Study population and design
The Whitehall II study was initiated in 1985 and recruited 10,308 participants (3,413 women) aged 35–55 years, with a response rate of 73%, from 20 London-based Civil Service departments . The initial visit (phase 1) included a clinical examination and a self-administered questionnaire in 1985–1988. During the follow-up, 5-yearly clinical examinations were performed (Phase 3: 1991–1994; Phase 5: 1997–1999; Phase 7: 2002–2004; Phase 9: 2007–2009) and additional postal questionnaire-only phases were conducted (Phase 2: 1988–1990; Phase 4: 1995–1996; Phase 6: 2001; Phase 8: 2006). The University College London Medical School Committee on the Ethics of Human Research provided ethical approval for the study, and written informed consent was obtained from all participants.
As a 75 g OGTT was first performed in phase 3; this provides the baseline for the current analysis. Of the 8,411 South Asian and white participants at baseline, 7,032 (282 South Asian and 6,750 white participants) remained free of diabetes during the follow-up and were eligible for the present study. After the exclusion of 319 participants with unavailable glucose levels, insulin values or missing covariates, 417 non-fasting (<8 h) participants, and 317 persons because of an afternoon sampling, the final dataset included 5,979 participants (230 South Asian and 5,749 white, i.e. 13,063 person-examinations) (Fig. 1).
Of the potential 28,128 person-examinations that included four repeat measures for each of the 7,032 eligible participants, we excluded 6,665 observations with unavailable glucose and insulin values or missing covariates, 5,919 observations with non-fasting sampling (<8 h) and 2,481 observations due to afternoon sampling. Thus, the final dataset included 13,063 person-examinations for the 5,979 participants who took part in at least once of the four examinations every 5 years. Data were used for 87 South Asian participants with one observation, 70 with two observations, 48 with three observations and 25 with four observations, and for 1,692 white participants with one observation, 1,898 with two observations, 1,532 with three observations and 627 with four observations (Fig. 1).
Fasting and 2 h post-load venous blood samples were taken during a 75 g OGTT according to standardised protocols during all phases of the study. Blood glucose was measured with the glucose oxidase method (YSI Corporation, Yellow Springs, OH, USA). Serum insulin was measured with an in-house human insulin RIA and later with a DAKO ELISA kit (Dako Cytomation Ltd, Ely, UK) .
Prevalent cases of diabetes (South Asian n = 31, white n = 100) at phase 3 as well as incident cases were excluded based on a diagnosis either by an FG ≥7.0 mmol/l or a PLG ≥11.1 mmol/l during the OGTT (South Asian n = 40, white n = 391), by the use of glucose-lowering medication (South Asian n = 21, white n = 108) or by reporting doctor-diagnosed diabetes (South Asian n = 40, white n = 263) at screening or during the additional questionnaire phases .
HOMA2-%S and HOMA beta cell function (HOMA2-%B, a marker of insulin secretion) were calculated with the HOMA2 calculator version 2.2 (www.dtu.ox.ac.uk/homacalculator/index.php) using FG (acceptable range 3–25 mmol/l) and FINS (acceptable range 20–400 pmol/l) measurements . The BMI (the body weight in kilograms divided by the height in metres squared [kg/m2]) was measured according to standardised protocols.
Ethnicity was defined according to the Office for National Statistics 1991 census types. Self-reported ethnicity at phase 5 was mainly used; missing data were complemented by observer-assigned ethnicity from phase 1 . The rates of agreement between observer-assigned and self-reported ethnicity were high (93% for South Asian and 99.3% for white individuals). South Asians were defined as participants of Indian, Sri Lankan, Pakistani or Bangladeshi ethnic origin.
British Civil Service employment grade was used as a measure of occupational position and was grouped into three categories: high (senior administrators), intermediate (executives, professionals and technical staff) and low (clerical and office support staff) .
Physical activity was assessed by the answers to questions on the frequency and duration of participation in moderate or vigorous physical activity at phase 3. At phases 5, 7 and 9, these questionnaires included 20 items on the frequency and duration of different physical activities that were used to calculate the hours per week spent at each intensity level. Physical activity levels were classified as active (≥2.5 h/week of moderate or ≥1 h/week of vigorous physical activity), inactive (≤1 h/week of moderate and ≤1 h/week of vigorous physical activity) or moderately active (if not active or inactive) .
Dietary patterns were based on the type of bread and milk most frequently consumed and the frequency of fruit and vegetable consumption (on an 8-point scale). First, each indicator was scored from 1 to 3 points. For the type of bread, this was: healthy (wholemeal, wheat meal or other brown bread) = 1, moderately healthy (both healthy and unhealthy) = 2 or unhealthy (white bread) = 3. For the type of milk, the scoring was: healthy (no milk, skimmed milk or other) = 1, moderately healthy (semi-skimmed milk) = 2 or unhealthy (whole milk) = 3. For the frequency of fruit and vegetable consumption, the categories were: healthy (daily or ≥2 times per day) = 1, moderately healthy (≥1 times per week) = 2 or unhealthy (<3 times per month) = 3. The dietary score was calculated as the sum of the previous points classified into three groups: healthy (3 points), moderately healthy (4–7 points) or unhealthy (≥8 points) .
Two-sample t tests and χ
2 tests were conducted to compare means and proportions, respectively, at baseline between South Asian and white individuals. FINS and PLINS levels, HOMA2-%S and HOMA2-%B were natural log-transformed because of their skewed distribution.
Linear mixed models were used to assess age-related trajectories of glycaemic traits including FG, PLG, FINS, PLINS, HOMA2-%S and HOMA2-%B. These models use all available data and take into account the interrelationship between within-individual data points. First, we modelled differences in trajectories by ethnicity (South Asian or white) using the participants’ age (centred at 50 years) as the underlying time variable adjusted for sex, including ethnicity, age, age squared and their interaction terms and sex into the model. As the age squared by ethnicity interaction did not improve the model fit and the term itself was not significant, we dropped this term from all the models. Furthermore, our results suggested that the FG trajectory was following a linear increase, and thus the quadratic age term was also dropped from this model.
Second, we further adjusted for occupational grade and time-varying BMI, physical activity and dietary score. For easier comparability of the models, we used the same specification for these multivariable adjusted models as for the sex-adjusted model.
Due to the fact that some people had available data on only one of the six glycaemic traits used for analysis, the final number of participants for the individual glycaemic traits is lower than of the total population (n = 5,979): 5,849 for FG, 5,608 for PLG, 5,591 for FINS, 5,596 for PLINS and 5,344 for HOMA2-%S and HOMA2-%B (Fig. 1).
All statistical analyses were performed in Stata 12.1 (StataCorp, College Station, TX, USA), and statistical significance was inferred at a two-tailed p < 0.05.