The breadth of this study required multiple data sources and methods, with data covering several years sometimes combined to increase sample size. Data prior to 2012 were used when 2012 data were unavailable. Estimates of diabetes prevalence and diagnosed status by state, age, sex, and insurance type came from national surveys. Analysis of medical claims provided information on medical expenditures and the proportion of diabetes cases that were T2DM, were receiving medication and recommended exams, had indications of poorly controlled diabetes, and had diabetes-related complications. This study used secondary data sources and received an exemption from the New England Institutional Review Board.
Estimating prevalence of diagnosed and undiagnosed diabetes
The approach to estimate prevalence of diagnosed and undiagnosed diabetes by population age (20–34, 35–44, 45–54, 55–59, 60–64, 65–70, and over 70), sex, insurance type (commercial, Medicare, Medicaid, and uninsured), and state is documented elsewhere, though detailed prevalence estimates by insurance type have not been previously published [3, 4, 10]. Using the 2012 American Community Survey (ACS, n = 2,375,715), we constructed a population file where each person residing in the community was matched with a similar person from the combined 2011 and 2012 Behavioral Risk Factor Surveillance System (BRFSS, n = 982,154) and each person residing in a nursing home was matched with a similar person from the 2004 National Nursing Home Survey (NNHS, n = 14,017). Detailed information on file construction is published elsewhere, but the ACS-BRFSS match criteria used state, age, sex, race/ethnicity, medical insurance status, and household income [3, 4]. The ACS-NNHS match criteria used age, sex, and race/ethnicity. The 2011 and 2012 BRFSS files were combined to increase sample size and assume little change in diabetes diagnosis rates between 2011 and 2012 controlling for patient demographics. The 2004 NNHS is the most recently available person-level file for the nursing home population, but the underlying rate of diagnosed diabetes was adjusted to reflect a 2011 national study that found diabetes prevalence was close to 33% among nursing home residents [11].
Diagnosed diabetes status (and hypertension and hyperlipidemia status) from BRFSS was determined based on previously having been told by a health professional that the respondent had the condition. Diabetes status from NNHS was determined by ICD-9 diagnosis code (250.xx); hypertension and hyperlipidemia status were determined by ICD 401.xx and 272.xx, respectively. Applying ACS sampling weights provides state-level estimates of adults with diagnosed diabetes and prevalence of conditions such as hypertension and hyperlipidemia among this population by demographic and insurance type.
To estimate prevalence of undiagnosed diabetes by state, age group, sex, and insurance type, we extrapolated national rates to states using the detailed state population characteristics available in the constructed ACS-BRFSS-NNHS file. We first analyzed national data in the combined 2009–2010 and 2011–2012 files of the National Health and Nutrition Examination Survey (NHANES) for adults who (a) had not previously been told by a health professional they had diabetes; (b) were not taking insulin; (c) were not pregnant; and (d) had laboratory results for hemoglobin A1c, fasting plasma glucose test, 2-h oral glucose tolerance test, or a combination of tests. We estimated a logistic regression where the dependent variable was diabetes (n = 649) as determined by laboratory results (see Additional file 1) exceeding thresholds for diabetes, versus no diabetes (n = 9,296). The explanatory variables were age group, sex, race/ethnicity, body weight category, insurance type, current smoker, year, and previous history or diagnosis of asthma, arthritis, heart attack, stroke, cancer, hypertension, hypercholesterolemia, and cardiovascular disease. We applied regression results (see Additional file 1) to the constructed ACS-BRFSS-NNHS 2012 population file to estimate the overall prevalence of undiagnosed diabetes by state.
Estimating prevalence of T2DM among the insured population
We analyzed 2011–2012 medical and pharmacy claims for commercially insured adults in OptumInsight’s de-identified Normative Health Information (dNHI) database) (n = 29,948,496), for the Medicare population using the 2011 Medicare Standard Analytical File 5% sample (n = 2,805,812), and for the Medicaid population using the 2008 Mini-Max file (n = 3,095,634). All analyses were done by state, age, and sex. The dNHI database consists of longitudinally linked and de-identified individual-level data from one of the nation’s largest private insurance plans. The Medicare extract contains medical and prescription claims. Mini-Max is a 5% sample of the Medicaid Analytic eXtract data-a set of person-level data files on service utilization and payments for more than 60 million Medicaid enrollees extracted from the Medicaid Statistical Information System. Patients analyzed in each of these databases were continuously enrolled in a fee-for-service coverage type plan with no more than one gap in enrollment of up to 45 days during the measurement year for the commercially insured population, and were enrolled for all 12 months for the Medicare and Medicaid populations.
We identified patients with diabetes if the patient had at least one emergency department visit or hospitalization or two separate ambulatory visits with a diabetes diagnosis (ICD-9 of 250.xx) submitted during the year, or if the patient used insulin or other diabetes-related medications. Sample inclusion and exclusion criteria and the algorithm for distinguishing whether a patient had type 1 or type 2 diabetes are described in Additional file 1. Patients with a diagnosis of gestational diabetes were excluded. This analysis assumes that within strata defined by age group and sex, the proportion of diabetes cases that are T2DM remained relatively constant between 2012 and 2008 for the Medicaid population and between 2012 and 2011 for the Medicare population.
Estimating the proportion of T2DM patients receiving medication and exams
Using pharmacy claims from the dNHI, Medicaid Mini-MAX data, and Medicare Part D files for each population strata, we calculated the percentage of patients with claims for insulin, non-insulin injectables, or oral antidiabetic agents. We calculated the percentage of patients with at least one claim for anti-hypertensive medications, statins, and angiotensin-converting enzyme (ACE) inhibitors/angiotensin II receptor blockers (ARB). Using procedure codes (see Additional file 1) we calculated the percentage of patients in each population stratum with indication of at least one cholesterol screening test, urine albumin test, or retinal eye exam during the year. These medications and exams are measures recommended by the HEDIS Comprehensive Diabetes Care 2012, and HEDIS is the source of the drug codes and procedure codes used.
Estimating the characteristics and medical expenditures of treated T2DM patients by controlled status
Whether people have their diabetes under control is generally determined by hemoglobin A1c levels – with A1c < 7% often considered tight control, A1c > 9% considered uncontrolled, and recommended individual patient targets as high as 8.5% depending on a patient’s circumstances [1]. Lab results with A1c values were available only for a subset of the commercially insured patients and unavailable for Medicare and Medicaid beneficiaries. Therefore, to identify patients with uncontrolled diabetes, we used ICD-9 diagnosis codes of 250.x2 and 250.x3 in any claim during the year. For discussion purposes, we refer to “uncontrolled” or “poorly controlled” status for anyone with an ICD-9 code indicating uncontrolled diabetes at some point during the year, and use the term “controlled” to define patients with no indication of uncontrolled diabetes. Use of one claim for uncontrolled could overestimate the number of patients with uncontrolled diabetes; however, use of a diagnosis code rather than A1c results could also miss some patients with uncontrolled diabetes, so there are potentially errors in both directions. Later we discuss the limitations of using ICD-9 versus A1c to define uncontrolled status.
We used primary ICD-9 diagnosis codes (see Additional file 1) in medical claims to identify the presence of eight categories of diabetes complications: neurological symptoms, peripheral vascular disease, cardiovascular disease, renal complications, endocrine/metabolic complications, ophthalmic complications, other complications, and orthopedic problems. Using medical claims to indicate comorbidity presence could undercount prevalence of complications. Diabetes is one of multiple risk factors for these complications, and we report total medical expenditures by category because estimating the proportion of complications attributable to diabetes is beyond the scope of this study. We calculated the total annual medical expenditures and pharmacy expenditures for all medical conditions, with all costs inflated to 2012 dollars using the medical and pharmacy cost components of the Consumer Price Index.