Normal Fasting Plasma Glucose and Risk of Type 2 Diabetes Diagnosis

CLINICAL RESEARCH STUDY Normal Fasting Plasma Glucose and Risk of Type 2 Diabetes Diagnosis Gregory A. Nichols, PhD, Teresa A. Hillier, MD, MS, Jonathan B. Brown, PhD, MPP Center for Health Research, Kaiser Permanente Northwest, Portland, Oregon. ABSTRACT PURPOSE: The study compares the risk of incident diabetes associated with fasting plasma glucose levels in the normal range, controlling for other risk factors. METHODS: We identified 46,578 members of Kaiser Permanente Northwest who had fasting plasma glucose levels less than 100 mg/dl between January 1, 1997, and December 31, 2000, and who did not previously have diabetes or impaired fasting glucose. After assigning subjects to 1 of 4 categories 85, ( 85-89, 90-94, or 95-99 mg/dl), we followed them until they developed diabetes, died, or left the health plan, or until April 30, 2007. We used Cox regression analysis to estimate the risk of incident diabetes, adjusted for age, sex, body mass index, blood pressure, lipids, smoking, cardiovascular disease, and hypertension. RESULTS: Subjects developed diabetes at a rate of less than 1% per year during a mean follow-up of 81.0 months. Each milligram per deciliter of fasting plasma glucose increased diabetes risk by 6% (hazard ratio [HR] 1.06, 95% confidence interval [CI], 1.05-1.07, P.0001) after controlling for other risk factors. Compared with those with fasting plasma glucose levels less than 85 mg/dl, subjects with glucose levels of 95 to 99 mg/dl were 2.33 times more likely to develop diabetes (HR 2.33; 95% CI, 1.95-2.79; P.0001). Subjects in the 90 to 94 mg/dl group were 49% more likely to progress to diabetes (HR 1.49; 95% CI, 1.23-1.79; P.0001). All other risk factors except sex were significantly associated with a diabetes diagnosis. CONCLUSIONS: The strong independent association between the level of normal fasting plasma glucose and the incidence of diabetes after controlling for other risk factors suggests that diabetes risk increases as fasting plasma glucose levels increase, even within the currently accepted normal range. 2008 Elsevier Inc. All rights reserved. The American Journal of Medicine (2008) 121, 519-524 KEYWORDS: Diabetes; Fasting glucose; Incidence; Prediction; Risk factors; Screening Diabetes mellitus, a chronic illness characterized by hyperglycemia, is typically diagnosed when fasting plasma glu- same risk as the higher criterion. However, a 3-fold induced level of fasting plasma glucose does not generate the 5 1 cose levels exceed 125 mg/dl. Fasting plasma glucose creased risk of diabetes at this lower level of impaired levels that are higher than normal but below this criterion fasting glucose (relative to those with normal plasma glucose) was reported as early as 1998. Indeed, the same study 2 increase the risk of developing diabetes. The American 6 Diabetes Association (ADA) terms this intermediate state demonstrated a clear risk gradient even among those with impaired fasting glucose, initially defining it as 110 to 125 normal ( 100 mg/dl) fasting plasma glucose but was un- mg/dl. 3 In 2003, the ADA lowered the threshold for impaired fasting glucose to 100 mg/dl to better predict diabetes development. 4 The reduction of the impaired fasting glucose cut-point generated international controversy, in part because the re- Requests for reprints should be addressed to Gregory A. Nichols, PhD, Center for Health Research, 3800 N. Interstate Avenue, Portland, OR 97227-1098. E-mail address: greg.nichols@kpchr.org able to account for other important diabetes risk factors. More recently, an Israeli study that adjusted for a number of other risk factors (family history of diabetes, smoking, hypertension, physical activity levels, triglycerides, and the ratio of total to high-density lipoprotein cholesterol) found that fasting plasma glucose levels significantly increased diabetes risk among young men (aged 26-45 years) with fasting plasma glucose levels less than 100 mg/dl. 7 We recently reported that fasting plasma glucose independently increased diabetes risk among a community-based sample 0002-9343/$ -see front matter 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.amjmed.2008.02.026

520 The American Journal of Medicine, Vol 121, No 6, June 2008 of men and women of all ages with impaired fasting glucose levels after controlling for other risk factors. 8 We report whether this association extends to patients in the same setting with normal plasma glucose levels. MATERIALS AND METHODS Kaiser Permanente Northwest (KPNW) is a group-model health maintenance organization that provides integrated health care to approximately 475,000 members in the Portland, Oregon area. KPNW maintains electronic databases containing information on all inpatient admissions, pharmacy dispenses, outpatient visits, and laboratory tests. All such information is stored in an electronic medical record in use since 1996. To complete each patient visit, the clinician is required to enter in the electronic medical record at least 1 diagnosis and may enter up to 20 diagnoses. The organization also provides online medical guidelines to assist clinicians in patient management. One such guideline recommends lipid screening for men aged more than 35 years and women aged more than 45 years. Fasting plasma glucose tests are routinely ordered with these lipid panels. For this study, we identified KPNW members aged more than 40 years who had a fasting plasma glucose test result of less than 100 mg/dl between January 1, 1997, and December 31, 2000, and who had health plan eligibility for at least 6 months before and 12 months after the test (n 53,356). For patients with multiple tests, we selected the first available result. We excluded 4856 potential subjects who had a previous test result of more than 100 mg/dl and 1922 potential subjects who had diagnosed diabetes. We followed the remaining 46,578 patients until they developed diabetes (inpatient or outpatient International Classification of Diseases, 9th Revision-Clinical Modification diagnosis 250.xx or fasting plasma glucose test result 125 mg/dl), died, or left the health plan, or until April 30, 2007. To determine whether the risk of diabetes increased with fasting plasma glucose, we analyzed fasting plasma glucose both as a continuous variable and in 4 categories: less than 85 mg/dl, 85 to 89 mg/dl, 90 to 94 mg/dl, and 95 to 99 mg/dl. To isolate the independent contribution of fasting plasma glucose to diabetes risk, we collected a number of control variables. From diagnoses contained in the electronic medical record, we identified comorbidities (International Classification of Diseases, 9th Revision-Clinical Modification codes) present at the time of the glucose test: myocardial infarction (410.xx); stroke (430.xx-432.xx, CLINICAL SIGNIFICANCE The risk of developing diabetes is associated with fasting plasma glucose levels that currently are considered normal. Obesity, hypertension, low high-density lipoprotein cholesterol, high triglycerides, and smoking increase the risk of developing diabetes associated with normal fasting plasma glucose. Closer surveillance for diabetes development within those with normal fasting plasma glucose levels might be warranted, especially if additional cardiovascular risk factors or established cardiovascular disease is present. 434.xx-436.xx, 437.1); other atherosclerotic cardiovascular disease (411.1, 411.8, 413.xx, 414.0, 414.8, 414.9, 429.2); and congestive heart failure (428.xx). These comorbidities were combined into a single variable indicating cardiovascular disease. We also identified hypertension diagnoses (401.xx-404.xx). Age was calculated as of the date of the fasting plasma glucose test. Smoking history, height, weight, and blood pressure were obtained from the electronic medical record. Lipid values were extracted from the laboratory database. For this study, we used the height, weight, blood pressure, and lipid values recorded on the date closest to the fasting plasma glucose test date either before or up to 3 months after the test date. By using these variables, we constructed a Cox regression model to estimate the adjusted risk of diabetes incidence. Interaction terms for fasting plasma glucose with age, sex, and body mass index (BMI) were found to be nonsignificant so were not included in the final model. All data were analyzed using SAS, version 8.2 (SAS Institute, Cary, NC). RESULTS Study subjects age averaged 57.5 years, and 40.4% were men (Table 1). Mean follow-up time was similar across fasting plasma glucose categories, averaging 81.0 months. Mean age, BMI, systolic blood pressure, low-density lipoprotein and total cholesterol, and triglycerides all increased with fasting plasma glucose, whereas high-density lipoprotein cholesterol decreased (P.001 for all comparisons). The proportion who were male, had cardiovascular disease, or had hypertension also increased with fasting plasma glucose. The characteristics of the 4.0% (n 1854) of subjects who developed diabetes are shown in Table 2. Diabetes developed after a mean of 54.6 months. The mean time to diabetes did not differ among categories, but the proportion developing diabetes increased significantly with fasting plasma glucose (P.001 for all comparisons, except between the 85 mg/dl and 85-89 mg/dl categories). All risk factors were less favorable among individuals who developed diabetes compared with the total sample. However, their risk profiles were similar regardless of initial fasting plasma glucose levels. Mean hemoglobin A 1c at diagnosis was greater than 7% in all categories. Kaplan-Meier plots of the unadjusted cumulative diabetes incidence for each of the 4 categories are displayed in Figure 1. Subjects in the 95 to 99 mg/dl category developed

Nichols et al Normal Fasting Glucose and Diabetes Risk 521 Table 1 Baseline Characteristics of Total Study Sample by Fasting Plasma Glucose Category 85 mg/dl 85-89 mg/dl 90-94 mg/dl 95-99 mg/dl Total n (%) 8705 (18.7%) 10,983 (23.6%) 13,704 (29.4%) 13,186 (28.3%) 46,578 Age* 55.4 (11.0) 56.6 (10.7) 57.8 (11.0) 59.1 (11.1) 57.5 (10.9) % Male* 28.7% 35.5% 42.9% 49.4% 40.4% BMI* 28.0 (5.8) 28.6 (5.8) 29.2 (5.9) 29.9 (6.0) 29.0 (5.9) Systolic blood pressure* 128 (19) 130 (18) 131 (19) 134 (19) 131 (19) Diastolic blood pressure* 79 (10) 79 (10) 80 (10) 81 (10) 80 (10) HDL cholesterol* 57 (17) 56 (17) 54 (16) 52 (16) 54 (16) LDL cholesterol* 123 (34) 126 (34) 129 (35) 130 (34) 128 (34) Triglycerides* 142 (122) 147 (115) 156 (106) 164 (114) 154 (114) Total cholesterol* 208 (42) 210 (39) 213 (41) 214 (40) 212 (41) Current smoker 20.3% 19.4% 20.4% 21.2% 20.4% Cardiovascular disease 4.8% 5.2% 6.6% 7.6% 6.2% Hypertension* 21.6% 24.0% 26.7% 31.7% 26.5% Mean months of follow-up (SD) 79.0 (28.4) 80.1 (29.8) 82.0 (30.1) 82.1 (31.1) 81.0 (30.0) BMI body mass index; HDL high-density lipoprotein; LDL low-density lipoprotein; SD standard deviation. *All comparisons of glucose categories are statistically significantly different (P.001). All comparisons of glucose categories are statistically significantly different (P.001), except 85 mg/dl versus 85-89 mg/dl (not significant). Data are means (standard deviation) or percent. diabetes at a rate of 9.9 per 1000 person-years (95% confidence interval [CI], 9.3-10.6). Subjects in the 90 to 94 mg/dl category developed diabetes at a rate of 5.6 per 1000 person-years (95% CI, 5.1-6.1). Subjects in the 85 to 89 mg/dl and less than 85 mg/dl groups developed diabetes at rates of 3.6 (95% CI, 3.2-4.1) and 3.1 (95% CI, 2.6-3.5), respectively. The log-rank test indicated statistically significant differences (P.0001) in the incidence of diabetes among the 4 categories. In a multivariable model that included all subjects, each milligram per deciliter of fasting plasma glucose was associated with an increased risk of diabetes incidence of 6% (hazard ratio [HR] 1.06; 95% CI, 1.05-1.07) (Table 3). Except for male sex, all other risk factors were highly statistically significant. Each kilogram per meter squared of BMI was associated with an increased risk of diabetes of 8% (HR 1.08; 95% CI, 1.07-1.09; P.0001), whereas the presence of hypertension was associated with an increased risk of 51% (HR 1.51; 95% CI, 1.35-1.68; P.0001). Compared with subjects with fasting plasma glucose levels less than 85 mg/dl, those in the 85 to 89 mg/dl category were not at significantly greater risk of diabetes Table 2 Baseline Characteristics of Subjects Who Developed Diabetes 85 mg/dl 85-89 mg/dl 90-94 mg/dl 95-99 mg/dl Total n (%) 175 (9.4%) 264 (14.2%) 520 (28.1%) 895 (48.3%) 1,854 Proportion of baseline FPG category* 2.0% 2.4% 3.8% 6.8% 4.0% Age 56.6 (9.7) 57.4 (9.5) 58.7 (10.4) 59.1 (10.5) 58.5 (10.3) % Male 42.9% 45.5% 46.5% 49.8% 47.6% BMI 32.7 (6.7) 33.5 (7.2) 33.4 (7.3) 33.0 (6.8) 33.2 (7.0) Systolic blood pressure 134 (18) 137 (18) 136 (20) 136 (19)% 136 (19) Diastolic blood pressure 82 (10) 82 (11) 82 (11) 82 (10) 82 (10) HDL cholesterol 49 (15) 46 (15) 47 (14) 47 (14) 47 (14) LDL cholesterol 123 (37) 124 (36) 127 (37) 125 (34) 125 (36) Triglycerides 239 (364) 213 (257) 212 (149) 209 (131) 213 (191) Total cholesterol 216 (54) 210 (45) 216 (48) 212 (43) 213 (46) Current smoker 27.4% 25.8% 25.0% 22.0% 23.9% Cardiovascular disease 9.7% 8.3% 11.9% 11.7% 11.1% Hypertension 41.1% 49.6% 43.7% 43.6% 44.2% Months to diabetes development 59.0 (28.3) 54.6 (28.3) 54.4 (28.3) 53.8 (29.9) 54.6 (29.2) Mean fasting glucose at diagnosis 150 (63) 145 (52) 141 (53) 142 (51) 143 (53) Mean HbA 1c at diagnosis 7.3% (1.5) 7.1% (1.5) 7.3% (1.8) 7.1% (1.5) 7.2% (1.6) FPG fasting plasma glucose; BMI body mass index; HDL high-density lipoprotein; LDL low-density lipoprotein; HbA 1c hemoglobin A1c. *All comparisons of glucose categories are statistically significantly different (P.001), except 85 mg/dl versus 85 to 89 mg/dl (not significant).

522 The American Journal of Medicine, Vol 121, No 6, June 2008 12 95-99 mg/dl 10 90-94 mg/dl 9.9/1,000 person-years 95% CI, 9.3-10.6 Cumulative Incidence (%) 8 6 4 < 85 mg/dl 3.1/1,000 person-years 95% CI, 2.6-3.5 85-89 mg/dl 3.6/1,000 person-years 95% CI, 3.2-4.1 5.6/1,000 person-years 95% CI, 5.1-6.1 2 log-rank p<0.0001 0 0 6 12 17 21 24 28 32 35 38 41 45 48 51 55 58 62 66 70 74 77 81 85 89 92 98 102 108 Months of Follow-up Figure 1 Kaplan-Meier plot of cumulative diabetes incidence by category of normal fasting plasma glucose. after adjustment for other risk factors (Figure 2). However, those in the 90 to 94 mg/dl group had a 49% greater diabetes risk relative to the subjects in the less than 85 mg/dl group (HR 1.49; 95% CI, 1.23-1.79), and diabetes risk was 2.33 (1.95-2.79) times greater in the 95 to 99 mg/dl group. Other risk factors performed identically to the model presented in Table 3. Table 3 Cox Regression Analysis of Diabetes Incidence HR 95% CI P value Fasting plasma glucose (per mg/dl) 1.06 1.05-1.07.0001 Age (per year) 1.01 1.00-1.02.001 Male sex 1.01 0.90-1.13.837 BMI (per kg/m 2 ) 1.08 1.07-1.09.0001 Systolic blood pressure (per 5 1.02 1.01-1.03.008 mm Hg) HDL cholesterol (per 5 mg/dl) 0.90 0.88-0.92.0001 LDL cholesterol (per 10 mg/dl) 0.97 0.96-0.99.0001 Triglycerides (per 50 mg/dl) 1.09 1.07-1.10.0001 Current smoker 1.37 1.22-1.54.0001 Diagnosed cardiovascular disease 1.65 1.40-1.93.0001 Diagnosed hypertension 1.51 1.35-1.68.0001 HR hazard ratio; CI confidence interval; BMI body mass index; HDL high-density lipoprotein; LDL low-density lipoprotein. CONCLUSIONS In an observational cohort analysis of 46,578 communitybased health maintenance organization subjects, we found a strong association between level of normal fasting plasma glucose and diabetes incidence after controlling for a large number of other known risk factors. The overall risk of diabetes (4.0%) among these subjects was relatively low compared with patients with impaired fasting glucose levels in the same study setting (11.3%). 9 However, we found that each milligram per deciliter of fasting plasma glucose was associated with a 6% increased risk of diabetes in the subjects with fasting plasma glucose levels in the normal range, similar to the 7% risk increase we reported in patients with impaired fasting glucose levels. Furthermore, the other risk factors analyzed in the current study performed nearly identically to the previous study. Thus, fasting plasma glucose levels seem to impart diabetes risk that begins well below the currently accepted normal level. In a study conducted on the island of Mauritius, Shaw et al 10 concluded that the risk of diabetes (when diagnosed by fasting plasma glucose alone) starts to increase at a fasting plasma glucose level of greater than 5.2 mmol/l ( 94 mg/dl). 10 Our results suggest that the threshold is lower. In our data, a fasting plasma glucose level of 90 to 94 mg/dl conferred a 49% greater risk of developing diabetes compared with a level less than 85 mg/dl. Furthermore,

Nichols et al Normal Fasting Glucose and Diabetes Risk 523 3.00 2.33* (1.95-2.79) 2.00 Hazard Ratio 1.00 1.00 (referent) 1.08 (0.87-1.33) 1.49* (1.23-1.79) 0.00 < 85 mg/dl 85-89 mg/dl 90-94 mg/dl Baseline Fasting Plasma Glucose Category *Statistically significantly different from <85mg/dL, p<0.0001. 95-99 mg/dl Figure 2 HRs (95% CIs) from Cox regression analysis of diabetes incidence, adjusted for risk factors displayed in Table 3. although the HR of 1.08 in the 85 to 89 mg/dl category did not reach statistical significance, it was nonetheless elevated, suggesting that the upper portion of this range may also carry additional risk. This is consistent with the study of young Israeli men, which found significantly greater risk at the level of 87 to 90 mg/dl (relative to 81 mg/dl). 11 Up to 70% of individuals with abnormal glucose regulation, defined as impaired fasting glucose or impaired glucose tolerance, may ultimately progress to diabetes. 2 Whether diabetes risk is best identified by fasting or postchallenge glucose tests remains controversial, but it is agreed that impaired fasting glucose and impaired glucose tolerance do not define the same individuals. 12 Because no subjects in the current study had impaired fasting glucose, it is likely that some of those who progressed to diabetes had impaired glucose tolerance. This would not be surprising given the higher sensitivity of impaired glucose tolerance in predicting progression to diabetes. 13 Indeed, the lowering of the criterion for impaired fasting glucose from 110 to 100 mg/dl was done primarily to equalize the population risk of developing diabetes between impaired fasting glucose and impaired glucose tolerance states. 4,12 The ADA currently recommends the use of fasting glucose tests to diagnose diabetes. Thus, from the standpoint of US clinical practice, monitoring fasting plasma glucose in at-risk patients is necessary. In the present study, all subjects had apparently normal fasting glucose levels that did not suggest diabetes risk by the current definition. However, those who developed diabetes had other adverse characteristics that may help identify their increased risk, namely, high BMI, hypertension, and poor lipid profiles. The univariate comparisons of these risk factors across fasting plasma glucose categories among those who developed diabetes yielded no significant differences. Thus, a consistent profile of predictive characteristics emerged that may assist clinicians in identifying patients for targeted diabetes screening. Although widespread screening with oral glucose tolerance tests is burdensome, performing this test on patients with the high-risk profile we identified might be an appropriate course of action. In addition to this risk profile, smoking increased the risk of diabetes by 36% independently of other factors, a result consistent with previous studies that have found an association between smoking and diabetes. 14-17 Patients with this high-risk profile would likely benefit from lifestyle modifications known to reduce diabetes risk. 18-20 Our study has several limitations. First, we defined diabetes as the entry into the electronic medical record of a

524 The American Journal of Medicine, Vol 121, No 6, June 2008 diagnosis or by a single fasting plasma glucose level greater than 125 mg/dl. Neither of these criteria necessarily defines clinically confirmed diabetes. Oral glucose tolerance tests, which may be a stronger predictor of diabetes, were not available for our subjects. Among those who developed diabetes by our criteria, however, the mean hemoglobin A 1c at diagnosis was more than 7%, a level that strongly suggests that abnormal glucose metabolism has been maintained for several months. Second, because of the observational design, we cannot conclude that higher fasting plasma glucose within the normal range actually causes diabetes; we can only report the strong independent association. Third, it is possible that risk factors we could not include, such as insulin resistance and family history of diabetes, would have accounted for the reported difference in diabetes incidence across fasting plasma glucose categories. We also did not assess the use of drugs known to affect glucose levels (eg, thiazide diuretics, glucocorticosteroids) as covariates. Fourth, we did not randomly select participants for fasting plasma glucose testing. Although most of our results were ascertained from routine lipid screening, the study was nonetheless limited to patients who sought health care. Thus, whether our study sample is representative of all individuals with normal fasting plasma glucose levels cannot be determined. The debate over the correct cut-point for defining impaired fasting glucose has not been settled. Our results demonstrate that increased risk of diabetes extends well below the ADA s current maximum limit of normal plasma glucose of 99 mg/dl. Although impaired fasting glucose may be a useful construct, seemingly normal fasting plasma glucose levels also provide information regarding the future risk of diabetes, especially in conjunction with other known risk factors. ACKNOWLEDGMENT We thank Martha Swain for editorial assistance. References 1. American Diabetes Association. Standards of Medical Care in Diabetes 2007. Diabetes Care. 2007;30:S4-S41. 2. Nathan DM, Davidson MB, DeFronzo RA, et al. Impaired fasting glucose and impaired glucose tolerance: implications for care. Diabetes Care. 2007;30:753-759. 3. American Diabetes Association. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care. 1997;20:s1-s15. 4. Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Follow-up Report on the Diagnosis of Diabetes Mellitus. Diabetes Care. 2003;26:3160-3167. 5. Borch-Johnsen, Colagiuri S, Balkan B, et al. Creating a pandemic of prediabetes: the proposed new diagnostic criteria for impaired fasting glucose. Diabetologia. 2004;47:1396-1402. 6. Dinneen SF, Maldonado III D, Leibson CL, et al. Effects of changing diagnostic criteria on the risk of developing diabetes. Diabetes Care. 1998;21:1408-1413. 7. Tirosh A, Shai I, Tekes-Manova D, et al. Normal fasting plasma glucose levels and type 2 diabetes in young men. N Engl J Med. 2005;353:1454-1462. 8. Nichols GA, Hillier TA, Brown JB. Progression from newly acquired impaired fasting glucose to type 2 diabetes. Diabetes Care. 2007;30: 228-233. 9. Nichols GA, Hillier TA, Brown JB. Progression from newly acquired impaired fasting glucose to type 2 diabetes. Diabetes Care. 2007;30: 228-233. 10. Shaw JE, Zimmet PZ, Hodge AM, et al. Impaired fasting glucose: how low should it go? Diabetes Care. 2000;23:34-39. 11. Tirosh A, Shai I, Tekes-Manova D, et al. Normal fasting plasma glucose levels and type 2 diabetes in young men. N Engl J Med. 2005;353:1454-1462. 12. Unwin N, Shaw J, Zimmet P, Alberti KGMM. Impaired glucose tolerance and impaired fasting glycaemia: the current status on definition and intervention. Diabet Med. 2002;19:708-723. 13. Shaw JE, Zimmet PZ, de Courten M, et al. Impaired fasting glucose or impaired glucose tolerance. What best predicts future diabetes in Mauritius. Diabetes Care. 1999;22:399-402. 14. Meisinger C, Doring A, Thorand B, Lowel H. Association of cigarette smoking and tar and nicotine intake with development of type 2 diabetes mellitus in mean and women from the general population: the MONICA/KORA Augsburg Cohort Study. Diabetologia. 2006;49: 1770-1776. 15. Rimm EB, Manson JE, Stampfer MJ, et al. Cigarette smoking and the risk of diabetes in women. Am J Pub Health. 1993;83:211-214. 16. Manson JE, Ajani UA, Liu S, et al. A prospective study of cigarette smoking and the incidence of diabetes mellitus among US male physicians. Am J Med. 2000;109:538-542. 17. Willi C, Bodenmann P, Ghali WA, et al. Active smoking and the risk of type 2 diabetes: A systematic review and meta-analysis. JAMA. 2007;298:2654-2664. 18. Pan XR, Li GW, Hu YH, et al. Effects of diet and exercise in preventing NIDDM in people with impaired glucose tolerance. The Da Qing IGT and Diabetes Study. Diabetes Care. 1997;20:537-544. 19. Tuomilehto J, Eriksson JG, Valle TT, et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med. 2001;344:1343-1350. 20. Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2 diabetes with life-style intervention or metformin. N Engl J Med. 2002;346:393-403.