Validity of the primary care diagnosis of diabetes in veterans in the southeastern United States

diabetes research and clinical practice 91 (2011) 395–400

Contents lists available at ScienceDirect

Diabetes Research and Clinical Practice journ al h omepage: www .elsevier.co m/lo cate/diabres

Validity of the primary care diagnosis of diabetes in veterans in the southeastern United States Jennifer G. Twombly a,b, Qi Long c, Ming Zhu c, Lisa-Ann Fraser d, Darin E. Olson a,b, Peter W.F. Wilson a,e, K.M. Venkat Narayan f,g, Lawrence S. Phillips a,b,* a

Veterans’ Affairs Medical Center, Decatur, GA, United States Division of Endocrinology and Metabolism, Emory University School of Medicine, Emory University, Atlanta, GA, United States c Department of Biostatistics and Bioinformatics, Rollins School of Public Health, Emory University, Atlanta, GA, United States d Division of Endocrinology & Metabolism, Department of Medicine, University of Western Ontario, London, Ontario, Canada e Division of Cardiology, Emory University School of Medicine, Emory University, Atlanta, GA, United States f Department of Medicine, Emory University School of Medicine, Emory University, Atlanta, GA, United States g Hubert Department of Global Health, Rollins School of Public Health, Emory University, Atlanta, GA, United States b

article info

abstract

Article history:

Aims: To determine the validity of diagnosis of diabetes in primary care.

Received 24 June 2010

Methods: Patients with initial primary care diagnosis (ICD-9 code 250.xx) were compared to

Received in revised form

matched controls (without code or diabetes drugs), and patients meeting VA Diabetes

22 October 2010

Epidemiology Cohort (DEpiC) criteria (any 250.xx twice, or diabetes drug) in ‘‘diagnostic

Accepted 1 November 2010

accuracy’’ (whether hyperglycemia preceded diagnosis) and ‘‘predictive accuracy’’ (whether

Published on line 26 November 2010

diabetes drug or A1c 6.5% followed diagnosis). Results: Only 1.8% of primary care diagnoses met ADA criteria, while nonstandard non-

Keywords:

fasting morning glucose 126 mg/dl or A1c 6.5% were utilized in 51.5%; broad ‘‘diagnostic

Type 2 diabetes

accuracy’’ criteria were met in 53% of 2980 primary care vs. 2% of 13,397 control ( p < 0.001),

Diagnosis

and 60% of 2456 DEpiC patients ( p < 0.001). ‘‘Predictive accuracy’’ was 88% in primary care

Delay

diagnosis vs. 12% control ( p < 0.001) and 93% DEpiC patients ( p = 0.08), but was higher if ADA

Primary care

criteria were met. Delay from hyperglycemia to diagnosis averaged 12.5 months in primary

Reminders

care vs. 20.1 months in DEpiC patients ( p < 0.001). Conclusions: While generally not based on ADA criteria, the primary care diagnosis of diabetes is valid, and identifies patients earlier than detection by DEpiC criteria. Either primary care diagnosis or DEpiC criteria could be used to trigger electronic reminders aimed to facilitate management. # 2010 Published by Elsevier Ireland Ltd.

* Corresponding author at: Division of Endocrinology, Emory University, 101 Woodruff Circle, WMRB Room 1027, Atlanta, GA 30322, United States. Tel.: +1 404 727 1392; fax: +1 404 727 1300. E-mail addresses: [email protected], [email protected] (L.S. Phillips). Abbreviations: ADA, American Diabetes Association; A1c, Hemoglobin A1c; Black, Self-identification; FDA, Food and Drug Administration; NDDG, National Diabetes Data Group; NGSP, National Glycohemoglobin Standardization Program; NHANES, National health and nutrition examination survey; OGTT, Oral glucose tolerance test; VA, Veterans’ Affairs; VISN, Veterans’ Integrated Service Network; White, Self-identification. 0168-8227/$ – see front matter # 2010 Published by Elsevier Ireland Ltd. doi:10.1016/j.diabres.2010.11.001

396 1.

diabetes research and clinical practice 91 (2011) 395–400

Introduction

Diabetes is a major public health problem. The prevalence of diabetes in the U.S. was estimated at 12.9% of the adult population aged 20 years based on NHANES 2005–2006 [1], the number of patients with diabetes is projected to rise to 48 million by 2050 [2], and the 2007 national economic burden of diabetes and prediabetes was estimated to be $218 billion in direct and indirect costs [3]. Estimated marginal yearly costs [3,4] are lower in individuals with undiagnosed diabetes ($2864) and prediabetes ($443) compared to patients with diagnosed type 2 diabetes ($9975). Therefore, it is possible that strategies aimed at prevention and effective management could be cost-effective. One possible strategy would be to focus on patients who are early in their natural histories, where preventive management would be expected to slow the progression of the disease and the development of costly complications [5], and pharmacotherapy is more effective [6]. However, targeting patients with newly diagnosed diabetes could be appropriate only if the diagnosis of diabetes is valid. Although there are well established guidelines for diagnosing diabetes, the process of diagnosis in general practice is not well understood. Some practitioners may measure mainly random plasma glucose levels [7]. Using a VA database, we assessed the validity of the diagnosis in veterans in the southeastern U.S., part of the largest healthcare system in the U.S. We evaluated both ‘‘diagnostic accuracy’’ (whether the diagnosis was preceded by antecedent hyperglycemia) and ‘‘predictive accuracy’’ (whether the diagnosis was followed by unequivocal hyperglycemia or prescription of an antihyperglycemic drug).

2.

Subjects, materials, and methods

2.1.

Study population

This study was approved by the Emory University Institutional Review Board. The sample population was a retrospective cohort of patients in the Corporate Data Warehouse database for VISN 7 (VA medical centers in South Carolina, Georgia, and Alabama). We selected patients who had (i) diabetes diagnosed 10/01/02 or later (to allow time for antecedent laboratory assessment, since population of the database began 10/01/98), and (ii) consistent primary care (3 visits over 2 years prior to the ‘‘index date’’ of diagnosis, and 4 visits over 3 years after the diagnosis – including the ‘‘diagnostic’’ visit as one of the visits); the conservative requirement of at least one follow-up visit a year was designed to ensure that providers had some opportunity to interact with the patients. In VISN 7, 269,434 veterans had a primary care visit in 2008, and 44,806 had 7 consecutive years of follow-up with 1 primary care visit a year and 2 outpatient glucose measurements each year. We compared diagnostic and predictive accuracy in (i) 2980 ‘‘initial primary care diagnosis’’ patients – veterans who were initially diagnosed in primary care settings (in the electronic medical record, initial use of the 250.xx ICD-9 code at a primary care visit, without use of the code or prescription of a diabetes drug during the preceding year); (ii)

13,397 controls matched for age, gender, race, VA facility, and index date (assigned to provide a 2/3 ratio of antecedent and subsequent follow-up), who had no use of 250.xx or a diabetes drug prior to the index date; and (iii) 2456 ‘‘DEpiC’’ patients, veterans meeting VA Diabetes Epidemiology Cohort (DEpiC) criteria [8] – any use of 250.xx twice, or a diabetes drug. The date of diagnosis was taken as the date criteria were met for inclusion in groups (i) or (iii); assignment as ‘‘initial primary care diagnosis’’ or ‘‘DEpiC’’ was exclusive, according to which criteria were met first. Within 3 months before and 6 weeks after the date of diagnosis, A1c averaged 7.3  1.9% (mean  SD) in the 1869 initial primary care diagnosis and 7.3  1.9% in the 1616 DEpiC patients with available data ( p = 0.9). Each primary care diagnosis patient was matched with 1 control; while we present findings with this full dataset, results were similar in a subset of 2062 primary care diagnosis patients each matched 1:2 with controls (not shown).

2.2.

Diagnostic accuracy

To test diagnostic accuracy, we examined the extent to which the diagnosis reflected antecedent outpatient hyperglycemia (inpatient glucose values were excluded to limit artifactual glucose elevations reflecting stress from acute illnesses), as: (i) outpatient morning plasma glucose 126 mg/dl (sampled 0630–1000 h); (ii) outpatient random plasma glucose 200 mg/dl (1001–1800 h); (iii) fasting plasma glucose 126 mg/dl; (iv) 2 h OGTT plasma glucose 200 mg/dl; (v) A1c 6.5% [random plasma glucose 126 mg/dl or A1c 6.5% confers a high likelihood of having diabetes [9,10], and VA health care providers frequently check A1c levels in patients not known to have diabetes]; (vi) diagnosis outside of the VA system, at another VA, or use of a diabetes medication. ADA criteria at that time included (ii–iv) [11]. Diagnostic accuracy was examined both by chart review and database analyses. For chart review, trained research nurses examined the records of 300 patients at the Atlanta VA with respect to criteria (i)–(vi) – 100 each of randomly selected patients meeting initial primary care diagnosis, control, and DEpiC criteria as above. Subsequent database analyses focused on criteria (i)–(v), since criterion (vi) required analysis of chart narrative.

2.3.

Predictive accuracy

To test predictive accuracy, we examined the extent to which the diagnosis was followed by criteria reflecting the presence of the disease – initiation of pharmacologic treatment, or the presence of laboratory values that would justify treatment for most patients with diabetes. This assessment utilized only database analyses.

2.4.

Assays

Glucose and A1c were measured in VA clinical chemistry laboratories. Glucose was assessed with FDA-approved platforms such as the Beckman DXC or LX20 (Beckman Coulter, Fullerton, CA) or Roche P or COBAS (Roche Diagnostics, Indianapolis, IN), and A1c with NGSP-approved methods,

397

diabetes research and clinical practice 91 (2011) 395–400

Table 1 – Diagnostic accuracy – percent of patients with antecedent outpatient hyperglycemia prior to the diagnosis of diabetes (initial outpatient primary care diagnosis, or meeting DEpiC criteria for diagnosis). Primary care diagnosis Any criterion once (D1) Atlanta VA 83% VISN 7 74% Two criteria or any criterion twice (D2) Atlanta VA 73% VISN 7 53%

p*

Control

DEpiC

(76–90%) (73–76%)

5% (1–9%) 9% (9–10%)

91% (85–97%) 79% (77–80%)

0.14 <0.001

(64–82%) (52–55%)

0% 2% (2–3%)

76% (68–84%) 60% (58–62%)

0.75 <0.001

Shown are proportions and 95% confidence intervals (CI). Criteria for antecedent hyperglycemia are provided in Table 2. In each instance, the prevalence of meeting criteria for antecedent hyperglycemia was significantly greater in primary care diagnosis and DEpiC patients compared to controls (all p < 0.001). * Significances shown are comparisons between primary care diagnosis and DEpiC populations.

mainly HPLC such as the Tosoh G7 (Tosoh Bioscience, South San Francisco, CA), but also immunological methods (Beckman and Roche, above).

2.5.

Statistical analysis

For descriptive statistics, continuous variables were analyzed by t-tests and categorical variables by Chi-square or Fisher’s Exact tests. Diagnostic accuracy was compared by Chi-square tests, and predictive accuracy was assessed both by Chisquare tests and Kaplan–Meier analyses. All analyses utilized SAS version 9.2 (SAS Institute, Cary, NC).

3.

Results

Table 1 in the online Appendix shows the demographic breakdown of the three study groups for the Atlanta VA and for VISN 7. The groups were roughly similar, and in VISN 7 overall, the studied patients had a mean age of 63 years and a BMI of 30 kg/m2, and were largely male; about 38% were black, 22% were white, and the remainder had unknown race. Table 1 in this manuscript shows that many patients are diagnosed with diabetes without having had antecedent outpatient hyperglycemia. Criteria for antecedent hyperglycemia (Table 2) were met more often for Atlanta VA patients than VISN 7 patients, in part because of narrative (found on chart review, but not available in the database) describing

diagnoses made outside the VA, use of diabetes drugs on presentation, etc. However, there were 2 occurrences of antecedent hyperglycemia in only 53% of primary care diagnosis and 60% of DEpiC patients in VISN 7 ( p < 0.001), both p < 0.001 compared to 2% in controls. Moreover, as shown in Table 2, even when antecedent hyperglycemia was present, the glucose levels often did not meet American Diabetes Association (ADA) criteria for hyperglycemia – fasting plasma glucose 126 mg/dl and/or 2 h OGTT plasma glucose 200 mg/dl and/or random plasma glucose 200 mg/dl [11]. Among the 1588 primary care diagnosis patients who had two episodes of antecedent hyperglycemia, the diagnosis was preceded by strict ADA criteria in only 3.3%, while 13.9% of the diagnoses were preceded by a combination of ADA and nonstandard criteria, and 82.7% only by criteria that were not recommended by the ADA at the time – two of nonfasting morning plasma glucose 126 mg/dl and/or A1c 6.5%. Among the 8 VA facilities in SC, GA, and AL, delay of diagnosis after two occurrences of antecedent hyperglycemia was consistently greater in DEpiC than in primary care diagnosis patients (21  17 vs. 12  15 months, mean  SD, p < 0.001, see online Appendix, Table 2). Despite uneven ‘‘diagnostic accuracy’’ (prevalence of antecedent hyperglycemia), ‘‘predictive accuracy’’ was high – subsequent unequivocal hyperglycemia or prescription of a diabetes drug was frequent. As shown in Table 3, within 3 years after the diagnosis, 70% of primary care diagnosis

Table 2 – Antecedent hyperglycemia in initial primary care diagnosis patients. Criterion

Morning plasma glucose 126 mg/dl Random plasma glucose 200 mg/dl A1c 6.5% Fasting plasma glucose 126 mg/dl 2 h OGTT plasma glucose 200 mg/dl

Morning plasma glucose 126 mg/dl

Random plasma glucose 200 mg/dl

A1c 6.5%

Fasting plasma glucose 126 mg/dl

694

84 22

431 63 189

15 3 49 19

2 h OGTT plasma glucose 200 mg/dl 3 0 7 9 0

For the 1588 initial primary care diagnosis patients whose diagnoses were preceded by ADA criteria for antecedent hyperglycemia (random plasma glucose 200 mg/dl or fasting plasma glucose 126 mg/dl or 2 h OGTT glucose 200 mg/dl) or nonstandard criteria (plasma glucose 126 mg/dl before 10 am or A1c 6.5%), shown are the numbers of pairs of occurrences of meeting the criteria (one criterion twice, or two criteria once).

398

diabetes research and clinical practice 91 (2011) 395–400

Table 3 – Predictive accuracy. Criterion A1c 7.0% or use of diabetes medication within 3 years after diagnosis date A1c 6.5% or use of diabetes medication within 4 years after diagnosis date

Primary care diagnosis

Control

70% (68–71%) 88% (87–90%)

2% (1–2%) 12% (9–14%)

DEpiC 75% (73–77%) 93% (91–96%)

Shown are the prevalences of either being prescribed an antidiabetic medication, or exhibiting an A1c level sufficient to justify initiating treatment in many patients with diabetes, in primary care diagnosis, control, and DEpiC groups (mean and 95% CI).

patients and 75% of DEpiC patients had A1c 7.0% or were prescribed a diabetes medication ( p < 0.001), both p < 0.001 vs. 2% of controls. Four years after the diagnosis, 88% of primary care diagnosis patients and 93% of DEpiC patients had A1c 6.5% or were using a diabetes medication ( p = 0.08), both p < 0.001 vs. 12% of controls. Evidence of hyperglycemia in controls presumably reflects the incident development of diabetes. Online Appendix Fig. 1 shows that primary care diagnosis and DEpiC patients exhibited comparable rates of development of A1c 6.5%, but DEpiC patients were more likely to be prescribed a diabetes medication during the first several years after diagnosis. As shown in Fig. 1, among primary care diagnosis patients who had 4 years of follow-up, 100% of the 24 patients who met ADA diagnostic criteria twice also met 4-year ‘‘predictive accuracy’’ criteria (above), vs. 98.5% of the 133 who met a combination of ADA and nonstandard diagnostic criteria, 93.0% of the 797 who met only nonstandard diagnostic criteria ( p = 0.90 and p = 0.001 for combination and nonstandard groups vs. the ADA criteria group), and only 80.0% of the

[()TD$FIG]

Fig. 1 – Predictive accuracy – the fulfillment over time of ‘‘predictive accuracy’’ criteria (initial prescription of a diabetes drug and/or A1c levels I6.5%) in the primary care diagnosis patients who had 4 years of follow-up; n = 24 patients who exhibited two occurrences of antecedent ADA diagnostic criteria, n = 221 patients who exhibited two occurrences of a mixture of ADA and nonstandard diagnostic criteria, n = 797 patients who exhibited two occurrences only of nonstandard diagnostic criteria, and n = 690 patients who did not exhibit two occurrences of either ADA and/or nonstandard antecedent diagnostic criteria. Dashes plus dots: no criteria met twice; dots: nonstandard criteria met twice; dashes: ADA criteria met once, and nonstandard criteria met another time; solid line: ADA criteria met twice.

690 who did not meet antecedent diagnostic criteria twice ( p < 0.001 vs. each of the other groups).

4.

Discussion

Despite the extensive attention given to developing and promulgating guidelines for the diagnosis of diabetes, there is little knowledge of the process of diagnosis in real-world primary care settings. Our study demonstrates that the diagnosis of diabetes is often not based on ADA guidelines, and ‘‘diagnostic accuracy’’ overall is only modest. The initial diagnosis of diabetes in VA primary care settings in the southeastern U.S. was preceded by two occurrences of outpatient hyperglycemia that met ADA guidelines [11] in only 1.8% of patients; 51.5% of patients had antecedent nonfasting morning plasma glucose 126 mg/dl and/or A1c 6.5% [levels that confer a high likelihood that diabetes would be found if an OGTT were performed [9,10]], and 47% of patients had lower glucose levels. However, ‘‘predictive accuracy’’ was much higher – the diagnosis was often followed by unequivocal hyperglycemia or prescription of an antidiabetic medication, even in patients who did not meet antecedent diagnostic criteria. Moreover, the primary care diagnosis of diabetes was also generally comparable to the more extensive VA Diabetes Epidemiology Cohort criteria [any use of 250.xx twice, or use of a diabetes drug [8]] in both diagnostic and predictive accuracy. Thus, even though the diagnosis of diabetes in primary care settings may not be based on standard ADA criteria, the diagnosis appears to be valid 88% of the time within four years of follow-up. However, our findings also support the use of the current diagnostic criteria – and possibly the use of screening approaches to prompt assessments with greater specificity – since predictive accuracy was significantly higher when antecedent diagnostic criteria were met. Few previous studies have examined how diabetes is diagnosed in actual clinical practice, and whether that diagnosis is valid. Ealovega et al. [7] reported that only 69% of patients in a managed care practice in Michigan had any glucose measurements over 1998–2000, and that 95% of the initial glucose measurements were random plasma glucose values; 3% were fasting plasma glucose, 2% were A1c, and <1% were OGTTs. Our data cannot be compared directly with theirs because of differences in design, but 17% of their subjects with abnormal glucose values (fasting plasma glucose 110 mg/dl, random plasma glucose 130 mg/dl, A1c 6.4%, or 2 h OGTT 140 mg/dl) were diagnosed with diabetes within 6 months – evidence that diabetes may be diagnosed without fulfillment of strict ADA criteria in other settings. Miller et al. [12] found

diabetes research and clinical practice 91 (2011) 395–400

that VA patients with a single outpatient diagnosis of diabetes in 1997–1999 had a 40% prevalence of self-reported diabetes, and a 37% prevalence of antecedent A1c 7.0%, but did not assess other measures of antecedent hyperglycemia, or subsequent hyperglycemia; they did note that a single outpatient diagnosis provided comparable sensitivity and specificity but somewhat less frequent use of medication compared to any two diagnoses – similar to our findings. O’Connor et al. [13] reported that use of 250.xx twice at primary care visits to a multispeciality practice in Minnesota in 1992–1994 was often associated with self-reported diabetes, but when it was not, NDDG criteria for diabetes were generally not met. However, antecedent hyperglycemia was assessed in only a small proportion of subjects, there was no evaluation of subsequent hyperglycemia, and there was no evaluation of a single outpatient diagnosis. In a subsequent study of patients diagnosed with diabetes in 1993–1996 [14], those investigators found that 72% of diagnoses did not meet ADA criteria – consistent with our findings – but they did not determine whether the diagnosis was accurate in such patients. To evaluate such ‘‘false-positives’’, two of the authors (J.G.T. and L.-A.F.) examined a convenience sample of 44 records of Atlanta VA patients who were diagnosed with diabetes but failed to meet our ‘‘predictive accuracy’’ criteria for unequivocal hyperglycemia during the subsequent 4-year period. Three patients developed comorbid disease (cancer, etc.) which was associated with weight loss and improved glucose levels, but all of the others had A1c levels of 6.0–6.4% (while managed with diet alone) – suggesting that the diagnosis may have been accurate in such patients as well. Other recent reports [15,16] describe the changing prevalence of diagnosed diabetes, and one noted the ADA criteria that were to have been used for the diagnosis [15], but neither assessed the validity of the diagnosis. Validation is important because of the potential of the initial diagnosis of diabetes to be ‘‘actionable’’, a basis for monitoring and directing the provision of care. If the diagnosis were to be invalid – because it had little relation to antecedent or subsequent hyperglycemia – then it should not be actionable. However, even though the diagnoses in our population were rarely based on ADA criteria, our demonstration of empirical diagnostic and predictive validity should allow the diagnosis to be used to evaluate and prompt management – both to permit assessment of subsequent provision of care, and to prompt provider adherence to guidelines for management. The strengths of our study include substantial numbers of subjects; access to a robust database which included linked laboratory, medication, clinical, and demographic data as well as diagnostic codes; and the opportunity to examine clinical records as well as administrative data. Since the study was retrospective, there was no way to ensure that subjects would have fasting plasma glucose or OGTT assessments, as would have been needed to permit the diagnosis to have been made more systematically and more often by ADA criteria, but our findings show that a non-ADA-based approach can still meet reasonable criteria for accuracy. We found that diagnostic accuracy based on manual record review was somewhat higher than accuracy based only on database analysis (Table 2), but we were unable to examine a larger number of records

399

due to the time and cost involved. We also found the delay of diagnosis to be quite variable across the 8 VA facilities in SC, GA, and AL (online Appendix Table 1), but the studies needed to understand the basis for such variability were beyond the scope of our analysis. Finally, further studies would also be needed to determine the extent to which our findings apply to other VA regions or to practice settings outside the VA. The median A1c at diagnosis in one U.K. general practice was 8.2% in symptomatic and 7.45% in asymptomatic patients [15], and mean A1c was 9.9% and 8.1%, respectively, in Minnesota [14] – compared to 7.3% overall in the present studies. To the extent that A1c at diagnosis reflects delay after initial development of hyperglycemia [17,18], the delay may be greater in other settings than in the VA. Why the diagnosis is delayed is a separate question. When patients exhibit typical symptoms in conjunction with high glucose levels, there may be little delay in making the diagnosis. For asymptomatic patients, providers may gradually notice that ‘‘random’’ blood glucose levels are 126 mg/dl, and make the diagnosis – often without definitive diagnostic tests. Our analyses indicate that although the initial outpatient primary care diagnosis of diabetes according to use of the 250.xx ICD-9 code in veterans in the southeastern U.S. may often not be based on strict ADA criteria, its diagnostic and predictive accuracy is only slightly less than that of more elaborate criteria such as those of the VA Diabetes Epidemiology Cohort. Either of these approaches for identification of diabetes should be sufficiently accurate to justify use in electronic medical record systems to trigger processes for improvement in diabetes management.

Conflict of interest Dr. Phillips has had relationships including consultancies (Boehringer-Mannheim, Merck, Novartis); honoraria (Merck, Novartis); and research grants (Novo Nordisk, Amylin, Eli Lilly, GlaxoSmithKline, Sanofi-Aventis, Roche, Diasome, Merck, Sankyo). These are relationships that existed within the past 3 years; none have anything to do with the submitted manuscript. Other authors had no potential conflicts of interest to report.

Acknowledgements This work was supported in part by NIH award DK066204, and VA HSR&D awards SHP 08-144 and IIR 07-138. We thank Christine Jasien, Johnita Byrd-Sellers, Jane Caudle, and Circe Tsui (systems and database support), Margaret Jenkins, Jennifer Ieong, and Martha Forrester (chart reviews), and Jennifer Michaels (research staff support) for their assistance. This work was presented in part at the national meeting of VA Health Services Research and Development investigators in Baltimore, February 2009, and at the national meeting of the Endocrine Society in San Diego, June 2010. The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. Dr. Phillips

400

diabetes research and clinical practice 91 (2011) 395–400

had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The authors declare that there is no conflict of interest associated with this manuscript.

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.diabres.2010.11.001.

references [1] Cowie CC, Rust KF, Ford ES, Eberhardt MS, Byrd-Holt DD, Li C, et al. Full accounting of diabetes and pre-diabetes in the U.S. population in 1988–1994 and 2005–2006. Diab Care 2009;32(2):287–94. [2] Narayan KM, Boyle JP, Geiss LS, Saaddine JB, Thompson TJ. Impact of recent increase in incidence on future diabetes burden: U.S., 2005–2050. Diab Care 2006;29(9):2114–6. [3] Dall TM, Zhang Y, Chen YJ, Quick WW, Yang WG, Fogli J. The economic burden of diabetes. Health Aff (Millwood) 2010;29(2). doi: 10.1377/hlthaff.2009.0155. [4] Dall TM, Mann SE, Zhang Y, Quick WW, Seifert RF, Martin J, et al. Distinguishing the economic costs associated with type 1 and type 2 diabetes. Popul Health Manag 2009;12(2):103–10. [5] Pogach LM, Brietzke SA, Cowan Jr CL, Conlin P, Walder DJ, Sawin CT. Development of evidence-based clinical practice guidelines for diabetes: the Department of Veterans Affairs/ Department of Defense guidelines initiative. Diab Care 2004;27(Suppl. (2)):B82–9. [6] Nichols GA, Alexander CM, Girman CJ, Kamal-Bahl SJ, Brown JB. Treatment escalation and rise in HbA1c following successful initial metformin therapy. Diab Care 2006;29(3):504–9. [7] Ealovega MW, Tabaei BP, Brandle M, Burke R, Herman WH. Opportunistic screening for diabetes in routine clinical practice. Diab Care 2004;27:9–12.

[8] Miller DR, Pogach LM. Longitudinal approaches to evaluate health care quality and outcomes: the Veterans Health Administration diabetes epidemiology cohorts. J Diab Sci Technol 2008;2:24–32. [9] Ziemer DC, Kolm P, Foster JK, Weintraub WS, Vaccarino V, Rhee MK, et al. Random plasma glucose in serendipitous screening for glucose intolerance: screening for impaired glucose tolerance study 2. J Gen Intern Med 2008;23(5):528–35. [10] The International Expert Committee. International Expert Committee report on the role of the A1c assay in the diagnosis of diabetes. Diab Care 2009;32(7):1327–34. [11] American Diabetes Assocation. Diagnosis and classification of diabetes mellitus. Diab Care 2009;32(Suppl. (1)):S62–7. [12] Miller DR, Safford MM, Pogach LM. Who has diabetes? Best estimates of diabetes prevalence in the Department of Veterans Affairs based on computerized patient data. Diab Care 2004;27(Suppl. (2)):B10–21. [13] O’Connor PJ, Rush WA, Pronk NP, Cherney LM. Identifying diabetes mellitus or heart disease among health maintenance organization members: sensitivity, specificity, predictive value, and cost of survey and database methods. Am J Manag Care 1998;4(3):335–42. [14] O’Connor PJ, Gregg E, Rush WA, Cherney LM, Stiffman MN, Engelgau MM. Diabetes: how are we diagnosing and initially managing it? Ann Fam Med 2006;4(1):15–22. [15] Evans P, Langley P, Gray DP. Diagnosing type 2 diabetes before patients complain of diabetic symptoms—clinical opportunistic screening in a single general practice. Fam Pract 2008;25(5):376–81. [16] Nielen MM, Schellevis FG, Verheij RA. Inter-practice variation in diagnosing hypertension and diabetes mellitus: a cross-sectional study in general practice. BMC Fam Pract 2009;10:6. [17] Mason CC, Hanson RL, Knowler WC. Progression to type 2 diabetes characterized by moderate then rapid glucose increases. Diabetes 2007;56(8):2054–61. [18] Tabak AG, Jokela M, Akbaraly TN, Brunner EJ, Kivimaki M, Witte DR. Trajectories of glycaemia, insulin sensitivity, and insulin secretion before diagnosis of type 2 diabetes: an analysis from the Whitehall II study. Lancet 2009;373(9682):2215–21.