity was tested by comparing the deviances of linear and categorical models and by the inclusion of polynomial components (quadratic and cubic terms). Other analyses were performed using SAS v9.2 (SAS Institute, Cary, NC, USA). All p values reported are two-sided, with the 5% threshold used to determine significance.
For the random subcohort, the ability of amino acids to discriminate between those who will and those who will not go on to suffer each of the three adverse outcomes were estimated, in the context of model 2, using c statistics for 5 year risk, accounting for censoring. In addition, the ability of amino acids to reclassify participants according to 5 year risk, using the continuous net reclassification index (NRI), was assessed by methods suitable for survival data, using bootstrapping [27].
Primary results came from use of all available data; sensitivity analyses using only participants with complete data were also performed.