Elsa.org.br

C l i n i c a l C a r e / E d u c a t i o n / N u t r i t i o n
O R I G I N A L
Detection of Undiagnosed Diabetes and
Other Hyperglycemia States
The Atherosclerosis Risk in Communities Study
ARIA INˆES SCHMIDT, MD, PHD
DAVID COUPER, PHD
ment of diabetes (2– 4), providing a ratio- RUCE B. DUNCAN, MD, PHD
FREDERICK BRANCATI, MD
LVARO VIGO, MSC
ARON R. FOLSOM, MD
screening these hyperglycemic states will AMES PANKOW, PHD
FOR THE ARIC INVESTIGATORS
inevitably find cases of undiagnosed dia- HRISTIE M. BALLANTYNE, MD
betes and impaired fasting glucose (IFG),benefit from their early detection andtreatment has not been directly docu-mented with randomized trials.
OBJECTIVE — To evaluate screening strategies based on fasting plasma glucose (FPG), clin-
ical information, and the oral glucose tolerance test (OGTT) for detection of diabetes or other hyperglycemic states—impaired fasting glucose (IFG) and impaired glucose tolerance— screening for glucose abnormalities using either fasting glucose or the oral glucosetolerance test (OGTT) in individuals aged RESEARCH DESIGN AND METHODS — We studied 8,286 African-American and
white men and women without known diabetes, aged 53–75 years, who received an OGTTduring the fourth exam of the Atherosclerosis Risk in Communities Study. Using a split sample technique, we estimated the diagnostic properties of various clinical detection rules derived from logistic regression modeling. Screening strategies utilizing FPG, these detection rules, and/or the OGTT were then compared in terms of both the fraction of hyperglycemia cases detected and the additional clinical information. Postchal- sample fraction receiving different screening tests and identified as screen positive.
lenge hyperglycemia has been shown inlarge observational studies to be more RESULTS — Screening based on the IFG cut point (Ն6.1 mmol/l), followed by a clinical
detection rule for those below this value, detected 86.3% of diabetic case subjects and 66.0% of all hyperglycemia cases, identifying 42% of the sample as screen positive. Applying an OGTT for those positive by the rule provides diagnostic labeling and reduces the fraction that is screen positive to 29%. Another strategy, to apply an OGTT to those with an FPG cut point between 5.6and 6.1 mmol/l, also identifies 29% of the sample as screen positive, although it detects slightly CONCLUSIONS — Screening strategies based on FPG, complemented by clinical detection
rules and/or an OGTT, are effective and practical in the detection of hyperglycemic states for the detection of three hyperglycemicstates meriting clinical intervention: un- Diabetes Care 26:1338 –1343, 2003
RESEARCH DESIGN AND
Type2diabetesisaleadingcauseof priorityworldwide(1).Recentclinicaltri- METHODS— The Atherosclerosis
als demonstrated that lifestyle interven- tions in individuals with impaired glucose ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● From the 1Graduate Studies Program in Epidemiology, Federal University of Rio Grande do Sul, Porto munities in 1987–1989. Of these, 11,656 Alegre, Rio Grande do Sul, Brazil; the 2Department of Epidemiology, School of Public Health, University ofNorth Carolina, Chapel Hill, North Carolina; the 3Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, Minnesota; the 4Department of Medicine, Baylor College of Medicine, Houston, Texas; the 5Department of Biostatistics, University of North Carolina, School of Public Health, Chapel Hill, North Carolina; and the 6School of Medicine, Johns Hopkins University, Baltimore, Maryland.
Address correspondence and reprint requests to Bruce B. Duncan, School of Medicine, UFRGS, R. Ramiro Barcelos, 2600/414, Porto Alegre, RS 90035-003. E-mail: [email protected].
Received for publication 11 October 2002 and accepted in revised form 22 January 2003.
known diabetes (report of physician diag- Abbreviations: ARIC, Atherosclerosis Risk in Communities; FPG, fasting plasma glucose; HDL-C, HDL
cholesterol; IFG, impaired fasting glucose; IGT, impaired glucose tolerance; OGTT, oral glucose tolerance test; ROC, receiver operator characteristic.
permitting ascertainment of diabetes sta- A table elsewhere in this issue shows conventional and Syste`me International (SI) units and conversion tus at all four ARIC visits; 52 lacking in- 2003 by the American Diabetes Association.
DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003 Schmidt and Associates
Table 1—Diagnostic properties of different fasting glucose cut points in the detection of
postchallenge diabetes or IGT in the testing sample of 4,143 men and women aged 53–75 in the
were neither white nor African American.
ARIC Study
After these exclusions, 8,286 individualsremained. Comparison of these subjects not to have diabetes (4,6,17) showed that likely to be African American (35.4 vs.
more frequently hypertensive (39 vs.
vs. 5.5 mmol/l) at ARIC baseline. Institu- Sens, sensitivity (percentage screen positive among case subjects); Spec, specificity (percentage screen negative among non– case subjects).
obesity (32%), central obesity (63%), hy- ethnicity, age, parental history of diabe- fasting and 2-h postchallenge plasma glu- Ͻ30 kg/m2, and obesity as Ն30 kg/m2. waist circumference), systolic blood pres- (32%) IFG or IGT. By 2-h plasma glucoseWaist girth was measured at the umbilical variables having lesser impact on the di- agnostic properties of the risk functions, as reflected by changes in the area under erties of fasting glucose cut points in the the receiver operator characteristic (ROC) IGT. The cut point for IFG (Ն6.1 mmol/l) by a hexokinase assay, insulin by nonspe- categorized, following a similar modeling cific radioimmunoassay, and triglycerides function into a simplified clinical score by further increased the detection rate (sen- approximating the function’s ␤ coeffi- sitivity 61.5 and 84.1%, respectively) at Ն11.1 mmol/l; IFG as a fasting value from sample as screen positive. Screening only those with BMI Ն25 kg/m2, a serial strat- each subject’s probability of having diabe- tes or IGT from the risk equation (13). We Cholesterol Education Program’s defini- determined diagnostic properties of rules tion rate at each glucose cut point.
based on fasting glucose cut points and on defined central obesity as a waist circum- cut points of these probabilities. The case ing steps that could follow an initial fast- Ͼ102 cm (40 in) for men; high triglycer- portion of total cases detected by the rule function models in the learning half of theides as Ն150 mg/dl (1.70 mmol/l); low (sensitivity). All analyses were performed sample with fasting glucose Ͻ6.1 mmol/l.
Factors offering greatest predictive ability RESULTS — This sample of 8,286 par-
Ն130/85 mmHg or the use of antihyper- ticipants consisted of 3,165 (38%) white pressure and the use of antihypertensivetensive medication.
cumference, parental history of diabetes, betes or IGT based on logistic regression ethnicity, and sex. For practical purposes, domly selected half of the sample (learn- DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003 Detection of diabetes
Table 2—Diagnostic properties of strategies based on fasting glucose and clinical factors in the
detection of diabetes or IGT in the testing sample of 3,428 men and women with fasting glucose
tic properties diminished rapidly with the <6.1 mmol/l in the ARIC Study
addition of further factors. When weightwas used instead of waist circumference the screening properties of fasting glucose cut points and clinical detection rules de- fined by the probability of having diabetes probabilities are shown in the footnotes.
on the most readily available clinical in- formation, thus excluding lipids. Clinical rules were generated to identify a range of positivity (30 –50%). To allow compari- identifying as screen-positive proportions tivity of 34.4%, equal to that of the 5.6 Sens, sensitivity; Spec, specificity. *Risk functions composed of fasting plasma glucose and clinical variables mmol/l cut point, detected 48.9% (vs.
(age, height, weight, parental history of diabetes, diagnosed hypertension, systolic blood pressure), or these variables plus HDL-C and triglycerides. Risk function with fasting plasma glucose and clinical variables (area under the curve ϭ 0.69): probability (diabetes or IGT) ϭ Ϫ 27.5758 ϩ 0.0590 ϫ age (years) ϩ 0.3382 ϫ parental history of diabetes ϩ 0.0586 ϫ fasting glucose (mmol/l) ϩ 0.00926 ϫ systolic blood pressure superior diagnostic properties for a clini- (mmHg) 0.2182 ϫ hypertensive medication use ϩ 0.00265 ϫ weight (kg) ϩ 0.2298 ϫ height (cm) Ϫ0.00081 ϫ height2 (cm2). Risk function including HDL-C and triglycerides (area under the curve ϭ 0.73): cal rule with 60.4% screen positivity. De- probability (diabetes or IGT) ϭ Ϫ 32.3529 ϩ 0.0631 ϫ age (years) ϩ 0.3537 ϫ parental history of diabetes ϩ 0.0555 ϫ fasting glucose (mmol/l) ϩ 0.00881 ϫ systolic blood pressure (mmHg) ϩ 0.2123 ϫ hyper- information on lipids was included in the tensive medication use ϩ 0.00218 ϫ weight (kg) ϩ 0.2498 ϫ height (cm) Ϫ 0.00084 ϫ height2 (cm2) ϩ 0.0113 ϫ triglycerides (mg/dl) Ϫ 0.00001 ϫ triglycerides2 (mg/dl)2 ϩ 0.0324 ϫ HDL-C (mg/dl) Ϫ0.00017 ϫ HDL-C2 (mg/dl)2. †Estimated probability of having diabetes or IGT; ‡categorical score ϭ 1 point each for age Ն65 years, African-American ethnicity, parental history of diabetes, high waist circumference (women Ͼ88 cm or 35 in, men Ͼ102 cm or 40 in), hypertension (Ն140/90 or using antihypertensive medication), and fasting glucose Ն5.3 mmol/l (area under the curve ϭ 0.65); §metabolic syndrome score ϭ proportion (56.1%) of cases of diabetes.
1 point each for high waist circumference (women Ͼ88 cm or 35 in, men Ͼ102 cm or 40 in), raised blood pressure (Ն130/85 or using antihypertensive medication), low HDL-C (Ͻ40 mg/dl for men and Ͻ50 mg/dlfor women), and high triglycerides (Ն150 mg/dl) (area under the curve ϭ 0.64).
curve for the risk equation including lip-ids was similarly small (0.73 vs. 0.69; seefootnote, Table 2).
screen positive between cut points of this score. The second categorical rule, based positive and detect and explicitly classify egorical factors. The first one attributes one point each for age Ն65 years, a posi- (11), produced detection rates inferior to tive parental history of diabetes, being Af- the conventional IFG cut point identified (diabetes or IFG), detecting 68.8% of the score of Ն4 out of 6 total points identified Ͻ6.1 mmol/l as positive, 52.0% of the screening strategies in the entire testing identified 45.7% of subjects as screensample with diabetes or IGT, and 57.4% positive, increasing the detection rate to with diabetes. Of note are the large jumps certain gold standard for these strategies, DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003 Schmidt and Associates
Table 3—Overall yield of differing screening strategies calculated for the 4,143 individuals in the full testing sample
CDR, clinical detection rule, fasting glucose plus clinical factors, 30% positivity (see Table 2); 2hPG, 2-h postchallenge glucose. *Includes case subjects not labeledexplicitly as IFG, IGT, or diabetic either because they are detected only by a lowered (Ͻ6.1 mmol/l) glucose cut point or only by the clinical detection rule.
fraction explicitly labeled as diabetes or rates. All strategies assume that those with IFG. It detects 51.7% of IGT cases. If la- IFG will not be further classified by 2-h identify those at higher risk for cardiovas- Ͻ6.1 mmol/l, an OGTT could then be ap- were performed, then an additional cular disease while avoiding diagnostic la-plied. This approach decreases screen beling of predisease. Several studies have positivity to 28.8% while explicitly classi- analyzed strategies based on clinical in- CONCLUSIONS — Evidence from
als at high probability to have (18 –22) or five clinical trials that type 2 diabetes can develop (23) diabetes. However, little has screen all those with fasting glucose Ͻ6.1 been done to develop a practical means to pharmacological interventions in individ- of clinically available diabetes risk factors.
set to identify those at “high probability” screening will inevitably find cases of IFG early intervention are less clear. Screening cases of diabetes and 58.4% of the cases of based on combinations of fasting glucose, clinical detection rules, and the OGTT.
IGT cases in the sample. With this strat- For all strategies, a fasting glucose was the egy, 42.1% of the sample are identified as first step, followed by additional screen- cause mortality than does fasting glucose receive a diagnostic label of hyperglyce- (6,7). However, the clinical utility of the results— detecting Ͼ85% of the cases of mia (diabetes, IFG, or IGT). If a clinical desirable to initiate clinical preventive ac- variability, glucose load, time spent in the tained with two strategies: the first one tions, then an additional step could apply then applied a clinical detection rule to and simpler. It is considered equally pre- those below this cut point; the second one this clinical rule. This step does not in- dictive of future diabetes as a 2-h value (5). Our results indicate that almost two- screen positivity to 29.1%, and it permits thirds of those classified as IFG have ei- explicit classification of those with glu- providing further justification for the use used in the first of these strategies is based of this fasting category as the basis for pre- ventive action. The major limitation of the tors, it identifies higher-risk subjects who cose alone applied only to those with BMI IFG cut point is its low sensitivity to de- Ն25 kg/m2 reduced those tested to tect postchallenge hyperglycemia.
considerably lower overall case detection rules costs little and uses readily available DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003 Detection of diabetes
fort to pool data from several large studies internationally so as to develop and more provides a diagnostic hyperglycemia label for a clinical diagnosis. This diagnostic for all positives, which could be of moti- based on fasting glucose, clinical detec- states in our sample—and thus, in abso- use a categorical score instead of a contin- lute terms, it overestimates the yield of and practical in the detection of cases of screen-positive cases the strategies would produce. In this regard, because variabil- tion is likely to beneficial. The adoption of includes African-American ethnicity, it is ity is greatest for the OGTT and probably a given strategy will depend, however, not of less value in non-U.S. settings. A score lowest for predictive equations, screening only on its diagnostic properties but also on the effectiveness of interventions and ities, though less discriminating diagnos- affected and those using clinical detection on resources available for screening, diag- nostic confirmation, and intervention.
relevant information is available. It has subject glucose variability. An additional the advantage, as do our continuous vari- limitation, in this regard, is that we are Acknowledgments — Support for this study
able clinical detection rules, of greater was provided by National Heart, Lung, and generalizability to populations with other with the same glucose test. However, this use of the screening value as part of the practice. A similar limitation is the fact N01-HC-55022 and National Institute of Dia- betes, Digestive and Kidney Diseases Grant naires. Our clinical prediction rules were ceived support from a Centers of Excellence with the increasing availability of precise, Grant of CNPq (the Brazilian National Re-search Council).
rapid means of fasting glucose determina- The authors thank the staff and participants assistance of some form of calculator.
in the ARIC study for their important contri- strategy at a single visit, dismissing those sonal digital assistants and web-based cal- negative at the first step and proceeding data into a preestablished calculator may limitation of our study is that we lacked References
be more practical than all but the simplest tions were developed using readily avail- lence of obesity in our sample, as well as common, growing, serious, costly, andpotentially preventable public health prob- able variables. In this regard, we favored the fact that all participants were either lem. Diabetes Res Clin Pract 50 (Suppl. 2): the use of weight, rather than waist cir- cumference, in risk functions for contin- four U.S. communities, may limit the gen- uous variables, given its more widespread availability in clinical databases and the in a large sample of a practical means of type 2 diabetes with lifestyle intervention vide a clinically relevant increment in test or metformin. N Engl J Med 346:393– 403, 3. Tuomilehto J, Lindstrom J, Eriksson JG, Valle TT, Hamalainen H, Ilanne-ParikkaP, Keinanen-Kiukaanniemi S, Laakso M, istrative or patient care databases already Uusitupa M: Prevention of type 2 diabetes the clinical detection rule because it iden- tifies those presumably at greatest risk.
subjects with impaired glucose tolerance.
Additionally, if OGTT testing is desired, N Engl J Med 344:1343–1350, 2001 the IFG/clinical detection rule strategy re- risk profile for diabetes than the initial the same fraction of hyperglycemia cases.
fasting glucose or a clinical detection rule would present importantly different diag- BV: Effects of diet and exercise in prevent-ing NIDDM in people with impaired glu- nostic properties in those not available for study. A more important limitation of our Diabetes Study. Diabetes Care 20:537– studies in the literature, is that we define data have direct clinical application, an 5. American Diabetes Association: The pre- and validate cases of hyperglycemia based important future step is a collaborative ef- vention or delay of type 2 diabetes. Dia- DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003 Schmidt and Associates
undiagnosed diabetes in a health care set- 6. Qiao Q, Tuomilehto J: Diagnostic criteria 12. Campbell G: Advances in statistical meth- ting. Diabetes Care 22:213–219, 1999 of glucose intolerance and mortality. Min- and laboratory tests. Stat Med 13:499 – ple questionnaire to identify people at in- 13. Szklo M, Nieto FJ: Epidemiology: Beyond creased risk for undiagnosed diabetes.
the Basics. Gaithersburg, MD, Aspen Pub- 14. Chiasson JL, Josse RG, Gomis R, Hanefeld logistic regression equation to screen for ology: Collaborative analysis Of Diagnos- M, Karasik A, Laakso M: Acarbose for pre- diabetes: development and validation. Di- tic criteria in Europe. Lancet 354:617– STOP-NIDDM randomised trial. Lancet 21. Barriga KJ, Hamman RF, Hoag S, Marshall 8. Stern MP, Fatehi P, Williams K, Haffner SM: JA, Shetterly SM: Population screening for Predicting future cardiovascular disease: do 15. Buchanan TA, Xiang AH, Peters RK, Kjos glucose intolerant subjects using decision we need the oral glucose tolerance test? Di- tree analyses. Diabetes Res Clin Pract 34 9. Meigs JB, Nathan DM, D’Agostino RB Sr, ervation of pancreatic ␤-cell function and 22. Griffin SJ, Little PS, Hales CN, Kinmonth glycemia and cardiovascular disease risk: cological treatment of insulin resistance in wards earlier detection of type 2 diabetes the Framingham offspring study. Diabetes high-risk Hispanic women. Diabetes 51: in general practice. Diabetes Metab Res Rev 10. ARIC Investigators: The Atherosclerosis 16. Tuomilehto J: Point: a glucose tolerance 23. Stern MP, Williams K, Haffner SM: Iden- Risk in Communities (ARIC) study. Am J tification of persons at high risk for type 2 (Commentary). Diabetes Care 25:1880 – diabetes mellitus: do we need the oral glu- 11. Expert Panel on Detection, Evaluation, cose tolerance test? Ann Intern Med 136: 17. Davidson MB: Counterpoint: the oral glu- 24. Hunt KJ, Williams K, Haffner SM, Stern mentary). Diabetes Care 25:1883–1885, 18. Baan CA, Ruige JB, Stolk RP, Witteman JC, Dekker JM, Heine RJ, Feskens EJ: Per- Antonio Heart Study (Abstract). Diabetes (Adult Treatment Panel III). JAMA 285: formance of a predictive model to identify DIABETES CARE, VOLUME 26, NUMBER 5, MAY 2003

Source: http://www.elsa.org.br/downloads/Artigos%20em%20PDF/Detection%20of%20Undiagnosed%20Diabetes%20and%20Other%20Hyperglycemia%20States%20-%20ARIC.pdf