A combination of HbA1c, fasting glucose and BMI is effective in screening for individuals at risk of future type 2 diabetes: OGTT is not needed

Transcription

1 Journal of Internal Medicine 2006; 260: doi: /j x A combination of HbA1c, fasting glucose and BMI is effective in screening for individuals at risk of future type 2 diabetes: OGTT is not needed M. NORBERG 1,J.W.ERIKSSON 2,B.LINDAHL 3, C. ANDERSSON 4,O.ROLANDSSON 4,5, H. STENLUND 1 & L. WEINEHALL 1,6 From the Sections of 1 Epidemiology and Public Health Sciences, 2 Medicine, 3 Behavioral Medicine, 4 Family Medicine and 5 Nutritional Research, Department of Public Health and Clinical Medicine, Umeå University, Umeå; and 6 Research Department, National Institute of Public Health, Stockholm, Sweden Abstract. Norberg M, Eriksson JW, Lindahl B, Andersson C, Rolandsson O, Stenlund H, Weinehall L (Umeå University, Umeå; and National Institute of Public Health, Stockholm, Sweden). A combination of HbA1c, fasting glucose and BMI is effective in screening for individuals at risk of future type 2 diabetes: OGTT is not needed. J Intern Med 2006; 260: Objective. To identify a screening model that predicts high risk of future type 2 diabetes and is useful in clinical practice. Design and methods. Incident case-referent study nested within a population-based health survey. We compared screening models with three risk criteria and calculated sensitivity, specificity, positive (PPV) and negative (NPV) predictive values and attributable proportion. We used fasting plasma glucose (FPG) alone or with an oral glucose tolerance test (OGTT), glycosylated haemoglobin A (HbA1c) (normal range %), body mass index (BMI), triglycerides and family history of diabetes (FHD). Setting. Participants in a health survey at all primary care centres (n ¼ ) and subjects with diagnosed type 2 diabetes in primary and hospital care (n ¼ 6088) in Umeå during Subjects. Each of the 164 subjects who developed clinically diagnosed type 2 diabetes (median time to diagnosis of 5.4 years) and 304 sex- and agematched referents without diabetes diagnosis. Results. Screening models with at least one criterion present had sensitivities of , specificities of and PPVs of 8 9%. Combinations of the criteria, FPG 6.1 mmol L )1 (capillary plasma), HbA1c 4.7% and BMI 27 in men and BMI 30 in women, had sensitivities, specificities and PPVs of 0.66%, 0.93% and 32%, and 0.52%, 0.97% and 46% respectively. Using FHD as one of three risk criteria showed comparable results. Addition of triglycerides or OGTT did not improve the prediction. Conclusions. The combination of HbA1c, FPG and BMI are effective in screening for individuals at risk of future clinical diagnosis of type 2 diabetes. OGTT or FHD is not necessary. Keywords: body mass index, glucose metabolism, risk factors, type 2 diabetes mellitus. Introduction There is an increasing incentive, both from a public health and a clinical perspective, to detect people at risk of future type 2 diabetes as it is a strong risk factor of cardiovascular disease (CVD) [1], and also as several trials have reported success in preventing or delaying diabetes in people with impaired glucose tolerance (IGT) [2, 3]. Routine use of oral glucose tolerance test (OGTT) is seldom used in clinical practice as it is time consuming and expensive, and therefore simpler methods are needed [4]. Several screening models have been previously reported. Investigators from the San Antonio Heart Study showed that a regression equation with eight demographic and clinical variables performed better Ó 2006 Blackwell Publishing Ltd 263

2 264 M. NORBERG et al. than OGTT [5] and also that the metabolic syndrome predicts diabetes [6]. The Finish Diabetes Risk Score was developed with drug-treated diabetes as outcome and uses anthropometry, lifestyle variables and history of hypertension and hyperglycaemia and requires no blood testing [7]. A Chinese study suggested the pairing values of fasting blood glucose (FPG) and glycosylated haemoglobin A (HbA1c) for prediction of future type 2 diabetes [8] and recently a report from the Botnia Study suggested measurements of FPG and body mass index (BMI) and in addition family history of diabetes (FHD) [9]. Body mass index alone has been shown to be a prospective marker for development of type 2 diabetes [10, 11], but another study showed that obesity and/or heredity background of diabetes will fail to detect the majority of subjects with IGT [12]. Impaired fasting glucose (IFG) and IGT are major risk factors for type 2 diabetes, particularly when they coexist, whereas IGT is more strongly associated with CVD [1]. In 2003 the American Diabetes Association Expert Committee on Diagnosis and Classification of Diabetes Mellitus changed the definition of IFG. The lower limit was shifted from a fasting plasma glucose (FPG) 6.1 mmol L )1 (110 mg dl )1 ) to 5.6 mmol L )1 (100 mg dl )1 ). This decision has been questioned as it increases the number of individuals diagnosed with IFG without a clear impact on their health [13]. Glycosylated haemoglobin A is clinically widely used in diabetes care as a reliable marker of longterm glycaemic control, but at present is not recommended as a diagnostic tool [14]. However, some studies report that HbA1c alone [15] or in combination with FPG [16] is a reliable marker for prevalent diabetes. Moreover, HbA1c is an independent progressive risk factor for CVD regardless of diabetes status [17] and the Framingham Heart Study suggests that HbA1c adequately reflects glucose levels in subjects without diabetes [18]. In spite of this, HbA1c is seldom used in screening for type 2 diabetes. We aimed at finding a simple and clinically useful method to identify individuals at high risk of future type 2 diabetes without using OGTT. A second aim was to evaluate whether FPG 5.6, the new IFG criterion, is as effective as FPG 6.1, when applied in screening models of future type 2 diabetes. To achieve this we conducted an incident case-referent study nested within an ongoing population-based health survey. Materials and methods Setting The Västerbotten Intervention Programme (VIP), a community programme for prevention of CVD and diabetes, started in 1985 in the county of Västerbotten in northern Sweden [19]. At the ages of 40, 50 and 60 years, all inhabitants are invited to their primary care centre for a health survey. Participants answer a questionnaire on psychosocial conditions and lifestyle, biomedical measurements are performed and they are asked to donate blood for future research. At the same occasion information about the physical examination and blood testing results and individual counselling concerning lifestyle is given but no active interventions are initiated. When the programme was launched the efforts were targeted at reducing hypertension, hypercholesterolaemia and smoking, as these were established risk factors for CVD. Evaluations in 1999 showed that no active intervention or specific follow-up were targeted at prediabetes (data not published), therefore such practices were implemented in Study population The study is an incident case-referent study nested within VIP [20] (Fig. 1). We used two data sets from the population in Umeå in Northern Sweden ( inhabitants). The first data set included all VIP survey participants from 1989 to 2000, a total of , representing 52% (amongst men 49%, and amongst female 56%) of the eligible population. No consistent differences in social conditions have been found between participants and nonparticipants [19]. Subjects with prevalent diabetes at the health survey (n ¼ 1038) or with incomplete OGTT (n ¼ 3562) were excluded. The second data set included all individuals in the area with prevalent diabetes during the study period until January (n ¼ 6739, prevalence 5.2%). They were identified from the registers of diagnoses from the Departments of Internal Medicine and Cardiology at the only local hospital, i.e. Umeå University Hospital, and from all primary

3 SCREENING FOR TYPE 2 DIABETES 265 Oral glucose tolerance test missing (3562) Health Survey (33 336) Prevalent & incident diabetes at health survey (1038) Individuals without diabetes at health survey (28736) Record linkage known incident cases by end identifying type 2 of follow up clinically diabetes (6088) Cases Evaluation of case records * Not type 2 Diabetes = 40 excluded = 81 Data from the health survey Cases Exclusion Cases excluded = 73 excluded = 169 Data from frozen samples of blood Cases Fig. 1 Study design of incident case control study nested in the Västerbotten Intervention Programme. The outcome was type 2 diabetes. *Verification of diagnosis. Forty cases were excluded as they did not meet WHO type 2 diabetes criteria; 80 case-matched referents were consequently excluded; one referent was excluded because of a technical error. Exclusion of cases and referents as blood samples were not available: 34 cases had not donated any blood. Eleven cases with impaired glucose tolerance were excluded as they participated in an intervention study; one case was excluded because both referents were excluded; 17 cases were given priority to DNA extraction and 10 cases to other studies; 143 referents were excluded because corresponding matched cases were excluded; 16 referents were given priority to other studies; seven referents had not donated blood and for three the glycosylated haemoglobin A analysis failed. care centres. This ascertainment method should cover >95% of adult patients with diabetes in the area and this is also supported by the obtained diabetes prevalence reported from the WHO Northern Sweden MONICA project [21]. After exclusion of 651 subjects with type 1 diabetes, 6088 subjects with type 2 diabetes remained (prevalence 4.7%). Cases and referents in the study were then identified through register linkage of the two data sets. Cases were free from diabetes at the time of the health survey but were diagnosed with type 2 diabetes (n ¼ 277) attending routine care during the study period. For each case, two referents, that were not diagnosed with diabetes until 31 January 2001 were randomly assigned from the original health survey cohort. Matching was made based on sex, age and year of survey. Mean duration from health survey until 31 January 2001 for both cases and referents was 8.8 years (range ) and from the health survey until diabetes diagnosis amongst cases 5.4 years (range ). Case records were evaluated for verification of correct clinical type 2 diabetes diagnosis, according to 1998 WHO definitions [22]. Two hundred and thirty-seven cases with measurements and questionnaire data from the original VIP health survey were included in the study. After exclusion because of inadequate sample volumes due to priority to DNA extraction or to other studies or participation in an on-going diabetes prevention study, frozen samples of erythrocytes and plasma from the health survey were retrieved and analysed after end of follow-up in 164 cases and 304 referents. All calculations and interpretations were based on these subjects. The protocol was approved by the Research Ethics Committee of Umeå University and all participants gave informed consent. Measurements Weight and height were measured, and BMI (kg m )2 ) was calculated. Central obesity was not

4 266 M. NORBERG et al. assumed during the time period of this particular study, but since 2003 waist circumference has been measured in VIP. Blood pressure was measured with a mercury sphygmomanometer with subjects in a supine position after 5 min of rest. An OGTT was performed with a 75-g glucose load [22]. Glucose concentrations were measured on capillary plasma (PG, mmol L )1 ) on a Reflotron bench-top analyser (Boehringer Mannheim GmbH, Mannheim, Germany) in a fasting state (FPG) and 2 h after glucose administration (2 hpg). Glucose concentrations were classified as normal fasting glucose, (NFG, FPG <5.6), IFG [FPG_5.6, (FPG ) alternatively FFG_6.1 (FPG )], normal glucose tolerance (NGT, FPG <7.0 and 2 hpg <8.9) or IGT (FPG <7.0 and 2 hpg ). Subjects with diabetic glucose levels (FPG 7.0 or 2 hpg 12.2) were excluded. If approved by the subject, a venous sample of blood was drawn and stored at )80 C. Plasma lipids and HbA1c [HPLC method, normal range %, calibrated according to the Swedish MonoS standard, 1% below Diabetes Control and Complications Trial (DCCT)] were analysed on stored samples using routine methods at the Department of Clinical Chemistry at Umeå University Hospital. The storage time ranged from 2 to 12 years (mean 8.4 ± 2.5 years). HbA1c is not affected by freezer storage time [23]. Statistical analysis Characteristics of the subjects are presented as mean ± standard deviation. Statistical significance was tested for continuous and categorical variables by t-test and chi-squared test respectively (Table 1). The ability to predict incident type 2 diabetes was evaluated by univariate and multivariate conditional logistic regression analyses to estimate odds ratios (OR) with 95% confidence intervals. Based on this regressions, we evaluated screening models with three dichotomous risk criteria, i.e. there are eight possible ways to combine the criteria (ABC, ABC, ABC, ABC, ABC, ABC, ABC, ABC where A, B and C denote absence and A, B and C denote presence of the criterion; see Appendix S1). The plasma glucose criteria were either IFG or IGT (with or without concomitant IFG), and both were tested with either FPG_5.6 ( mmol L )1 ) or FPG_6.1 ( mmol L )1 ). Triglycerides were dichotomized by 1.7 mmol L )1. The obesity criterion was tested as BMI 27 or, alternatively, BMI 30. The cut-off point of BMI 27 was selected according to the Table 1 Characteristics at baseline for cases and referents in a nested case-referent study for risk of development of type 2 diabetes Men Women Cases (n ¼ 97) (n ¼ 182) P Cases (n ¼ 67) (n ¼ 122) P Age, years 51.0 ± ± ± ± BMI, kg m ) ± ± 3.6 < ± ± 4.3 <0.001 Systolic blood pressure, mmhg 139 ± ± 16 < ± ± Diastolic blood pressure, mmhg 89 ± ± 10 < ± 9 80 ± Hypertension a, % < Fasting plasma glucose b, mmol L )1 6.0 ± ± 0.6 < ± ± 0.8 < h plasma glucose b, mmol L )1 7.9 ± ± 1.5 < ± ± 1.5 <0.001 FFG_6.1 b, % < <0.001 FFG_5.6 b, % < <0.001 IGT b, % < <0.001 HbA1c c, % 4.7 ± ± 0.3 < ± ± 0.3 <0.001 Total cholesterol c, mmol L )1 6.1 ± ± ± ± HDL cholesterol c, mmol L )1 1.0 ± ± 0.3 < ± ± 0.3 <0.001 LDL cholesterol c, mmol L )1 4.0 ± ± ± ± Triglycerides c, mmol L )1 2.3 ± ± 0.6 < ± ± 0.7 <0.001 Heredity d, % Non/ex/daily smoker, % 35/36/29 53/31/ /22/23 60/20/ Values are mean ± SD or when indicated %. a Blood pressure 140/90 and/or on antihypertensive medication. b Glucose determined on capillary plasma. IFG, impaired fasting glucose. FFG_6.1: fasting plasma glucose FFG_5.6: fasting plasma glucose IGT, impaired glucose tolerance: fasting plasma glucose <7.0 and 2-h plasma glucose c Determined on samples stored at )80 C. HbA1c normal range is using the Swedish calibration standard Mono-S from EQUALIS (External Quality Assurance in Laboratory Medicine in Sweden). d History of diabetes in first-degree relatives.

5 SCREENING FOR TYPE 2 DIABETES 267 classification by Bray [24] and also represents the cut-off point with optimal sensitivity (0.71) and specificity (0.71) by receiver operating characteristic analysis (data not shown). The BMI 30 cut-off for obesity and the triglycerides cut-off of 1.7 are used in several working definitions of the metabolic syndrome [25]. The HbA1c criterion was either 4.5% or 4.7% (normal range %). The selected cut-off points were the 75th and 90th percentiles in the distribution of the study referents (5.5% and 5.7% with DCCT calibration). Hypertension was defined as systolic blood pressure 140 mmhg and/or diastolic 90 mmhg or ongoing antihypertensive medication at the time of the health survey. FHD was only addressed with respect to first-degree relatives and this was based on one question in the health survey questionnaire: Diabetes in anyone of parents or siblings (yes or no). The risk analyses were based on crude OR for later development of type 2 diabetes and Rothmans definition of interaction and the relative excess risk caused by interaction [26]. Interaction implies that the effect from the exposure to several risk factors leads to more, or, alternatively, less cases than could be predicted by the additive effect from the same risk factors. This is appropriately applied to case-referent studies [27]. An algorithm was developed and has been previously described in detail [28]. We calculated ORs for future type 2 diabetes for each of the three single criteria, and ORs for interaction and for additive effects for each of the combinations of three risk criteria in each screening model. The attributable proportions due to each single risk criterion and to each possible combination of the three criteria in each model were also calculated [29]. This analysis was not matched. Finally we compared two different approaches to define risk of future diabetes: (i) presence of at least one of the three criteria or (ii) combinations of the three criteria. Sensitivities and specificities were calculated. Positive (PPV) and negative predictive values (NPV) were calculated based on an estimated prevalence of type 2 diabetes of 4.7%. Statistical analyses were performed with SPSS 11.5 (SPSS Inc., Chicago, IL, USA) and Stata 8.2 (StataCorp, College Station, TX, USA). Results Baseline characteristics are summarized in Table 1. Regression analyses showed that all risk criteria were significant predictors of type 2 diabetes in the bivariate analyses (data not shown). In the multivariate analyses, significant ORs were retained only for HbA1c 4.7%, BMI 30 and IFG (WHO definition) for all participants. In women IGT also was predictive (Table 2). Screening models using the three criteria of (i) IFG or IGT, (ii) BMI 27 and (iii) HbA1c 4.7% and defining risk as fulfilling any combination with at least two of the three criteria showed a proportion of attributable cases of 48 54%. If risk instead was defined as having at least one criterion, the proportion was 82 92%. In contrast, the PPVs were 17 27% and 8 9%, respectively, and the NPVs were % in all models. Because of low PPVs for screening models that applied one risk criterion, they were not further evaluated. For screening models that applied combinations of criteria, stratifying for sex improved the results for men with proportions of attributable cases reaching 54 57%. The screening model with FPG_6.1 (without using 2 hpg) had the highest PPV of 32% (Table 3). In women, the resultant sensitivities were %, specificities %, attributable proportions 42 52%, PPVs 12 21% and NPVs 97 98% (data not shown). Changing the obesity criterion to BMI 30 improved the result in women, the PPVs were 20 46%, and the screening model with FPG_6.1 displayed the highest PPV (Table 3). The overlap between the three-factor screening models using only FPG and models using also 2 hpg from the OGTT, otherwise identical, was considerable. For combinations of BMI 30, HbA1c 4.7 and either FPG_6.1 or FPG_5.6, the result from adding the 2HPG was only two and four more subjects respectively. Introducing FHD into three-factor screening models, including BMI 27 and one of the four glucose criteria, resulted in low PPVs of 15 18%. Changing the obesity criterion to BMI 30, the PPVs increased. The screening model with FPG_6.1 (no OGTT), BMI 30 and FHD had a PPV of 30, but with only 16% of cases attributable to these combinations. These models could not be stratified by sex, as referents were missing for at least one of the four possible combinations in both sexes. Other three-factor screening models using plasmaglucose 6.1, either HbA1c 4.5 or 4.7, FHD, BMI 30 and triglycerides 1.7 were tested. The best performing models were FHD and BMI 30 combined with

6 268 M. NORBERG et al. Table 2 Multivariate conditional logistic regression analyses in men and women. The outcome is clinical diagnosis of type 2 diabetes after a mean time of 5.4 ± 8.4 years. The case and referent (Ref) subjects were matched for sex, age and year of health survey Men Women No. observations Multivariate model No. observations Multivariate model Risk factor Cases Ref OR 95% CI Cases Ref OR 95% CI HbA1c, % a < Body mass index, kg m )2 < Impaired fasting glucose b No Yes Impaired glucose tolerance b No Yes Hypertension c No Yes Triglycerides, mmol L )1 < Family history of diabetes No Yes a Swedish Mono-S calibration, normal range %, equivalent to % DCCT calibrated. b WHO definitions: impaired fasting glucose is fasting plasma glucose of mmol L )1 and impaired glucose tolerance is fasting plasma glucose of <7.0 mmol L )1 and 2-h plasma glucose of mmol L )1 measured on capillary plasma. c WHO definition: systolic blood pressure 140 mmhg and/or diastolic blood pressure 80 mmhg and/or on antihypertensive medication. Screening model Men Sensitivity Specificity Attributable proportion, % PPV, % NPV, % FPG_5.6, BMI 27, HbA1c OGTT_5.6, BMI 27, HbA1c FPG_6.1, BMI 27, HbA1c OGTT_6.1, BMI 27, HbA1c Women Table 3 Sensitivity, specificity, attributable proportion, positive (PPV) and negative predictive values (NPV) for screening models where risk for developing type 2 diabetes is defined by fulfilling at least two of the three criteria FPG_5.6, BMI 30, HbA1c OGTT_5.6, BMI 30, HbA1c FPG_6.1, BMI 30, HbA1c OGTT_6.1, BMI 30, HbA1c Plasma glucose measured on capillary plasma. FPG, fasting plasma glucose; OGTT, oral glucose tolerance test. FPG_5.6: FPG and no 2 hpg. OGTT_5.6: FPG and/or 2 hpg FPG_6.1: FPG and no 2 hpg. OGTT_6.1: FPG and/or 2 hpg either HbA1c 4.5 or HbA1c 4.7 and they had PPVs of 26% and 35% respectively. The attributable proportions were 35% and 32%, respectively, and lower than models presented in Table 3. Discussion We have shown that screening models using combinations of dichotomous values of HbA1c, BMI and

7 SCREENING FOR TYPE 2 DIABETES 269 FPG accurately identify individuals at risk of future clinically diagnosed type 2 diabetes. The models should be stratified for sex with respect to use of a BMI cut-off of 27 in men and 30 in women. For use in this model, the glucose criterion should be mmol L )1, as the new 5.6 IFG limit proposed by the American Diabetes Association (ADA) decreased the PPV without an obvious increase of the proportion of subjects at risk. There is little additional value in adding OGTT versus the use of FPG alone for the prediction of type 2 diabetes in clinical practice. Finally, a highly specific cut-off for HbA1c, should be used, in this population 4.7%, corresponding to 5.7% with DCCT calibration. Our data do not in any respect indicate that OGTT could be abolished as a diagnostic test for prevalent diabetes. Family history of diabetes is a major risk factor for type 2 diabetes and therefore it could seem surprising, that family history in first-degree relatives (FHD) did not improve the screening models. This might be the result of underreporting that could lead in turn to negative confounding. Parents of subjects aged 40 may be at an age where they have not yet been diagnosed with diabetes in spite of a genetic predisposition. However, including second-degree relatives with diabetes, according to the Botnia study, does not markedly change the risk for type 2 diabetes [8]. We suggest that our screening models, with the use of FPG, BMI and HbA1c, identifies the majority of subjects with a high genetic risk of type 2 diabetes as this can be manifested as obesity and dysglycaemia. Several screening models that included FHD could not be analysed by sex, because there were too few referents with combinations of all three criteria. We think this illustrates that subjects with FHD and simultaneous other risk factors are in fact very close to meeting a diabetes diagnosis. Interestingly, in the Botnia study [8], where the subjects were selected on the basis of FHD, the PPVs for the combinations of FHD, BMI 30 and either FPG 5.6 or FPG 6.1 were 21% and 29%, respectively, and this is similar to our results of the same models, 19% and 30% respectively. The models we propose, with the use of FPG, BMI and HbA1c, seem to be more robust as the PPVs are higher and as information of FHD is not necessary. A combination of several tests makes it possible to benefit from both a high sensitivity in one test and a high specificity in another. In this analysis, we tried to optimize the PPV. Our result agrees with a previous study based on NHANES III, that suggests the combined use of FPG, HbA1c and BMI in order to reduce the number of OGTTs needed, for identification of individuals eligible for type 2 diabetes prevention programmes [30]. However, further studies using other populations are needed to confirm the validity of the algorithm before general clinical use can be recommended. Glycosylated haemoglobin A is significantly correlated with 2 hpg at OGTT but not with FPG in subjects without diabetes [31]. Postprandial glucose contributes relatively more than fasting glucose to HbA1c at the lowest levels of HbA1c [32] and this should be particularly relevant in individuals without diabetes. Afternoon and evening PG are also more highly correlated with HbA1c than morning PG [33]. These findings indicate that, in a prediabetic state, FPG and HbA1c in combination might better reflect the integrated glucose level, both basal and postprandial glucose levels, and this might also support replacing OGTT with FPG and HbA1c when screening individuals for risk of future diabetes. A limitation of our study might be that IGT is more predictive of CVD than IFG. Although the overlap between models shows that OGTT added only a few subjects at risk, further research is needed to evaluate the risk of CVD for subjects with combinations of IFG, BMI 27 or 30 and HbA1c 4.7. An objection could be that the progression to diabetes might be underestimated, as the outcome was based on clinical diagnosis and not on follow-up testing. On the other hand, regular follow-up testing of adults are not, to our knowledge, routine anywhere and our aim was to find a practical method for predicting risk of clinical diabetes diagnosis, and therefore our findings could be particularly relevant from a clinical point of view. Measurement of abdominal obesity might also have been informative, but this was not available in the VIP survey. It should also be borne in mind that the exact number of first-degree relatives afflicted with diabetes was not registered nor was the type of diabetes amongst first-degree relatives ascertained. Therefore, a small number of type 1 diabetes relatives might have biased the results to a minor extent. The participation rate might also be a limitation. The suboptimal participation rate is partly explained by discontinuation of VIP for periods of time in some primary care centres because

8 270 M. NORBERG et al. of lack of resources. This should not bias the selection of subjects undergoing health surveys [19]. The participation rate was relatively high (70%) during those periods when all planned health surveys were performed. In summary, these findings suggest that measurements of FPG, with the cut-off value of 6.1 mmol L )1, in combination with HbA1c 4.7% and BMI 27 in men and BMI 30 in women, offers a model that makes OGTT unnecessary for screening for individuals at risk of future type 2 diabetes development. As these measurements are readily available, this approach can be used in daily clinical practice with the aim of identifying individuals who might benefit from preventive efforts. Conflict of interest statement Margareta Norberg, Bernt Lindahl, Christer Andersson, Olov Rolandsson, Hans Stenlund and Lars Weinehall have no conflict of interest to declare. Jan W. Eriksson is adviser at AstraZeneca. Acknowledgements This study was supported by grants from the Västerbotten County Council, the Swedish Research Council (Medicine, grant to JWE) and from AstraZeneca (to OR). We thank Åsa Ågren, research assistant at the Northern Sweden Medical Research Bank, Umeå University, for collecting data. References 1 The DECODE Study Group, European Diabetes Epidemiology Group. Is the current definition for diabetes relevant to mortality risk from all causes and cardiovascular and noncardiovascular diseases? Diabetes Care 2003; 26: Tuomilehto J, Lindstrom J, Eriksson JG et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med 2001; 344: Knowler WC, Barrett-Connor E, Fowler SE et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N Engl J Med 2002; 346: Narayan KM. Targeting people with pre-diabetes. BMJ 2002; 325: Stern MP, Williams K, Haffner SM. Identification of persons at high risk for type 2 diabetes mellitus: do we need the oral glucose tolerance test? Ann Intern Med 2002; 136: Lorenzo C, Okoloise M, Williams K, Stern MO, Haffner SM. The metabolic syndrome as predictor of type 2 diabetes: the San Antonio heart study. Diabetes Care 2003; 26: Lindstrom J, Tuomilehto J. The Diabetes Risk Score: a practical tool to predict type 2 diabetes risk. Diabetes Care 2003; 26: Ko GT, Chan JC, Tsang LW, Cockram CS. Combined use of fasting plasma glucose and HbA1c predicts the progression to diabetes in Chinese subjects. Diabetes Care 2000; 23: Lyssenko V, Almgren P, Anevski D et al. Predictors of and longitudinal changes in insulin sensitivity and secretion preceding onset of type 2 diabetes. Diabetes 2005; 54: Rolandsson O, Hagg E, Nilsson M, Hallmans G, Mincheva- Nilsson L, Lernmark A. Prediction of diabetes with body mass index, oral glucose tolerance test and islet cell autoantibodies in a regional population. J Intern Med 2001; 249: Colditz GA, Willett WC, Rotnitzky A, Manson JE. Weight gain as a risk factor for clinical diabetes mellitus in women. Ann Intern Med 1995; 122: Lindahl B, Weinehall L, Asplund K, Hallmans G. Screening for impaired glucose tolerance. Results from a population- based study in individuals. Diabetes Care 1999; 22: Schriger DL, Lorber B. Lowering the cut point for impaired fasting glucose: where is the evidence? Where is the logic? Diabetes Care 2004; 27: American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care 2004; 27 (Suppl. 1): S Rohlfing CL, Little RR, Wiedmeyer HM et al. Use of GHb (HbA1c) in screening for undiagnosed diabetes in the U.S. population. Diabetes Care 2000; 23: Anand SS, Razak F, Vuksan V et al. Diagnostic strategies to detect glucose intolerance in a multiethnic population. Diabetes Care 2003; 26: Khaw KT, Wareham N, Bingham S, Luben R, Welch A, Day N. Association of hemoglobin A1c with cardiovascular disease and mortality in adults: the European prospective investigation into cancer in Norfolk. Ann Intern Med 2004; 141: Meigs JB, Nathan DM, Cupples LA, Wilson PW, Singer DE. Tracking of glycated hemoglobin in the original cohort of the Framingham Heart Study. J Clin Epidemiol 1996; 49: Weinehall L, Hallgren CG, Westman G, Janlert U, Wall S. Reduction of selection bias in primary prevention of cardiovascular disease through involvement of primary health care. Scand J Prim Health Care 1998; 16: Essebag V, Genest J Jr, Suissa S, Pilote L. The nested casecontrol study in cardiology. Am Heart J 2003; 146: Eliasson M, Lindahl B, Lundberg V, Stegmayr B. No increase in the prevalence of known diabetes between 1986 and 1999 in subjects years of age in northern Sweden. Diabet Med 2002; 19: WHO. Definition, diagnosis and classification of diabetes mellitus and its complications. Report of a WHO consultation Rolandsson O, Marklund SL, Norberg M, Agren A, Hagg E. Hemoglobin A1c can be analyzed in blood kept frozen at )80 degrees C and is not commonly affected by hemolysis in the general population. Metabolism 2004; 53: Bray G. An approach to the classification and evaluation of obesity. In: Björntorp P, Brodoff BN, eds. Obesity. Philadelphia, PA: J. B. Lippincott, 1992:

9 SCREENING FOR TYPE 2 DIABETES Grundy SM, Cleeman JI, Daniels SR et al. Diagnosis and management of the metabolic syndrome. An American Heart Association/National Heart, Lung, and Blood Institute scientific statement. Circulation 2005; 112: Rothman K. Modern Epidemiology. Boston, MA: Little Brown, Hallqvist J, Ahlbom A, Diderichsen F, Reuterwall C. How to evaluate interaction between causes: a review of practices in cardiovascular epidemiology. J Intern Med 1996; 239: Emmelin M, Weinehall L, Stegmayr B, Dahlgren L, Stenlund H, Wall S. Self-rated ill-health strengthens the effect of biomedical risk factors in predicting stroke, especially for men - an incident case referent study. J Hypertens 2003; 21: Hosmer DW, Lemeshow S. Confidence interval estimation of interaction. Epidemiology 1992; 3: Saydah SH, Byrd-Holt D, Harris MI. Projected impact of implementing the results of the diabetes prevention program in the U.S. population. Diabetes Care 2002; 25: Davies MJ, Raymond NT, Day JL, Hales CN, Burden AC. Impaired glucose tolerance and fasting hyperglycaemia have different characteristics. Diabet Med 2000; 17: Monnier L, Lapinski H, Colette C. Contributions of fasting and postprandial plasma glucose increments to the overall diurnal hyperglycemia of type 2 diabetic patients: variations with increasing levels of HbA(1c). Diabetes Care 2003; 26: Rohlfing CL, Wiedmeyer HM, Little RR, England JD, Tennill A, Goldstein DE. Defining the relationship between plasma glucose and HbA(1c): analysis of glucose profiles and HbA(1c) in the Diabetes Control and Complications Trial. Diabetes Care 2002; 25: Correspondence: Dr Margareta Norberg, Epidemiology By 9, Umeå University Hospital, SE Umeå, Sweden. (fax: ; margareta.norberg@epiph.umu.se). Supplementary Material The following supplementary material is available for this article online: Appendix S1 Method for calculation of the number of aetiological cases This material is available as part of the online article from