Discussion
These data provide contemporary population-based estimates of diabetes incidence among 7 million adults insured through 11 US integrated health-care delivery systems. The incidence rates we found are higher than those seen in US surveillance data that are based on self-reported diagnoses, probably because our comprehensive definition included persons with diagnostic laboratory values who might not yet have been diagnosed by their clinician. Our incidence rates are also higher than those found in other epidemiologic cohorts, but they are based on larger and more diverse sampling frames than have previously been used.
As expected, diabetes incidence varied widely as a function of age, BMI, and race/ethnicity in our data. Overall, we found no evidence that diabetes incidence increased between 2006 and 2010, although incidence may have increased slightly in 2011. These findings were consistent across all 11 sites and are consistent with those of other studies. For example, in the United Kingdom, diabetes incidence increased steadily from 2006 through 2009 before declining slightly in 2010. In Denmark, diabetes incidence stabilized between 2004 and 2007. The US Centers for Disease Control and Prevention reported a dramatic increase in annual new self-reported cases of diabetes from 1997 to 2007, but little change from 2008 to 2010. A recent study of self-reported diabetes incidence in New York City also suggested a decline between 2004 and 2008.
The development of diabetes is a biological process, the precise timing of which can only be accurately assessed with frequently repeated glycemia testing. Our incidence estimates included cases identified from occasional testing given during the course of clinical care, as well as other indications for diabetes, such as provider-assigned diagnosis codes or filled prescriptions for diabetes medications. Identification of diabetes in observational data is subject to variation in screening practices, and several considerations could influence year-to-year screening and subsequent identification of new cases of diabetes. The proportion of our sample that received any form of glycemia testing increased over time, which may itself increase incidence, but changes in how diabetes was identified clinically may also have played a role in our findings.
In 2010, the American Diabetes Association added HbA1c testing as a means of diagnosing diabetes, having previously recommended only FPG tests, random glucose tests (if symptoms were present), or oral glucose tolerance tests as valid diagnostic tools. This was somewhat controversial, because these laboratory tests measure different phenotypes of dysglycemia that do not perfectly overlap. Other studies have suggested that at the recommended diagnostic threshold of 6.5% (48 mmol/mol), HbA1c concentration would identify fewer persons with diabetes than FPG or oral glucose tolerance tests. In our data, however, the shift in testing practice to greater use of HbA1c testing corresponded with an increase in incidence. Though some of the rise in incidence could be attributed to increased testing, the shift in testing methods from FPG to HbA1c may also have introduced artifactual variation in incidence rates. Because this was an observational analysis, we could not determine whether the people identified as having diabetes through FPG testing would also have been identified at the same point in time through HbA1c testing or vice versa.
The increase in overall testing also makes drawing conclusions about the impact of changes in testing modality difficult, and the degree to which this shift in practice may have contributed to variation in diabetes incidence deserves careful consideration. It is possible that some patients who would have been identified with earlier methods are now being missed by increased reliance on HbA1c testing or that patients now being identified would have been missed by previous methods. Although such patients would probably be diagnosed eventually, the resulting delay in treatment could affect the long-term development of diabetes complications. Studies evaluating the question of diagnostic delays and the impact on future public health are critically needed. Furthermore, because we applied the same criteria for incidence throughout the study, it is possible that we "diagnosed" persons with HbA1c values prior to 2010 who would not have been recognized as having diabetes. There is evidence that many clinicians were already using HbA1c for diagnosis prior to 2010, but because we required 2 separate events to define diabetes and the sole use of HbA1c for diagnosis was infrequent, it is unlikely that incidence rates in the earlier years were artificially raised by premature application of the HbA1c criterion.
Trends in incidence rates during the study period differed by race and ethnicity. It is widely known that blacks, Asians, Hispanics, and Native Americans are at higher risk for developing diabetes, and in a previous study of one of the participating health plans, Karter et al. reported a much higher incidence of diabetes in many minority groups (e.g., Pacific Islanders, South Asians, Filipinos, Koreans) than in non-Hispanic whites. However, this does not explain why incidence rates would increase over time within select racial or ethnic groups. Investigators in the Diabetes Prevention Program reported higher HbA1c values among non-Hispanic black, Hispanic, American Indian, and Asian participants compared with whites both before and after adjustment for a wide range of possible confounders. Our findings are consistent with the concern that HbA1c values vary by race/ethnicity at similar glucose levels.
Another potential concern is that true incidence may be rising faster than observed incidence but will appear to decline or remain flat over time if the cohort is from a closed population (fixed cohort) in which the persons at highest risk are identified early in the observation period, leaving only lower-risk individuals in the at-risk pool. Although the 11 health-care systems in our study form a dynamic cohort insofar as new members can join every year, we observed that the mean number of years of health-plan enrollment prior to the analysis year of the at-risk pool increased between 2006 and 2011, from 6.2 years to 9.4 years. This suggests that our analyses preferentially followed the same individuals over time, and that the enrollment of new members at risk for diabetes occurred more slowly than the disenrollment of existing at-risk members. Conversely, longer periods of observation would allow greater opportunity for screening and diagnosis, which should produce higher incidence rates in later years—the opposite of what we found. Thus, despite aging, we suspect that longer-term members who had not yet developed diabetes were probably at lower risk than those diagnosed in earlier years, because those at greatest risk had already been recognized, and the age-specific incidence rates we observed were consistent with this hypothesis. This may explain the flat trend in incidence from 2006 through 2011.
Strengths of our study include the sample size of nearly 7 million adult health-plan members from geographically distributed integrated health-care delivery systems across the United States; defined populations that included both Medicare and Medicaid patients; the availability of electronic health record data; and analysis of 7 consecutive years of health-plan enrollment. Observational studies have inherent limitations, including the aforementioned variation in screening and detection over time. Other differences among our study sites in how electronic health records are used and variation in the completeness of data could lead to inaccuracies, potentially biasing our estimates of diabetes incidence. However, our findings were consistent across the participating sites.
To match our estimates to the methodology used by the Centers for Disease Control and Prevention, we considered Hispanic persons to be Hispanic regardless of race, based primarily on self-reported information. Approximately 22% of our population were missing data on race/ethnicity, and 34% were missing BMI data. Rates of missing data differed over time and across sites, which could have biased our incidence estimates to the extent that characteristics of members with missing data are also associated with diabetes risk. The membership of our health-care systems may not generalize to less integrated systems or to uninsured individuals, and market penetration varies among the systems included. However, previous studies have shown that members of integrated health-care systems reflect the diversity of the populations from which they are drawn and demonstrate similar disenrollment rates regardless of diabetes status. We did not distinguish between cases of type 1 diabetes and cases of type 2 diabetes, but limiting our study population to adults aged 20 years or older meant that the vast majority of incident cases were type 2. Our case definition included abnormal laboratory values regardless of the presence of a physician-coded diagnosis. Thus, it is possible that some patients with incident cases had undiagnosed diabetes. To identify incident diabetes, we required an extended period of prior health-plan eligibility without any indication of diabetes. This would have excluded some incident cases that did not meet the pre-eligibility criterion, resulting in more conservative incidence estimates.
In conclusion, we found no significant increase in diabetes incidence among adults between 2006 and 2010 and a small but significant increase in 2011 that coincided with a shift in diagnostic testing practice from FPG testing to use of the HbA1c assay. The increase in diabetes incidence rates was significant only among racial and ethnic minority groups and among very obese persons. As a result, disparities in diabetes incidence by race and ethnicity appear to be increasing over time. Future trends in overall diabetes incidence in the United States are likely to be driven in part by the aging of the population, increasing racial and ethnic diversity, and persistently high rates of obesity. Furthermore, because diabetes remission is rare and mortality among people with diabetes is declining, prevalence continues to increase. The diabetes epidemic is far from over.