Serum true insulin concentration and the risk of clinical non-insulin dependent diabetes during long-term follow-up

Transcription

1 International Epidemiological Association 1999 Printed in Great Britain International Journal of Epidemiology 1999;28: Serum true insulin concentration and the risk of clinical non-insulin dependent diabetes during long-term follow-up Ivan J Perry, a S Goya Wannamethee, a A Gerald Shaper a and K George MM Alberti b Background There is considerable evidence that insulin resistance with compensatory hyperinsulinaemia is an early and modifiable defect in the pathogenesis of non-insulin dependent diabetes (NIDDM). Current data, however, are largely based on studies that have used insulin assays which cross-react with proinsulin and other insulin precursors. Using a specific assay, we have addressed the hypothesis that an elevation of serum true insulin concentration, reflecting insulin resistance, is an early event in the pathogenesis of NIDDM. Methods Results Conclusion Keywords We have used a prospective cohort study design in which a group of 5550 nondiabetic men aged years, from 18 British towns, have been followed for incident cases of physician-diagnosed NIDDM for an average of period of 14.8 years (range years). We have estimated the incidence of physiciandiagnosed NIDDM by quintile of non-fasting serum true insulin concentration at entry into the study. There were 168 cases of clinically diagnosed NIDDM among the group of 5550 men during follow-up. Mean serum insulin at entry (geometric mean and 95% range, adjusted for time of sampling) was significantly higher in men who subsequently developed NIDDM than in the rest of the cohort, 19.5 mu/l ( ) versus 12.2 mu/l ( ), P There was a highly significant linear trend of increasing risk of NIDDM by quintile of serum insulin which was not attenuated substantially after adjustment for age and body mass index (BMI) and additional lifestyle and biological factors associated with serum insulin and risk of NIDDM. However, in men with non-fasting serum glucose 6.1 mmol/l at baseline (80th percentile, n = 1125, 82 cases), the risk of NIDDM, adjusted for age and BMI, was higher in the first quintile of serum insulin than in all other quintiles. These findings are consistent with the hypothesis that the majority of cases of adult onset NIDDM in this population are characterized by the early development of insulin resistance with compensatory true hyperinsulinaemia. NIDDM, serum true insulin, prospective study Accepted 13 January 1999 There is continuing debate regarding the primary defect in the pathogenesis of non-insulin dependent diabetes (NIDDM). Is it insulin hyposecretion due to a primary pancreatic β-cell defect or insulin resistance with hyperinsulinaemia and subsequent pancreatic exhaustion. 1 2 It is now increasingly accepted that a Department of Primary Care and Population Sciences, The Royal Free Hospital School of Medicine, London, UK. b Department of Medicine, University of Newcastle upon Tyne, UK. Reprint requests to: Professor IJ Perry, Department of Epidemiology & Public Health, Distillery House, University College Cork, Republic of Ireland. i.perry@ucc.ie both defects are required to produce clinically manifest NIDDM. It is likely that much of the controversy in this area reflects the varying sophistication of different methods of assessing early defects in insulin sensitivity and pancreatic hyposecretion. 3 Regardless of the primary defect in NIDDM, there is considerable evidence that insulin resistance with compensatory hyperinsulinaemia is an early and modifiable defect in the pathogenesis of this condition. 4 However, current data suggesting that hyperinsulinaemia is an early event in the pathogenesis of NIDDM 5 9 are largely based on insulin assays which have been unable to distinguish true insulin from proinsulin and other insulin precursors. It has therefore been argued that hyperinsulinaemia in 735

2 736 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY subjects who subsequently develop NIDDM is largely due to increased secretion of insulin precursors due to pancreatic β-cell failure. 10 In two prospective studies, involving Japanese-Americans 11 and elderly Finns, 12 proinsulin emerged as a stronger predictor of subsequent diabetes than true insulin during follow-up for 5 and 4 years, respectively. These findings may be regarded as a challenge to the concept of insulin resistance with compensatory true hyperinsulinaemia as an early abnormality in the pathogenesis of NIDDM, given that elevated proinsulin levels may be a marker for β-cell dysfunction. 10 It is clear however, that prospective data on insulin levels and risk of NIDDM from studies with longer follow-up are required. In this study we have measured serum insulin levels, using a specific assay, over a decade on average before the onset of clinically manifest diabetes. The objective of this study was to examine the relation between serum true insulin levels at screening and the risk of NIDDM during an average of 14.8 years follow-up. The primary focus was on the pathogenesis of NIDDM, on the question of whether true insulin levels were higher or lower in men who subsequently developed this condition. To assess possible heterogeneity in the aetiology of NIDDM between obese and non-obese men in this population, the insulin-niddm association has also been examined separately in each tertile of the distribution of body mass index. Subjects and Methods In the British Regional Heart Study (BRHS), 7735 men aged years, were selected at random (using an age-sex register) from one general practice in each of 24 towns in England, Wales and Scotland and examined between January 1978 and June 1980 for a prospective study of cardiovascular disease. The criteria for selecting the towns, general practices and subjects, and details of the respondents and data collection have been described previously. 13 Men with cardiovascular or other disease or those receiving regular medication were not excluded. The overall response rate was 78%. Aliquots of serum from the men in the 7th to the 24th towns visited, a total of 5661 men, were stored at 20 C. With exclusion of 111 men with either known diabetes at screening (n = 85), a diagnosis of diabetes in the calendar year in which they were screened (n = 9), or a non-fasting serum glucose concentration 11.1 mmol/l (200 mg/dl; n = 17),followup data were available for a group of 5550 non-diabetic men. Baseline assessment Research nurses administered a standard questionnaire and completed an examination of each man, which included a resting electrocardiogram. 14 The questionnaire included questions on smoking habits, alcohol intake, the usual pattern of physical activity, medical history and regular medication, including use of antihypertensive drugs. Details of the classification of smoking habits, alcohol intake, social class, the measurement of blood pressure and other physical measurements have been reported. 13 Body mass index (BMI) calculated as weight/height 2 was used as an index of relative weight. Heart rate was determined from the electrocardiogram. A physical activity score, was derived from the exercise questionnaire administered at the screening examination, based on the frequency and intensity of the activities reported. 15 Based on the score, the men were grouped into six broad physical activity categories: inactive, occasional, light, moderate, moderately-vigorous and vigorous. Forced expiratory volume in one second (FEV 1 ) was measured in the seated position, using a Vitalograph spirometer and values were height standardized. Prevalent coronary heart disease (CHD) at screening was defined on the basis of any or all of the following criteria: recall of doctor diagnosis of angina or heart attack, a WHO (Rose) questionnaire response indicating angina or possible myocardial infarction and electrocardiographic evidence of definite or possible myocardial ischaemia or infarction. 14,16 Non-fasting blood samples were obtained between 8.30 am and 6.30 pm. The time of arrival at the examination centre was noted and the estimated time of venepuncture (at the end of the examination) was 35 min later. 17 Details of venepuncture, serum separation and storage and the methods of analysis for serum lipids, uric acid and haematocrit have been described Insulin and glucose measurement Serum insulin concentration was determined by a two-site enzyme-linked immuno-adsorbent assay (ELISA) using commercially available monoclonal antibodies raised against human insulin (Novo Nordisk A/S: Denmark) which do not cross react with proinsulin. 21 Analyses were performed in the Department of Medicine, University of Newcastle upon Tyne, UK, on nonfasting samples which were stored at 20 C for years. In this laboratory, no change in insulin levels was detected in repeat assays (using a standard radioimmunoassay method) of 34 samples, stored at 20 C over an 8-year period (mean difference 0.19 mu/l, paired t = 0.7, P = 0.5). The lower limit of detection for the ELISA was 1 mu/l and the interassay coefficients of variation were 5.5% at 8.8 mu/l, 5.9% at 21.6 mu/l and 7.5% at 44.8 mu/l. The distribution of serum insulin was markedly skewed, range mu/l, median 11.6 mu/l, geometric mean (log sd) 12.4 (0.76) mu/l. There was diurnal variation in serum insulin levels, with peaks at 08.00, 09.00, 13.00, 14.00, and h, troughs at and h and intermediate levels at 10.00, 12.00, and h. Peak levels (hourly geometric mean) ranged from 13.6 mu/l to 18.4 mu/l, the trough levels were 10.3 mu/l and 10.4 mu/l and the intermediate levels ranged from 11.6 mu/l to 11.7 mu/l. 22 Glucose was analysed in serum at screening using a commercially available automated analyzer (Technicon SMA 12/60). Diurnal variation in glucose levels was modest, with a peaktrough difference of 0.4 mmol/l. 17,19 Follow-up All men have been followed up for all cause mortality and for cardiovascular morbidity up to December 1993, a mean period of 14.8 years (range years). 23 Information on death was collected through the established tagging procedures provided by the National Health Service registers in Southport (England and Wales) and Edinburgh (Scotland). New cases of NIDDM were ascertained by means of: (a) a postal questionnaire sent to the men at year 5 of follow-up for each individual, (b) systematic reviews of primary care records in 1990, 1992 and 1994 looking specifically for cases of NIDDM, (c) a further questionnaire to 6483 surviving members of the cohort resident in Britain in 1992, and (d) review of all death certificates for any mention of diabetes. 24 The questionnaire at year 5 achieved a

3 SERUM TRUE INSULIN AND NIDDM 737 response rate of 98% and the 1992 questionnaire a response rate of 91%. In the primary care record review, the records of each study participant (including discharge letters from hospital) were examined for a number of specific diagnoses, including diabetes. Inconsistencies between the questionnaire data and the clinical records were resolved by means of further review of the primary care records. A diagnosis of diabetes was not accepted on the basis of questionnaire data unless confirmed in the primary care records. Statistical analysis Cox s proportional hazards models were used to assess the independent contribution of serum insulin concentration at baseline to the risk of NIDDM at 14.8 years follow-up (range years) and to estimate the relative risk of NIDDM in each quintile of insulin relative to the first quintile, adjusted for other risk factors. 25 Age, BMI, systolic blood pressure, heart rate, uric acid, FEV 1, high density lipoprotein (HDL) cholesterol and triglyceride concentration were fitted as continuous variables in the proportional hazards model. Haematocrit was fitted as four dummy variables for five haematocrit groups, based on absolute levels of haematocrit: 42.0%, %, %, % and 48.0%. Physical activity was fitted as five dummy variables (six categories), and alcohol as four variables (five categories). Pre-existing CHD was fitted as a dichotomous (yes/no) variable. As insulin, glucose and triglyceride concentrations were not normally distributed, log transformation and geometric means were used. Because of the marked diurnal variation in serum insulin and triglyceride levels, 17,22 the log transformed data on these variables was adjusted for time of sampling, using the mean level of each variable for each hour in which samples were taken and the overall mean (Appendix). 22 To illustrate the separate effects of allowing for key lifestyle and biological variables, the insulin-niddm relation was adjusted for potential confounding factors in three cumulative stages: (a) age and BMI, (b) pre-existing CHD, physical activity, alcohol consumption, heart rate, FEV 1, systolic blood pressure, uric acid, haematocrit and HDL-cholesterol level and (c) triglyceride level (Table 2). Possible interactions between insulin and BMI and insulin and glucose in the development of NIDDM were explored in stratified analyses and by fitting interaction terms in Cox s proportional hazards models (Tables 3 and 4). In further analyses, to investigate possible bias due to time of sampling, the association between serum insulin concentration, unadjusted for time of sampling, and risk of NIDDM was examined in three time strata, at times of trough, intermediate and peak insulin levels, as detailed above (Table 5). Table 1 Incidence of physician diagnosed non-insulin dependent diabetes mellitus (NIDDM) per 1000 person-years of follow-up by quintile of serum insulin (non-fasting) No. of No. of Rate/1000 Insulin (mu/l) a men events person-years (geometric mean and 95% range, adjusted for time of sampling) were significantly higher in men who subsequently developed NIDDM than in the rest of the cohort, 19.7 mu/l ( ) versus 12.3 mu/l ( ), P There was a similar difference in serum insulin levels (unadjusted for time of sampling) between NIDDM cases and the rest of the cohort, 19.5 mu/l ( ) versus 12.2 mu/l ( ), P Serum insulin levels were weakly associated with age (r = 0.03) and strongly associated with BMI (r = 0.33; P 0.001). 22 On adjustment for age and BMI, serum insulin levels (geometric mean) remained significantly higher in those who developed NIDDM relative to the rest of the cohort, although the difference between the groups was attenuated, 16.4 mu/l versus 12.3 mu/l, P Table 1 shows the incidence of physician diagnosed NIDDM per 1000 person-years of follow-up by quintile of serum insulin. Figure 1 shows the relative risk of NIDDM with 95% CI in each quintile of the serum insulin distribution relative to the first Results Among the group of 111 men excluded from follow-up because of known or probable diabetes there were 87 men who were not receiving insulin therapy. Serum insulin levels, geometric mean (95% range) adjusted for time of sampling, were higher in this group of men than in the group of 5550 non-diabetic men, 17.5 mu/l ( ) versus 12.4 mu/l ( ), P By December 1993, after 14.8 years mean follow-up, 168 new cases of NIDDM had become manifest in the group of 5550 non-diabetic men. Mean serum insulin levels at baseline Figure 1 Relative risk of clinical non-insulin dependent diabetes mellitus (NIDDM) (log scale) adjusted for age and body mass index (BMI) with 95% CI, by quintile of baseline non-fasting serum insulin. The shows each relation adjusted for age alone. The number of cases of NIDDM in each quintile of serum insulin are shown. The insulin data were adjusted for time of sampling

4 738 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Table 2 Relative risk (95% CI) of incident cases of physiciandiagnosed non-insulin dependent diabetes mellitus (NIDDM) by quintile of serum insulin (non-fasting) Insulin (mu/l) a A B C ( ) 0.6 ( ) 0.6 ( ) ( ) 1.7 ( ) 1.5 ( ) ( ) 1.5 ( ) 1.3 ( ) ( ) 2.6 ( ) 2.3 ( ) Trend b P = P = P = Adjustments A = Adjusted for age and BMI (n = 5549; 168 cases). B = Adjusted for the factors in A plus physical activity, alcohol consumption, systolic blood pressure, heart rate, uric acid, FEV 1, haematocrit, HDLcholesterol and pre-existing coronary heart disease (n = 5026; 154 cases). C = Adjusted for the factors above plus triglyceride level (n = 4932; 148 cases). b Test for linear trend from the first to the fifth serum insulin quintile. adjusted for age and separately for age and BMI. The data were consistent with a J-shaped association between serum insulin concentration and the subsequent risk of NIDDM. There was a steady increase in the risk of NIDDM from the 2nd to the 5th quintile of the insulin distribution with a non-significantly higher risk in the first compared with the second quintile. The age-adjusted association between insulin and risk of diabetes was substantially attenuated on adjustment for BMI, although it remained significant (Table 2, model A). Adjustment for additional potential confounders and mediating factors (including serum HDL-cholesterol and triglyceride, which were highly correlated with serum insulin; r = 0.27 and r = 0.4, respectively, both P ) led to further attenuation of the association between non-fasting insulin and the subsequent risk of NIDDM (Table 2, models B and C). However, in the full multivariate analysis there remained a significant independent association between insulin levels and risk of NIDDM, (Table 2). Despite the apparent J-shaped association, tests for linear trend of increasing risk of NIDDM by quintile of serum insulin were highly significant in all models (Table 2). There was no evidence of a sharp increase in risk of NIDDM in the upper decile of the insulin distribution. On adjustment for age and BMI the relative risk of NIDDM in the 9th decile of the serum insulin distribution relative to the first quintile was 2.9 (95% CI : ), and in the 10th decile it was 3.1 (95% CI : ), similar to the risk in the fifth relative to the first quintile, 3.0 (95% CI : ). Stratification by serum glucose concentration at baseline The insulin-niddm association was further examined in two strata of non-fasting glucose concentrations, 6.1 mmol/l (80th percentile) and mmol/l (Table 3). The data were adjusted for age and then separately for age and BMI. In the lower glucose stratum, as in the entire group, there was a significant linear trend of increasing risk of NIDDM by quintile of serum insulin, although the risk of NIDDM was lowest in the second rather than in the first quintile of the insulin distribution. By contrast, in the upper glucose stratum there was no evidence of a linear increase in risk of NIDDM with increasing insulin levels after adjustment for age and BMI and risk was higher in the first insulin quintile (serum insulin 6.7 mu/l) than in each of the other quintiles. Tests for interaction were non-significant. Insulin-NIDDM association stratified by body mass index Table 4 shows the age-adjusted risk of NIDDM in each quintile of insulin relative to the first by tertile of BMI. A J-shaped relation between serum insulin concentration and risk of NIDDM, similar to that seen overall was seen among men in the third BMI tertile (BMI 26.6 kg/m 2 ) with the lowest risk observed in the second quintile of serum insulin. However, as in the entire group, the difference in risk between the first and second quintile did not attain statistical significance, possibly because of the small number of cases in the lower quintiles of serum insulin. In the second BMI tertile, the findings were consistent with a simple graded association between serum insulin and risk of NIDDM. In the lowest BMI tertile (BMI 24 kg/m 2 ), there were only 20 cases of NIDDM. In this group of men no significant association between serum insulin concentration and risk of NIDDM was observed. However, on formal testing for interaction no significant difference in the insulin-niddm association across the BMI tertiles was detected (P = 0.3). Table 3 Relative risk (95% CI) of incident cases of physician-diagnosed non-insulin dependent diabetes mellitus (NIDDM) by quintile of serum insulin in two glucose strata. The data are adjusted separately for age and for age and body mass index (BMI) Glucose I ( 6.1 mmol/l) Adjustments Glucose II ( mmol/l) Adjustments Insulin (mu/l) a Age Age and BMI Age Age and BMI ( ) 0.6 ( ) 0.2 ( ) 0.2 ( ) ( ) 2.1 ( ) 0.4 ( ) 0.3 ( ) ( ) 1.6 ( ) 0.5 ( ) 0.3 ( ) ( ) 2.4 ( ) 1.0 ( ) 0.5 ( ) Trend b P P = 0.01 P = 0.02 P = 0.9 Glucose I = First to the fourth glucose quintiles combined (n = 4411, 86 cases). Glucose II = Upper glucose quintile (n = 1125, 82 cases). b Test for linear trend from the first to the fifth serum insulin quintile.

5 SERUM TRUE INSULIN AND NIDDM 739 Table 4 Age-adjusted relative risk (95% CI) of incident cases of physician-diagnosed non-insulin dependent diabetes mellitus (NIDDM) by quintile of serum insulin in three body mass index (BMI) strata (tertiles) BMI I BMI II BMI III Insulin (mu/l) a (n = 1862) (n = 1842) (n = 1845) ( ) 1.6 ( ) 0.3 ( ) ( ) 4.4 ( ) 1.4 ( ) ( ) 1.5 ( ) 2.1 ( ) ( ) 4.7 ( ) 4.3 ( ) Trend b P = 0.5 P = P BMI I ( kg/m 2 ; 20 cases). BMI II (24.06 kg/m 2 ; 43 cases). BMI III (26.6 kg/m 2 ; 105 cases). b Test for linear trend from the first to the fifth serum insulin quintile. Table 5 Relative risk (95% CI) of incident cases of physiciandiagnosed non-insulin dependent diabetes mellitus (NIDDM) by quintile of unadjusted serum insulin, stratified by time of day at which blood samples were obtained, i.e. times of trough, intermediate and peak insulin levels. The data are adjusted for age and body mass index (BMI) Insulin Trough Intermediate Peak mu/l (n = 1319) (n = 2003) (n = 2228) ( ) 1.2 ( ) 0.6 ( ) ( ) 1.3 ( ) 0.5 ( ) ( ) 2.9 ( ) 1.1 ( ) ( ) 4.0 ( ) 1.7 ( ) Trend a P = P = P = Trough times at and h; 52 cases. Intermediate times at 10.00, 12.00, and h; 50 cases. Peak times at , 13.00, 14.00, and h; 66 cases. a Test for linear trend from the first to the fifth serum insulin quintile. Insulin/glucose ratio and risk of NIDDM The association between insulin/glucose ratio and risk of NIDDM was similar to that between insulin and NIDDM with a lower age- and BMI-adjusted risk (relative risk [RR] = 0.6; 95% CI : ) in the second than in the first quintile of insulinglucose ratio but with a steady increase in relative risk from the second to the fifth quintile relative to the first (RR 0.6, 1.4, 1.7, 2.1). Similar trends were observed in full multivariate analysis. On exclusion of cases diagnosed within the first 8 years of follow-up from the analysis, the J-shaped association between both insulin and insulin/glucose ratio and risk of NIDDM was unchanged. Insulin-NIDDM association by time of sampling In further analyses the association between insulin, unadjusted for time of sampling, and risk of NIDDM was examined in three strata defined by the time of day at which blood samples were obtained (Table 5). In each stratum the risk of NIDDM was highest among men with serum insulin in the fifth quintile of the distribution. Discussion In this longitudinal study, men who developed NIDDM had higher true insulin levels than the rest of the cohort more than a decade (on average) before the onset of clinically manifest disease. However, among men with non-fasting serum glucose in the fifth quintile, who were at particularly high risk of developing NIDDM, increasing serum insulin levels did not predict NIDDM and in this group the risk was highest among those with low serum true insulin levels ( 6.6 mu/l, 1st quintile). It is clear that given the method of case ascertainment used in this study, based on doctor diagnosis and death certification, there will be underreporting and misclassification of cases. However, these errors are likely to be random, and therefore should not affect the overall relation between insulin and diabetes. It would also have been preferable to have had data on fasting and/or post-load insulin. However, given the increased withinsubject variability (i.e. random measurement error) associated with non-fasting insulin data, it is likely that the difference in true insulin levels between those who did and did not develop NIDDM during follow-up has been underestimated. Moreover, regardless of the time of sampling, whether at meal times, midmorning or mid-afternoon, a serum insulin concentration in the upper quintile of the distribution was associated with increased risk of NIDDM during follow-up (Table 5). NIDDM pathogenesis A two-step model for the development of NIDDM has been proposed. 26 It is suggested that the first step, involving a transition from normal to impaired glucose tolerance (IGT), depends mainly on the development of insulin resistance with compensatory hyperinsulinaemia whereas the second step, worsening from IGT to diabetes, is primarily dependent on the development of β-cell dysfunction. The findings in this study are consistent with this model. The finding that higher true insulin levels antedate clinically manifest NIDDM by more than a decade supports the concept of an early role for insulin resistance with compensatory true hyperinsulinaemia in the pathogenesis of this condition. Moreover, on the basis of the two-step model of pathogenesis, one would anticipate that in the high-risk subgroup of men with high serum glucose at baseline, many would have progressed to the second step, to the development of β-cell dysfunction. Thus, one would not expect that serum insulin would predict NIDDM in this group of men. Adjustment for other predictors of NIDDM The association between insulin and NIDDM was attenuated after adjustment for BMI, reflecting the importance of obesity as a predictor of current insulin levels and future risk of NIDDM in this cohort. 24 By contrast, the effect of adding additional important predictors of NIDDM, including serum triglyceride, 24 to the multivariate model was relatively modest (Table 2). We have not adjusted for serum glucose level in multivariate analyses. As one would expect, serum glucose is a major predictor of NIDDM in this cohort. 24 However, this study has addressed the question of whether serum true insulin levels are elevated as an early event in the pathogenesis of NIDDM, not the biologically meaningless issue of whether insulin is independent of glucose in multivariate models predicting NIDDM.

6 740 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Previous studies The finding of true hyperinsulinaemia over a decade before the diagnosis of NIDDM is consistent with data from previous studies. The current evidence that hyperinsulinaemia predicts NIDDM is based mainly on studies that have used assays which cross-react with proinsulin. 5 9 However, Haffner et al. have shown true hyperinsulinaemia and a normal insulin:proinsulin ratio in Mexican Americans, 27 a high-risk population for NIDDM in whom hyperinsulinaemia on standard radioimmunoassay is well documented. 28 There is also direct evidence that insulin resistance antedates NIDDM. 29,30 The J-curve We observed a J-shaped association between non-fasting serum insulin levels (adjusted for time of sampling) and risk of NIDDM with a graded increase in risk from the 2nd to the 5th quintile of the insulin distribution but with a higher (although nonsignificant) risk in the first than in the second quintile. This observation is also consistent with the two-step model of NIDDM pathogenesis as one would anticipate that the lowest insulin group at baseline would include some men who were at high risk of NIDDM, having progressed to the stage of β-cell dysfunction. Against the latter interpretation, however, is the fact that the J-curve was not abolished following exclusion of men diagnosed within the first 8 years of follow-up. It should be noted, however, that in the context of the natural history of NIDDM, the difference in duration of follow-up between these groups was relatively small. Clearly, this J-curve should be interpreted cautiously given the small number of events in the first and second insulin quintiles (n = 26), the non-significant difference in NIDDM risk between the first and second quintile and the fact that measurement of non-fasting insulin levels reflects to a variable extent both basal and post-load insulin secretion. However the biological significance of the J-curve with regard to β-cell dysfunction and insulin hypo-secretion is enhanced by the fact that although it was found in virtually all subgroups examined, it was most apparent in subjects from whom blood samples were obtained at meal-times, i.e. at times of peak insulin secretion (Table 5). There is considerable evidence that NIDDM is a heterogeneous syndrome that may include conditions characterized by an early major defect in insulin secretion. 31 Hence, it may be that the somewhat higher risk of NIDDM seen during follow-up in the first relative to the second quintile of the insulin distribution reflects the inclusion of individuals with an early β-cell defect in the former group. Obesity is likely to be less important in the development of NIDDM in such individuals. In this context, it is noteworthy that serum insulin concentration did not predict NIDDM in those with a BMI 24 kg/m 2. Again however, the small number of cases in this subgroup precludes firm conclusions. Conclusion In this cohort of middle-aged British men elevated circulating true insulin levels antedate clinically manifest NIDDM by more than a decade. This finding supports the concept of an early role for insulin resistance with compensatory true hyperinsulinaemia in the pathogenesis of this condition. However, it does not exclude additional early defects of post-load insulin secretion. 3,32 Arguments about the relative importance of insulin resistance and β-cell dysfunction in the pathogenesis of NIDDM are of limited relevance. That both abnormalities contribute to the development of NIDDM, interacting at multiple levels, is not disputed. 3 However, the current data suggest that in the majority of cases of NIDDM in this population the development of insulin resistance is an early event, extending over many years or decades before the onset of obvious pancreatic β-cell decompensation. These findings emphasize the need to focus on lifestyle and environmental factors which increase insulin resistance, such as obesity, physical inactivity 24 and possibly cigarette smoking, 33 in our efforts to understand and prevent this common disorder. Acknowledgements The British Regional Heart Study is a British Heart Foundation Research Group. Support is also provided by The Stroke Association and the Department of Health (England and Wales). Biochemical estimations (with the exception of insulin and triglyceride) were carried out at the Wolfson Research Laboratories, Queen Elizabeth Hospital, Birmingham. SGW is a British Heart Foundation Research fellow. IJP was supported by the Wellcome Trust. We are grateful to the British Diabetic Association for laboratory support. References 1 O Rahily S, Hattersley A, Vaag A, Gray H. Insulin resistance as the major cause of impaired glucose tolerance: a self-fulfilling prophesy? Lancet 1994;344: Taylor SI, Accili D, Imai Y. Insulin resistance or insulin deficiency: which is the primary cause of NIDDM? Diabetes 1994;43: Weir GC. Which comes first in non-insulin dependent diabetes: insulin resistance or Beta cell failure? Both come first. JAMA 1995;273: DeFronzo RA. The triumvirate: β-cell, muscle and liver. A collusion responsible for NIDDM. Diabetes 1988;37: Sicree RA, Zimmett PZ, King HOM, Coventry JS. Plasma insulin response among Nauruans: prediction of deterioration in glucose tolerance over 6 years. Diabetes 1987;36: Bergstrom RW, Newell-Morris LL, Leonetti DL, Shuman WP, Wahl PW, Fujimoto WY. Association of elevated fasting C-peptide level and increased intra-abdominal fat distribution with development of NIDDM in Japanese American men. Diabetes 1990;39: Haffner SM, Stern MP, Mitchell BD, Hazuda HP, Patterson JK. Incidence of type II diabetes in Mexican Americans predicted by fasting insulin and glucose levels, obesity and body fat distribution. Diabetes 1990;39: Charles MA, Fontbonne A, Thibult N, Warnet JM, Rosselin GE, Eschwege E. Risk factors for NIDDM in white population. Paris Prospective Study. Diabetes 1991;40: Mykkänen L, Kuusisto J, Pyorala K, Laakso M. Cardiovascular disease risk factors as predictors of type 2 (non-insulin-dependent) diabetes mellitus in elderly subjects. Diabetologia 1993;36: Temple RC, Carrington CA, Luzio SD et al. Insulin deficiency in Non- Insulin-Dependent Diabetes. Lancet 1989;i: Kahn SE, Leonetti DL, Prigeon RL, Boyko EJ, Bergstrom RW, Fujimoto WY. Proinsulin as a marker for development of NIDDM in Japanese-American men. Diabetes 1995;44: Mykkanen L, Haffner SM, Kuusisto J, Pyorala K, Hales CN, Laakso M. Serum proinsulin levels are disproportionately increased in elderly prediabetic subjects. Diabetologia 1995;38:

7 SERUM TRUE INSULIN AND NIDDM Shaper AG, Pocock SJ, Walker M, Cohen NM, Wale CJ, Thomson AG. The British Regional Heart Study: cardiovascular risk factors in middle-aged men in 24 towns. Br Med J 1981;283: Shaper AG, Cook DG, Walker M, Macfarlane PW. Prevalence of ischaemic heart disease in middle-aged British men. Br Heart J 1984; 51: Shaper AG, Wannamethee G. Physical activity and ischaemic heart disease in middle-aged British men. Br Heart J 1991;66: Phillips AN, Shaper AG, Pocock SJ, Walker M. The role of risk factors in heart attack occurring in men with pre-existing IHD. Br Heart J 1988;60: Pocock SJ, Ashby D, Shaper AG, Walker M, Broughton PMG. Diurnal variations in serum biochemical and haematological measurements. J Clin Path 1989;42: Thelle DS, Shaper AG, Whitehead TP, Bullock DG, Asbhy D, Patel I. Blood lipids in middle-aged British men. Br Heart J 1984;49: Cook DG, Shaper AG, Thelle DS, Whitehead TP. Serum uric acid, serum glucose and diabetes: relationships in a population study. Postgrad Med J 1986;62: Wannamethee G, Shaper AG. Haematocrit: relationships with blood lipids, blood pressure and other cardiovascular risk factors. Thrombosis Haemostasis 1994;72: Andersen L, Dinesen B, Jorgensen PN, Poulsen F, Roder ME. Enzyme immunoassay for intact human insulin in serum or plasma. Clin Chem 1993;39: Perry IJ, Wannamethee G, Whincup PH, Shaper AG, Walker MK, Alberti KGMM. Serum insulin and incidence of coronary heart disease in middle-aged men British men. Am J Epidemiol 1996;144: Walker M, Shaper AG. Follow-up of subjects in prospective studies in general practice. J R Coll Gen Pract 1984;34: Perry IJ, Wannamethee SG, Walker MK, Thomson AG, Whincup PH, Shaper AG. Prospective study of risk factors for development of noninsulin-dependent diabetes in middle-aged British men. Br Med J 1995;310: Cox DR. Regression models and life-tables. J R Stat Soc [B] 1972;34: Saad MF, Knowler WC, Pettitt DJ, Nelson RG, Charles MA, Bennett PH. A two-step model for development of non-insulin dependent diabetes. Am J Med 1991;90: Haffner SM, Bowsher RR, Mykkanen L et al. Proinsulin and specific insulin concentration in high- and low-risk populations for NIDDM. Diabetes 1994;43: Haffner SM, Stern MP, Hazuda HP, Pugh JA, Patterson JK. Hyperinsulinaemia in a population at high risk for non-insulin dependent diabetes mellitus. N Engl J Med 1986;315: Lillioja S, Mott DM, Spraul M et al. Insulin resistance and insulin secretory dysfunction as precursors of non-insulin-dependent diabetes mellitus. Prospective studies of Pima Indians. N Engl J Med 1993;329: Warram JH, Martin BC, Krolewski AS, Soeldner JS, Kahn R. Slow glucose removal rate and hyperinsulinemia precede the development of type 2 diabetes in the offspring of diabetic parents. Ann Int Med 1990;113: Alberti KGMM. Problems related to definitions and epidemiology of Type 2 (non-insulin dependent) diabetes mellitus: studies throughout the world. Diabetologia 1993;36: Haffner SM, Miettinen H, Gaskill SP, Stern MP. Decreased insulin secretion and increased insulin resistance are independently related to the 7-year risk of NIDDM in Mexican-Americans. Diabetes 1995; 44: Rimm EB, Chan J, Stampfer MJ, Colditz GA, Willett WC. Prospective study of cigarette smoking, alcohol use, and the risk of diabetes in men. Br Med J 1995;310: Appendix Method adjustment of serum insulin and triglyceride data for time of sampling Serum insulin and triglyceride levels were adjusted for time of sampling using a simple mathematical approach which makes no assumptions about the form of the association between these variables and time of sampling. The log-transformed data on these variables were adjusted for time of sampling, using the mean level of each variable for each hour in which samples were taken and the grand mean. From each individual value, the mean for the hour of sampling for that individual was subtracted, and the result was added to the grand mean. Hence, each value was adjusted to one an equal distance away from the grand mean as it was from the mean for the hour in which it was measured. The calculation for serum insulin was as follows: adjusted log insulin level = (unadjusted log insulin level the mean log insulin level for the hour of sampling) + the grand mean log insulin level.