The Whitehall II study is an occupational cohort of 10,308 British civil servants who have been followed up with clinical examinations every 5 years since 1985 . The current analysis utilised data from phase 7 (2002–2004) and phase 9 (2007–2009), when fasting plasma glucose (FPG), 2 h plasma glucose (2hPG) and HbA1c were measured. From the 6967 participants at phase 7, we excluded 671 (9.6%) who did not participate at phase 9, 392 (5.6%) with known diabetes at phase 7, 684 (9.8%) who could not be classified with normoglycaemia, prediabetes or diabetes at both phases on at least one criterion, and 27 (0.4%) with screen-detected diabetes according to all the three criteria at phase 7, leaving 5193 (74.5%) of the phase 7 participants for analysis (ESM Fig. 1). For the analyses of FPG and 2hPG, participants had to have fasted for ≥8 h before the clinical examinations. The University College London Ethics Committee reviewed and approved the study. Written informed consent was obtained from all participants at each study phase.
Measurements and definitions
Participants underwent a standard 75 g OGTT with measurement of FPG and 2hPG using the glucose oxidase method . HbA1c was measured in whole blood, drawn into BD Vacutainers (Becton Dickinson, Winnersh, UK), using the validated Tosoh G8 high-performance ion-exchange liquid chromatography platform (Tosoh Bioscience, Tessenderlo, Belgium).
Prediabetes was defined as FPG 5.6–6.9 mmol/l, 2hPG 7.8–11.0 mmol/l and/or HbA1c 39–47 mmol/mol (5.7–6.4%) according to criteria from the ADA . For each of the three criteria, normoglycaemia and diabetes were defined as values below and above the cut-off points for prediabetes, respectively. Diabetes could also be diagnosed by a doctor outside the study.
The primary outcome was a composite endpoint of a CVD event or death between 2007–2009 (phase 9) and the end of follow-up (30 June 2015). Adjudicated CVD events included fatal and non-fatal myocardial infarction and stroke . All-cause mortality was obtained from the NHS Central Registry, which provided information on the cause and date of death.
All analyses were conducted separately for the three glycaemic criteria, in accordance with the guidelines from ADA  and the International Expert Committee (IEC) , which suggest that the different criteria to define prediabetes and diabetes should not be combined. For each criterion, participants with screen-detected diabetes at baseline by the given glycaemic criterion were excluded.
Among individuals with prediabetes at phase 7, we calculated the probability of reversion to normoglycaemia at the phase 9 re-examination approximately 5 years later. Poisson regression analysis with log(person time) as offset was used to estimate and compare incidence rates of future CVD or death in individuals with prediabetes who did or did not revert to normoglycaemia. The follow-up period of each participant was split into 1 year age bands to account for the non-constant effect of age over time on CVD risk and mortality. Analyses were adjusted for age, sex, ethnicity and previous CVD. A complete case approach was used.
All participants with 2hPG available also had measurements of FPG and HbA1c (but not vice versa). Among participants with 2hPG-defined prediabetes and elevated FPG and/or HbA1c, we compared event rates between participants who did or did not revert to normoglycaemia on FPG and/or HbA1c, without normalising 2hPG (ESM Fig.1).
We further estimated the associations of changes in FPG, 2hPG or HbA1c from phase 7 to phase 9 with risk of future CVD or death, excluding participants with known diabetes at phase 9 because they were likely to be receiving treatment.
Statistical analyses were performed in R, version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria).