DIABETES MELLITUS type 2 is a multifactorial disease

Transcription

1 X/00/$03.00/0 Vol. 85, No. 9 The Journal of Clinical Endocrinology & Metabolism Printed in U.S.A. Copyright 2000 by The Endocrine Society Risk for Diabetes Mellitus in Middle-Aged Caucasian Male Participants of the PROCAM Study: Implications for the Definition of Impaired Fasting Glucose by the American Diabetes Association ARNOLD VON ECKARDSTEIN, HELMUT SCHULTE, AND GERD ASSMANN Institut für Klinische Chemie und Laboratoriumsmedizin, Zentrallaboratorium, Westfälische Wilhelms-Universität Münster, D Münster, Germany; and Institut für Arterioskleroseforschung an der Universität Münster, D Münster, Germany DIABETES MELLITUS type 2 is a multifactorial disease involving genetic predisposition and various environmental factors (1). Established risk factors include overweight, an unfavorable body fat distribution, hyperinsulinemia, and impaired glucose tolerance (2 14). Several prospective studies, mostly performed in elderly men or in populations with a high incidence of diabetes mellitus, have also observed that elevated blood pressure, elevated serum concentrations of total and low density lipoprotein (LDL) cholesterol and triglycerides, the presence of small dense LDL as well as low high density lipoprotein (HDL) cholesterol levels predict the future occurrence of diabetes mellitus type 2 (15 24). However, because of the important role of insulin resistance in the pathogenesis of both diabetes mellitus and these cardiovascular risk factors, these components of the metabolic syndrome are strongly confounded with overweight and glucose intolerance (25 27). It is therefore not clear whether they are independent risk factors. To solve this question we investigated prospectively by multivariate statistical methods in a subpopulation of 3737 men, aged yr, from the Prospective Cardiovascular Münster (PROCAM) Study the roles of genetic predisposition, age, Received December 22, Revision received April 12, Accepted May 16, Address all correspondence and requests for reprints to: Dr. Arnold von Eckardstein, Institut für Klinische Chemie und Laboratoriumsmedizin, Zentrallaboratorium, Westfälische Wilhelms Universität Münster, Albert Schweitzer Strasse 33, D Munster, Germany. vonecka@uni-muenster.de. ABSTRACT The criteria of the American Diabetes Association and the WHO for the diagnosis of diabetes mellitus are controversially discussed. In a prospective population study, we evaluated the data of 3737 men, aged yr, without diabetes mellitus and with fasting serum glucose levels less than 7 mmol/l at entry into the study who had at least 1 repeat examination during a follow-up of 4 10 yr. During a mean follow-up of 6.3 yr, 200 men developed diabetes mellitus. They differed significantly from 3537 men by body mass index, fasting serum levels of glucose, high density lipoprotein cholesterol, and family history positive for diabetes mellitus. Receiver operating curve analysis revealed that a glucose level of 5.72 mmol/l was the best discriminatory cut-off. Upon global risk estimation by multiple logistic function (MLF) analysis, 69.6% of all diabetes mellitus incidences occurred in the highest quintile as defined by the MLF algorithm. The relative risk of a men in this quintile was 8.7 compared to that in the residual population. The performance of risk assessment by MLF as estimated by the area under the receiver operator characteristic curve was similar to fasting glucose levels. Global risk estimation by multiple risk factors does not improve the prediction of diabetes mellitus by fasting glucose in middle-aged men. The lower discriminatory cut-off of 5.72 mmol/l glucose may help to reduce the previously reported discordance between impaired fasting glucose (American Diabetes Association) and impaired glucose tolerance (WHO) in diagnosis of the prediabetic state. (J Clin Endocrinol Metab 85: , 2000) overweight, glucose, dyslipidemia, hypertension, and hyperuricemia as risk factors for the future development of diabetes mellitus type 2. Moreover, as risk assessment of coronary heart disease has been much improved by multiple logistic function analysis of several risk factors (28, 29), we also tested whether the combination of information from potential risk factors of diabetes mellitus by this statistical method would improve the prediction of diabetes mellitus. Finally, the American Diabetes Association (ADA) and the WHO have defined new criteria for the diagnosis of diabetes mellitus and the prediabetic state (30, 31). ADA defines diabetes mellitus by the presence of clinical symptoms and/or the (repeated) finding of fasting glucose levels of 7 mmol/l or more. This cut-off was chosen because it has the best agreement with the diagnosis of diabetes mellitus by the finding of postprandial glucose levels of 11.2 mmol/l or more in the oral glucose tolerance test, which in epidemiological studies of the occurrence of microangiopathic complications of diabetes mellitus has evolved as a gold standard in the diagnosis of diabetes mellitus (32). By consensus, but without any epidemiological data basis, the ADA has defined the prediabetic state as the finding of impaired fasting glucose levels of mmol/l (30). Although accepting impaired glucose tolerance as a risk factor for diabetes mellitus, the ADA no longer recommends oral glucose testing because of practical reasons (30). By contrast, the WHO continues to require the performance of oral glucose tolerance testing for individuals with a fasting glucose level between 5.6 and 11.2 mmol/l and extends the definition of diabetes 3101

2 3102 VON ECKARDSTEIN, SCHULTE, AND ASSMANN JCE&M 2000 Vol. 85 No. 9 to the finding of 2 h glucose levels being higher than 11.2 mmol/l. A prediabetic state is defined as impaired glucose tolerance, with 2 h glucose levels ranging from mmol/l (31, 33). Previous studies, which compared the WHO and ADA criteria, showed a various degree of concordance with respect to diagnosis of manifest diabetes mellitus and great discordance with respect to diagnosis of the prediabetic state (34 38). Moreover, compared to impaired glucose tolerance (WHO), impaired fasting glucose (ADA) had a weaker association with cardiovascular morbidity and all-cause mortality (39, 40). It is hence assumed that the ADA criteria underestimate the burden of glucose disorders (41). Therefore and because of the lack of epidemiological data on this issue, we also analyzed the data for the best cut-off of fasting glucose that identifies men at risk for diabetes mellitus. Subjects and Methods Probands and follow-up The PROCAM study began in 1979 and examined people at work (employees of 52 companies and authorities) for cardiovascular risk factors, mortality, and cardiovascular events, including myocardial infarction and stroke (42). The recruitment phase was completed at the end of Full data records are held for 13,737 male participants, aged yr, and 5,961 women, aged yr. The examination at study entry included case history using standardized questionnaires, measurement of blood pressure and anthropometric data, a resting electrocardiogram, and collection of a blood sample after a 12-h fast for the determination of more than 20 laboratory parameters. The examination was carried out during paid working hours. Participation was voluntary (between 40 80% took part; average, 60%), and free of charge both to the volunteers and to their employers (apart from loss of work). All findings were reported to the participant s general practitioner, and the volunteer was told whether the results of the examination were normal or whether a check-up by the general practitioner might be necessary. The investigators neither carried out nor arranged for any intervention (42). For the present study we selected the data of 3,951 men, aged yr, who had at least 1 follow-up examination within 4 10 yr. Data for 199 men were excluded from the statistical analysis because they had diabetes mellitus at entry into the study (Table 1). Thus, we investigated the roles of various risk factors in the incidence of TABLE 1. Prevalence and incidence of diabetes mellitus in 36- to 60-yr-old men who were recruited by the PROCAM study and had follow-up examinations within 4 10 yr Total population of 36- to 60-yr-old men with repeat 3951 examinations during 4 10 yr of follow-up Excluded because of known diabetes mellitus without 46 drug treatment Excluded because of known diabetes mellitus treated 19 with oral anti-diabetic drugs Excluded because of known diabetes mellitus treated 11 with insulin Excluded because of fasting glucose 7 mmol/l 123 No diabetes mellitus and included in the prospective study 3737 Newly developed known diabetes mellitus without drug 20 treatment Newly developed known diabetes mellitus treated with 10 oral antidiabetic drugs Newly developed known diabetes mellitus treated with 1 insulin Newly developed but unknown diabetes mellitus detected 169 by fasting glucose 7 mmol/l Sum of cases with newly developed diabetes mellitus 200 ( cases) No diabetes mellitus during follow-up ( controls) 3537 diabetes mellitus in a cohort of 3,737 men who had no manifest diabetes mellitus at baseline and who underwent at least 1 other examination during follow-up. If a subject underwent more than 1 examination during this period, we used the data from the final visit for statistical evaluation. Proband and family history and anthropometric measurements A family history of diabetes mellitus was defined as positive if at least one first or second degree relative had diabetes mellitus. Anyone smoking at least one cigarette per day within the last 12 months was considered a current smoker. Systolic and diastolic readings were taken from the left arm with the subject seated and the arm at heart level. One measurement was taken at the start of the interview by the examining physician, and one was taken at the end of the interview. The second measurement was recorded (43). For determination of body weight the probands were dressed in only underwear. Body mass index was calculated as the ratio of weight (kilograms) to the square of height (meters). Blood taking and biochemical measurements Serum was taken by venipuncture after at least 12 h of fasting. Serum was prepared after 1 2 h of clotting time by centrifugation at 2000 g. All biochemical parameters were measured by enzymatic methods using assays and the Hitachi 737 autoanalyzer from Roche (Mannheim, Germany). Glucose was determined by the hexokinase method, uric acid by the uricase method, cholesterol by the CHOD-PAP method, triglycerides by the GPO-PAP method, and HDL cholesterol by the CHOD- PAP method after precipitation of apolipoprotein B-containing lipoproteins with phosphotungstic acid/mgcl 2. LDL cholesterol was calculated by the Friedewald formula, if triglycerides were less than 4.6 mmol/l (44). Imprecision was below 2% for glucose and uric acid, below 3% for total cholesterol, below 4% for triglycerides, and below 5% for HDL cholesterol. We compared glucose levels in serum and plasma of 50 volunteers with serum glucose levels ranging from mmol/l. At a coefficient of correlation of 0.99, the linear regression equation was: plasma glucose (mmol/l) serum glucose (mmol/l) 0.04 (mmol/l). TABLE 2. Mean values of age-standardized factors for male participants in the PROCAM study, aged yr with (DM ) and without (DM ) development of diabetes mellitus within 4 10 yr (mean, 6.3 yr) of follow-up DM (n 3537) DM (n 200) Follow-up (months) a Age (years) BMI (kg/m 2 ) b Systolic blood pressure b (mm Hg) Diastolic blood pressure b (mm Hg) Glucose (mmol/l) b Cholesterol (mmol/l) Triglycerides (mmol/l) c b HDL cholesterol (mmol/l) b LDL cholesterol (mmol/l) Uric acid ( mol/l) b DM, Diabetes mellitus; BMI, body mass index; HDL, high-density lipoprotein; LDL, low-density lipoprotein; PROCAM, Prospective Cardiovascular Münster. a P 0.05 (by Student s t test). b P (by Student s t test). c Triglycerides are presented as geometric means because of their non-gaussian frequency distribution. Student s t test was performed on lg e -transformed data.

3 RISK FACTORS FOR DIABETES MELLITUS 3103 TABLE 3. Relative age-standardized risks of 36- to 60-yr-old men and women for the development of diabetes mellitus within 4 10 yr (mean, 6.3 yr) according to tertiles of various risk factors Parameter and tertiles Tertile limits Definitions of diabetes mellitus, impaired fasting glucose, and arterial hypertension A subject was defined as affected by diabetes mellitus if this diagnosis was known to the patient or, according to the ADA definitions, if fasting serum glucose was 7 mmol/l or more. Impaired fasting glucose was defined as a fasting serum glucose level of 6.1 mmol/l or more but less than 7 mmol/l (30). A proband was considered hypertensive when he knew this diagnosis and was treated with antihypertensive drugs or, in accordance with the definition by the WHO (45), when systolic and diastolic blood pressures were 160 and/or 95 mm Hg or more, respectively. Borderline hypertension was defined by a systolic blood pressure of 140 mm Hg or more, but less than 160 mm Hg and/or a diastolic blood pressure of 90 mm Hg or more, but less than 95 mm Hg. Incidence (%) Relative risk (95% confidence interval) Glucose (mmol/l) (ref.) ( ) ( ) BMI (kg/m 2 ) (ref.) ( ) ( ) Age (yr) (ref.) ( ) ( ) Diastolic blood pressure (mm Hg) (ref.) ( ) ( ) Systolic blood pressure (mm Hg) (ref.) ( ) ( ) Triglycerides (mmol/l) (ref.) ( ) NS ( ) HDL cholesterol (mmol/l) ( ) ( ) NS (ref.) Uric acid ( mol/l) (ref.) ( ) NS ( ) Cholesterol (mmol/l) (ref.) ( ) NS ( ) NS LDL cholesterol (mmol/l) (ref.) ( ) NS ( ) NS BMI, Body mass index; HDL, high-density lipoprotein; LDL, low-density lipoprotein. P values were calculated by 2 test. Statistics An explorative analysis was performed using the statistical package for the social sciences (SPSS-X) (46) and the statistical analysis system (SAS Institute, Inc., Cary, NC) (47). Because of the strong age dependency of diabetes mellitus, all data analyzed by univariate methods were adjusted for age. Because of their non-gaussian frequency distribution, triglycerides data were evaluated after log e transformation. Comparisons between groups were performed with Student s t test for continuous variables and the 2 test for discrete variables. The relative risks were calculated using the category with the lowest incidence of diabetes mellitus as the reference. The simultaneous contributions of several factors to the risk of diabetes mellitus were analyzed using a multiple logistic model of those factors that upon univariate analysis had a significant association with the incidence of diabetes mellitus, namely age, body mass index (BMI), HDL cholesterol, triglycerides, uric acid, blood pressure, glucose levels, and family history positive for diabetes mellitus. Forward and backward selections were used to build up the logistic regression model. Both procedures were modified in that at each point of the selection process the partial significance of each term included in or excluded from the model was reviewed. In the analyses the criterion for a variable to enter and to remain in the model was that its initial probability value as well as its partial probability value in the presence of other variables should not exceed Maximum likelihood statistics were used for the selection process. Initially, the population was split in half to derive multiple logistic models from the data of one half of the population and to test it in the other half. As the results of the two procedures did not differ significantly, only results from a third model are presented, which included the data for the entire population. P

4 3104 VON ECKARDSTEIN, SCHULTE, AND ASSMANN JCE&M 2000 Vol. 85 No. 9 TABLE 4. MLF analysis of risk factors for the development of diabetes mellitus of 36- to 60-yr-old men during a 4- to 10-yr (mean, 6.3 yr) follow-up of the PROCAM study FIG. 1. Incidence of diabetes mellitus among male PROCAM participants, aged yr, over a 4- to 10-yr period according to risk estimated by MLF analysis. The MLF algorithm was based on the risk factors and their estimates (cf. Table 4) assessed in 3737 men, 200 of whom developed diabetes mellitus. The formula reads I 1/(1 exp( y)), with y age (yr) glucose (mmol/l) BMI (kg/m 2 ) HDL cholesterol (mmol/l) family history of diabetes mellitus (no 0; yes 1) hypertension (no 0; borderline 1; manifest 2). Results Univariate associations between risk factors and diabetes mellitus At entry into the PROCAM study 3737 men, aged yr, were free of diabetes mellitus as defined by their unawareness of this diagnosis and by the finding of fasting glucose levels being below 7 mmol/l. At a repeat examination after 4 10 yr of follow-up (mean, 6.3 yr), 200 men were diagnosed to have developed diabetes mellitus either because in the meantime an antidiabetic treatment has been started or because they had fasting glucose levels of 7 mmol/l or more (Table 1). Men who developed diabetes mellitus had a 2 months longer mean follow-up interval than men who did not develop diabetes mellitus (P 0.001). Compared to nondiabetic men, men who had become diabetic during follow-up had a 7% higher mean BMI, 5% higher mean systolic and diastolic blood pressures, 11% higher mean fasting serum levels of glucose, 35% higher median triglyceride levels, and 7% higher mean levels of uric acid as well as 7% lower mean levels of HDL cholesterol (Table 2; all P 0.001, by t test). Fasting glucose was impaired in 49% of men who later became diabetic compared to 9.6% who did not develop diabetes mellitus during follow-up (P 0.001, by 2 test). Thirty-one percent of men with subsequent diabetes mellitus, but only 19.4% of men without subsequent diabetes mellitus (P 0.001, by 2 test), had at least one first or second degree relative with diabetes mellitus. Parameter estimate SE P (by 2 test) Age ( yr) Body mass index ( 10 kg/m 2 ) Hypertension (yes 1/no 0) Glucose ( mmol/l) Family history of diabetes mellitus (yes 1/no 0) Triglycerides ( mmol/l) HDL cholesterol ( mmol/l) Uric acid ( mol/l) Table 3 describes the relative risks for future diabetes mellitus associated with tertiles of various potential risk factors for diabetes mellitus. As expected, the highest relative risk was associated with increasing glucose levels. The relative risk for diabetes mellitus also significantly increased with increasing age, BMI, systolic and diastolic blood pressures, triglycerides, and uric acid as well as with decreasing HDL cholesterol. Total cholesterol and LDL cholesterol had no statistically significant association with the risk of diabetes mellitus. Multivariate associations between risk factors and diabetes mellitus Upon multiple logistic function (MLF) analysis glucose, BMI (both P 0.001), age, glucose, HDL cholesterol, hypertension, and family history (all P 0.05), but not triglycerides and uric acid, were significantly and independently associated with the future occurrence of diabetes mellitus (Table 4). In Fig. 1 we stratified the incidence rate of diabetes mellitus for quintiles of estimated risk, which was calculated by MLF analysis. According to this model, 69.6% of all new diabetes mellitus cases occurred in the highest quintile, whereas only 3.6%, 8.2%, 5.2%, and 13.6% occurred in the first to fourth quintiles. Thus, the risk of future diabetes mellitus was increased by a factor of 19.3 in individuals in the highest quintile compared to individuals in the lowest quintile or by a factor of 8.7 compared to individuals in the first to fourth quintiles. In Fig. 2 we compared the receiver operating characteristics (ROC) curves of the MLF model and of some independent single risk factors for diabetes mellitus. The areas under the curve, which reflect the performance of a test, were similar for the MLF risk estimates [Fig. 2A; 79.3%; 95% confidence interval (CI), %] and glucose (Fig. 2B; 79.9%; 95% CI, %), but significantly higher than those for BMI (Fig. 2C; 66.3%; 95% CI, %) and HDL cholesterol (Fig. 2D; 59.5%; 95% CI, %). In agreement with the similar areas under the ROC curves, sensitivity and positive predictive value of glucose and MLF risk estimates were similar when specificity was defined as 80% or 90% (Table 5). Impaired fasting glucose, defined according to the ADA guidelines (31), had a diagnostic sensitivity and specificity toward the prediction of diabetes mellitus of 51% and 91%, respectively. At the same specificity, the diagnostic sensitivity of the MLF risk estimate was 53%. At this cut-off of estimated MLF risk, 76% of men had a fasting glucose of 6.1 mmol/l or more and less than 7 mmol/l; 28.5% of men with this cluster developed diabetes mellitus compared to 9.4% of men who had fasting glucose of 6.1 mmol/l or more but a MLF risk estimate below this threshold value. Thus,

5 RISK FACTORS FOR DIABETES MELLITUS 3105 FIG. 2. ROC curves of MLF risk estimates and single risk factors for diabetes mellitus. A, MLF model; B, glucose levels; C, BMI; D, HDL cholesterol. Sensitivity and specificity were calculated on the basis of data from 3737 men, 200 of whom developed diabetes mellitus within 4 12 yr of follow-up. Variables of the MLF models are described in Table 3. Areas under the ROC curves are 79.3% (95% CI, %) for MLF risk estimates (A), 79.9% (95% CI, %) for fasting glucose (B), 66.3% (95% CI, %) for BMI (C), and 59.5% (95% CI, %) for HDL cholesterol (D). The boxes within the curves together indicate the cut-off value with the best discriminatory power. TABLE 5. Sensitivity and positive predictive value of fasting glucose and MLF risk estimates in the prediction of diabetes mellitus during 4 10 yr of follow-up of 35- to 60-yr-old men % (95% confidence intervals) Fasting glucose MLF 80% specificity Sensitivity 67.5 ( ) 69.5 ( ) Positive predictive value % specificity Sensitivity 54.0 ( ) 57.0 ( ) Positive predictive value despite similar performance characteristics, the MLF risk estimate appears to provide prognostic information for diabetes mellitus in addition to that provided by fasting glucose testing. It is however important to note that a glucose level of 5.72 mmol/l, rather than 6.1 mmol/l, was the best discriminatory cut-off for the identification of men with future diabetes mellitus. At this cut-off, sensitivity and specificity were 75.0% and 72.7%, respectively. Interactions between risk factors for diabetes mellitus Further examples for the interaction of risk factors in the prognosis of diabetes mellitus are shown in Figs. 3 and 4. Figure 3 demonstrates the interaction between glucose levels and other risk factors for diabetes mellitus. In both groups of individuals with normal ( 6.1 mmol/l) and impaired fasting glucose (6.1 7 mmol/l), increases in BMI (Fig. 3A) or blood pressure (Fig. 3B) as well as a decrease in HDL cholesterol (Fig. 3C) increased the risk of diabetes mellitus. The joint occurrence of impaired fasting glucose with overweight, hypertension, or low HDL cholesterol had prevalences of 2.0%, 4.8%, and 3.0%, respectively, but accounted for about 30%, 38%, and 32% of all new cases of diabetes mellitus. Also, interactions among BMI, blood pressure, and HDL cholesterol formed gradients of increasing risk for diabetes mellitus (Fig. 4). The risk of diabetes mellitus associated with decreasing levels of HDL cholesterol was further increased by increasing BMI (Fig. 4A) and increasing blood pressure (Fig. 4B). Discussion In this prospective population study of middle-aged German men, we identified, in the order of associated relative risks, fasting glucose levels, BMI, family history positive for diabetes mellitus, age, diastolic and systolic blood pressures, triglycerides, HDL cholesterol, and uric acid as risk factors for future diabetes mellitus. These risk factors have also been identified in other prospective studies of risk factors for

6 3106 VON ECKARDSTEIN, SCHULTE, AND ASSMANN JCE&M 2000 Vol. 85 No. 9 FIG. 3. Determination of risk for diabetes mellitus by interaction of glucose levels with body mass index (A), hypertension (B), and HDL cholesterol (C). Normal glucose levels, below 6.1 mmol/l (n 3315); impaired fasting glucose levels, 6.1 mmol/l or more but less than 7 mmol/l (n 413). Manifest hypertension was defined as the report of diagnosed hypertension and antihypertensive drug treatment or the finding of systolic blood pressure above 160 mm Hg and/or diastolic blood pressure of 95 mm Hg or more (n 592). Borderline hypertension was defined by systolic blood pressure of 140 mm Hg or more but less than 160 mm Hg and/or diastolic blood pressure of 90 mm Hg or more but less than 95 mm Hg (n 979). The number of normotensive individuals was For the definition of tertiles, see Table 3. The numbers on the bars represent the prevalences (percentages) of the various conditions in the PROCAM population. diabetes mellitus (15 24). Those studies, however, have been performed in populations with higher risk of diabetes mellitus, such as Mexican Americans (1.5%), Finns, and Pima FIG. 4. Determination of risk for diabetes mellitus by interactions of HDL cholesterol with BMI (A) or hypertension (B). For the definition of tertiles, see Table 3. The numbers on the bars represent the prevalences (percentages) of the various conditions in the PROCAM population. Indians (3.6%), where the incidence of diabetes mellitus amounted to %/yr (3, 7, 9, 15, 18, 24) compared to 0.85%/yr in our study. Thus, our data are in agreement with those reported by Haffner and colleagues, who found that both metabolic parameters and measures of insulin sensitivity predicted diabetes mellitus equally well in high risk (i.e. Mexican Americans) and low risk (i.e. non-hispanic whites) populations (48). Upon multivariate analysis, only glucose, BMI, hypertension, low HDL cholesterol, and family history positive for diabetes mellitus were independent risk factors for diabetes mellitus. Of course, statistical independence does not imply biological independence, as obesity and insulin resistance play important roles in the pathogenesis of both diabetes mellitus and the above-mentionned metabolic cardiovascular risk factors (25 27). Clustering of several risk factors within one individual aggravates the risk of diabetes mellitus in a dose dependent-manner. Mykkänen and colleagues previously associated the number of risk factors with the risk of diabetes mellitus in 65- to 74-yr-old Finnish individuals (18). They found that the relative risk of diabetes mellitus increased from 3.6 in individuals with a single risk factor to 59

7 RISK FACTORS FOR DIABETES MELLITUS 3107 in individuals with four risk factors, namely impaired glucose tolerance, hypertriglyceridemia, low HDL cholesterol, and hypertension. However, this dichotomic approach does not take into account the graded relationships between risk factors and disease. Moreover, despite the strong associated relative risk, this approach has only limited sensitivity. For example, in the PROCAM study, 38% of individuals with impaired fasting glucose and hypertension developed diabetes mellitus. However, as this condition had a prevalence of only 2.9%, nearly 80% of individuals with future diabetes mellitus did not present with this cluster. These disadvantages in risk assessment by counting risk factors can be overcome by the use of MLF algorithms that take into account the graded relationship between risk factors and disease. On the basis of the coefficients of our multivariate analysis of single risk factors we calculated an MLF algorithm that helped to estimate the risk of 35- to 60-yr-old men to develop diabetes mellitus in the next 4 10 yr (mean, 6.3 yr). Thereby, we characterized a 20% subgroup of the population in which the incidence of diabetes mellitus was 18.0% compared to 2.07% in the residual population (relative risk, 8.7). In fact, 69% of all incidences of diabetes mellitus occurred in this subgroup. According to the area under the ROC curve the risk estimation by MLF was equal to that by fasting glucose (79.3% vs. 79.9%). Impaired fasting glucose, as defined by the new ADA recommendations (30), is thought to reflect a prediabetic state and hence to have a high sensitivity and specificity toward the identification of individuals with future diabetes mellitus. In the PROCAM population, a fasting glucose level of mmol/l was associated with a relative risk of 8.9 and had a diagnostic sensitivity of 51% and a specificity of 91% in the prediction of future diabetes mellitus. At the same specificity, the sensitivity of the risk estimation by MLF was slightly higher, namely 53%, and the associated relative risk was 9. This underscores the predictive value of glucose levels in the identification of men at risk for diabetes mellitus, even if they are measured only in the fasting state (30). Using ROC analysis, we found evidence that the optimal cut-off for the identification of men at increased risk for diabetes mellitus should be even lower, namely 5.72 mmol/l. Several studies revealed the reduced sensitivity of impaired fasting glucose (ADA) and impaired glucose tolerance (WHO) toward the prediction of diabetes mellitus (34 38) and death and the association with cardiovascular morbidity (39, 40). The lower discriminatory cut-off of 5.72 mmol/l found in our study may hence help to overcome this shortcoming of the ADA criteria. By method comparison we found that glucose levels in plasma are about 4% lower than those in serum, which was used for our measurement of glucose. As a consequence, the cut-off for fasting plasma glucose may be even lower, namely 5.5 mmol/l. The PROCAM study is a field study. As a consequence, we only measured fasting glucose and did not repeat glucose measurements for confirmation of the diagnosis. Because of the first limitation, we probably included diabetic individuals at baseline and overlooked other diabetic individuals during follow-up who would have been diagnosed as diabetic by oral glucose tolerance testing. For the same reason we probably also misclassified some individuals as glucose tolerant. Because of the second limitation we may have made a premature diagnosis of diabetes mellitus in some individuals. These limitations obviously weaken the strength of our data. However, it is also important to note that in previous studies, the relative risk associated with glucose intolerance, as defined by elevated fasting or postprandial glucose levels, ranged between 6 15 and was hence not much higher than that estimated by MLF or fasting glucose in our study (9, 17, 18, 48). Moreover, it is important to note that oral glucose tolerance tests are impractical in many clinical and outpatient situations, especially as screening tests in the general population. Actually, this problem has been a major motivation for the ADA not to demand glucose tolerance testing (30). The multiparametric screening of anthropometric and metabolic parameters needed for risk estimation by MLF analysis may turn out to be a more practical surrogate test instead of glucose tolerance testing or at least may serve as a kind of quality control for risk assessment by fasting glucose. Moreover, it is important to emphasize that our findings have been made in middle-aged Caucasian men and may not extend to women and individuals of different age and ethnicity. In conclusion, fasting glucose is a powerful measure to identify middle-aged Caucasian men at increased risk of diabetes mellitus. The predictive value would be even improved by the definition of a lower cut-off of 5.72 mmol/l. Finally, age, BMI, elevated blood pressure, low HDL cholesterol, as well as the occurrence of diabetes mellitus in the family history are independent factors that characterize a prediabetic state. The parallel assessment of these independent and easy to measure risk factors and the use of a statistical algorithm help to identify individuals who are at high risk for diabetes mellitus. References 1. Leslie RDG (ed) Causes of diabetes. Genetic and environmental factors. Wiley & Sons, Chicester. 2. Keen H, Jarrett RJ, McCartey P The ten year follow-up of the Bedford survey ( ). Glucose tolerance and diabetes. Diabetologia. 22: Haffner SM, Stern MP, Mitchell BD, Hazuda, HP, Patterson JK Incidence of type 2 diabetes in Mexican American predicted by fasting insulin and glucose levels, obesity and body fat distribution. Diabetes. 39: Charles MA, Fontbonne A, Thibault N, Warnet J-M, Roseda GE, Eschwege E Risk factors for NIDDM in white population. Paris Prospective Study. Diabetes. 40: Jarrett RJ, Keen H, Fuller JH, McCartney M Worsening of diabetes in men with impaired glucose tolerance ( borderline diabetes ). Diabetologia 16: King H, Zimmet P, Paper LR, Balkau LB The natural history of impaired glucose tolerance in the Micronesian population of Nauru. A six year follow-up study. Diabetologia 26: Saad ME, Knowler WC, Pettitt DJ, Nelson RG, Mott DM, Bennett PH The natural history of impaired glucose tolerance in Pima Indians. N Engl J Med. 319: S Icree RA, Zimmet PZ, King HOM, Coventry JS Plasma insulin response among Nauruans. Prediction of deterioriation in glucose tolerance over 6 years. Diabetes 36: Lillioja S, Mott DM, Spraul M, et al Insulin resistance and insulin secretory dysfunction as precursors of non-insulin dependent diabetes. Prospective studies in Pima Indians. N Engl J Med. 329: Warram JH, Martin BC, Krolewski AS, Soledner JS, Kahn R Slow glucose removal rate and hyperinsulinemia preceed the development of type 2 diabetes in the offspring of diabetic parents, Ann Intern Med. 113: Ohlson LO, Larsson B, Svärdssudd K, et al The influence of body fat distribution on the incidence of diabetes mellitus years of follow-up of the participants in the study of men born in Diabetes 34: Carey VJ, Walters EE, Colditz GA, et al Body fat distribution and risk of non-insulin dependent diabetes mellitus. The Nurses Health Study. Am J Epidemiol. 145:

8 3108 VON ECKARDSTEIN, SCHULTE, AND ASSMANN JCE&M 2000 Vol. 85 No Colditz GA, Willet WC, Rotnitzky A, Mansson JAE Weight gain as a risk factor for clinical diabetes mellitus. Ann Intern Med. 122: Jarrett RJ, Shipley MJ Type 2 (non-insulin-dependent) diabetes mellitus and cardiovascular disease: putative associations via common antecedents. Further evidence form the Whitehall study. Diabetologia. 31: Haffner SM, Stern PM, Hazuda HP, Mitchel BD, Patterson JK Cardiovascular Risk Factors in confirmed prediabetic individuals. Does the clock for coornary heart disease start ticking before the onset of clinical diabetes. JAMA. 263: Ohlson LO, Larsson B, Björntorp P, et al Risk factors for type 2 (noninsulin-dependent) diabetes mellitus. Thirteen and one-half years of follow-up of the participants in a study of Swedish men born in Diabetologia. 31: McPhillips JB, Barrett-Connor E, Wingard DL Cardiovascular disease risk factors prior to the diagnosis of impaired glucose tolerance and noninsulin dependent diabetes mellitus. Am J Epidemiol. 131: Mykkänen L, Kuusisto J, Pyörälä K, Laakso M Cardiovascular disease risk factors as predictors of type 2 (non-insulin-dependent) diabetes mellitus in elderly subjects. Diabetologia. 36: Austin MA, Mykkänen L, Kuusisto J, et al Prospective study of small LDLs as a risk factor for non-insulin-dependent diabetes mellitus in elderly men and women. Circulation. 92: Perry IJ, Wanamathee SG, Walker MK, Thomson AG, Whincup PH, Shaper AG Prospective study of risk factors for development of non-insulin dependent diabetes mellitus in middle aged men. Br Med J. 310: Wilson PWF, Anderson KM, Kannel WB Epidemiology of diabetes mellitus in the elderly. The Framingham Study. Am J Med. 80(Suppl A): Skarfors ET, Sellinus KI, Lithell HO Risk factors for developping non-insulin-dependent diabetes mellitus. A 10 year follow-up of men in Uppsala. Br Med J. 303: Feskens EJM, Kromhout D Cardiovascular risk factors and the 25 years incidence of diabetes mellitus in middle aged men. The Zutphen Study. Am J Epidemiol. 130: Njolstad I, Arnesen E, Lund-Larsen PG Sex-differences in risk factors for clinical diabetes mellitus in a general population. A 12 year follow-up of the Finnmark study. Am J Epidemiol. 147: Stern MP Diabetes and cardiovascular disease. The common soil hypothesis. Diabetes. 44: Meigs JB, D Agostino Sr RB, Wilson PW, Cupples LA, Nathan DM, Singer DE Risk variable clustering in the insulin resistance syndrome. The Framingham Offspring Study. Diabetes. 46: DeFronzo RA Pathogenesis of type 2 diabetes: metabolic and molecular implications for identifying diabetes genes. Diabetes Rev. 5: Assmann G, Carmena R, Cullen P, et al Coronary heart disease. Reducing the risk. A worldwide view. Circulation. 100: Grundy S, Pasternak M, Greenland R, Smith S, Fuster V Assessment of cardiovascular risk by the use of multiple risk factor assessment equations. Circulation. 100: Expert Committee on the Diagnosis and Classification of Diabetes Mellitus Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care. 20: Alberti KGMM, Zimmer PZ. Definition, diagnosis, and classification of diabetes mellitus and its complications. I. Diagnosis and classification of diabetes mellitus. Provisional report of a WHO consultation. Diabetes Med. 15: Peters AL Diagnosing diabetes in Curr Opin Endocrinol Diabetes. 7: WHO Diabetes mellitus report of a WHO study group. Geneva: WHO; Tech Rep Ser de Vegt F, Dekker JM, Sehouwer CDA, Nijpels G, Bouter LM, Heine RJ The 1997 American Diabetes Association criteria vs. the 1985 World Health Organization criteria for the diagnosis of abnormal glucose tolerance. Poor agreement in the Hoorn study. Diabetes Care. 21: Gimeno SGA, Ferreira SRG, Franco LJ, Junes M, Japanese-Brazilian Diabetes Study Group Comparison of glucose tolerance categories according to World Health Organization and American Diabetes Association diagnsotic criteria in a population-based study in Brazil. Diabetes Care. 21: Ramachandran A, Snehalata C, Latha E, Vijay V Evaluation of the use of fasting glucose as a new diagnositc criterion for diabetes in Asian Indian populations. Diabetes Care. 21: Harris MI, Eastman RC, Cowie CC, Flegal KM, Eberhardt MS Comparison of diabetes diagnostic categories in the US population according to 1997 American Diabetes Association and World Health Organization diagnostic criteria. Diabetes Care. 20: Wahl PW, Savage PJ, Psaty BM, Orchard TJ, Robbins JA, Russell PT Diabetes in older patients. Comparison of 1997 American Diabetes Association classification of diabetes mellitus with 1985 WHO classification. Lancet. 352: DECODE Study Group Glucose tolerance and mortality. Comparison of WHO and American Diabetes Association diagnostic criteria. Lancet. 354: Barzilay JI, Spiekerman CF, Wahl PW, et al Cardiovascular disease in older adults with glucose disorders: comparison of American Diabetes Association criteria for diabetes mellitus with WHO criteria. Lancet. 354: Davies M New diagnostic criteria for diabetes: are they doing what they should? Lancet. 354: Assmann G, Schulte H Results and conclusions of the Prospective Cardiovascular Münster (PROCAM) Study. In: Assmann G, ed. Lipid metabolism disorders and coronary heart disease. Primary prevention, diagnosis, and therapy guidelines for general practice, 2nd Ed. Munich: Medizin Verlag; Anonymous Standardization of blood pressure readings. Am Heart J. 8: Friedewald WT, Levy J, Fredrickson DS Estimation of the concentration of low-density-lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem. 18: Organisation Mondiale de la Santé L hypertension artérielle. Ser Rap Tech. 00: Nie NH SPSSX user s guide. New York: McGraw-Hill. 47. Ray AA SAS user s guide: basics. Cary: SAS Institute. 48. Haffner SM, Miettinen E, Stern MP Are risk factors for conversion to NIDDM similar in high and low risk populations? Diabetologia. 40:62 66.