Trends in incidence of youth-onset type 1 and type 2 diabetes in the USA, 2002-18: results from the population-based SEARCH for Diabetes in Youth study.

METHODS

SEARCH is a population-based study of diabetes from 2002–2018 of approximately 5 million youth aged <20 years under surveillance (85 million person-years), covering geographically defined populations in Colorado (all 64 counties plus selected American Indian reservations in Arizona and New Mexico), Ohio (eight counties), South Carolina (all 46 counties), Washington (five counties), and Kaiser Permanente Southern California (KPSC) health plan enrollees in seven counties. The combined populations in these areas resemble the US population with regards to race, ethnicity, age, parental education, and household income6. Eligible participants included nonmilitary and noninstitutionalized persons with diabetes other than gestational diabetes diagnosed at age <20 years and who resided in one of the study areas at the time of diagnosis. For persons in California, eligibility required membership in KPSC, and for American Indians, participation in Indian Health Services (IHS) at the time of diagnosis. All participating sites received Institutional Review Board approval for this research. The study protocol is located at https://searchfordiabetes.org.

Trained research staff identified new (incident) diabetes cases through physicians’ report, medical record review, and self-report. Cases were considered valid if the medical record indicated a physician diagnosis, the diagnosis was verified directly by a physician, or the participant was referred directly from a physician, while eligibility was based on age and area of residence in the year of diagnosis. Participants or their parent/guardian completed a survey that recorded age, date of diagnosis, sex, self-reported race and ethnicity, and place of residence at the time of diagnosis. Medical records served as the secondary source for these variables when survey data was unavailble. Diabetes type was the physician-assigned type six months after diagnosis. Date of diagnosis was from medical records or local clinical registries. Case presentation was based on the parent/guardian survey that included the question, “How did you find out you had diabetes?” Response categories were symptoms, checkup (i.e., routine health visit), community screening, or other.

The analyses included all cases with a diagnosis year of 2002–2018, the last complete year for which SEARCH ascertained incident diabetes. T1D cases included those aged birth-19 years of age at diagnosis. T2D cases included those aged 10–19 years. Although SEARCH conducted surveillance of all youth below age 20, few youth below the age of 10 with T2D were identified (225 total: 6 aged 0–4 years and 219 aged 5–9 years; approximately half [110] were aged 9 years). Youth with all other types of diabetes, including secondary forms (e.g., diabetes due to cystic fibrosis or glucocorticoid-induced diabetes) and those with missing/unknown/other types, were excluded from these analyses.

The annual population at risk (denominators) included youth younger than 20 years of age on December 31 of the incident year and civilian residents of the geographic study areas, members of KPSC residing in seven counties in southern California, or IHS beneficiaries at participating Native American reservations. For the geographically based centers, denominators used the bridged-race intercensal and vintage 2019 postcensal population estimates7. For KPSC, addresses were geocoded to the Census block level, and race and ethnic-group–specific proportions were applied to estimate the racial and ethnic-group composition of youths according to age and sex.

For Native American reservations, the IHS user population for the previous three years was used following IHS definitions. The last year of denominator data from the IHS was 2016; these same denominators were used for 2017 and 2018. Denominator estimates were then summed across all five centers.

Unadjusted annual incidence rates were calculated as the number of registered cases divided by the number of persons in the respective group in the surveillance networks over the same period. Rates are presented as 2-year moving averages expressed per 100,000 youth, overall, and according to age group, sex, and race or ethnic group. The 95% confidence intervals (CIs) for the annual unadjusted rates were calculated using the skew-corrected inverted score test, assuming a binomial distribution. Adjustments for age, sex, race, or ethnic group, and estimation of the annual rate of change were performed in a modeling framework.

Trends in incidence were tested with a generalized autoregressive moving average (GARMA) to capture the serial correlation between estimates8. Likelihood-ratio tests were performed to compare three possible formulations: a first-order autoregressive and first-order moving-average model (GARMA [1, 1]), a first-order autoregressive model (GARMA [1, 0]), and a first-order moving-average model (GARMA [0, 1]). Model selection suggested that the first-order moving-average model (GARMA [0, 1]) provided the best fit for most models. Models for the overall incidence estimate adjusted for age, race and sex, are shown in the Appendix. Since the GARMA [0, 1] model also captures the serial correlation between incidence estimates obtained for two consecutive years, the overall trend contains both the linear effect and the moving average effect. Consequently, our change estimates must be interpreted with caution because the dependence on the estimate observed in the previous year is not captured in the linear effect.

Overall trends were adjusted for age, sex, and race or ethnic group, and unadjusted trends in incidence were estimated. The model treated the observed number of cases each year as the outcome and the corresponding denominator as an offset. Homogeneity of effects over time across age, sex, and race or ethnicity was tested by including a group-by-time interaction term in the model. The GARMA model was fitted assuming incident counts follow a negative binomial distribution with a logarithmic link. Likelihood ratio tests for quadratic and cubic trends were also considered.

We assessed the completeness of case ascertainment for the four geographically based centers using the capture-recapture method9. The number of times an individual case was found in a hospital and other clinical setting was used to estimate the number of “recaptured” (and missed) cases. Capture-recapture was not performed in the health plan surveillance area since there were not two independent sources of cases.

Case presentation patterns were examined in 3-year periods to assess change over time, and were examined by month of diagnosis aggregated across all 17 years to explore possible explanations for seasonal (monthly) trends10.

Peak age at diagnosis of diabetes was examined within subgroups of sex, race, and ethnicity for two periods (2002–2009 and 2010–2018). Using a bootstrapping approach, we identified the peak over the entire age range by single year of age and generated 95% CIs. No modeling was conducted.

Seasonality of diagnosis was explored visually using polar seasonal plots which present the distribution of cases by month of diagnosis, by study site, and by two periods (2002–2009 and 2010–2018). Seasonality was tested using a GARMA model, where the monthly count of incident cases served as the outcome. The model was adjusted for age, sex, and race/ethnicity and included a 12-month lag term which was used to test for the presence of a seasonal effect.11

RESULTS

From 2002 through 2018, SEARCH identified 18,169 youth with T1D among 85 million person-years of youth aged 0–19 years and 5,293 youth with T2D among 44 million person-years of youth aged 10–19 years. Based on capture/recapture analysis, few cases were missed. Completeness ranged from 97–99% across the 17-year period for T1D and 91–96% for T2D, and was stable across the age range but dropped by approximately 20% among those aged 18 and 19 for both T1D and T2D.

Overall, the GARMA model suggested an increasing linear trend combined with a positive correlation between two consecutive incidence estimates for both T1D and T2D (Table 1, Figure 1, Appendix Tables 1–6). For T1D, a significant upward trend in the age-, sex- and race/ethnicity-adjusted incidence was observed from 19·5 cases per 100,000 youths per year in 2002–2003 to 22·2 cases per 100,000 youths per year in 2017–2018, with an annual rate of increase of 2·02%, (95% CI 1·54, 2·49%). For the overall estimate, the moving average component revealed a positive correlation between two consecutive incidence estimates (r=0·12; Appendix), suggesting that although the overall tendency is increasing over time, errors around this trend are not independently distributed. Variation in the incidence estimate in a given year also depends on the incidence observed in the preceding year.

The upward trend in the adjusted incidence of T2D was greater, from 9·0 cases per 100,000 youths per year in 2002–2003 to 17·9 cases per 100,000 youths per year in 2017–2018, with an annual rate of increase of 5·31% (95% CI 4·46, 6·17%). The correlation between incidence estimates was 0·20.

Annual rates of increase for T1D was highest for Asian/Pacific Islander, Hispanic, and non-Hispanic Black youth, measuring 4·84%, 4·14%, and 2·93%, respectively. Annual rates of increase for T2D were also highest for Asian/Pacific Islander, Hispanic, and non-Hispanic Black youth, with increases of 8·92%, 7·17%, and 5·99%, respectively. By 2018 in 15–19 year old youth, the overall incidence of T2D had exceeded that of T1D (19·7/100,000 vs 14·6/100,000, respectively).

For T1D, 94–95% of all cases were identified through symptoms, with no significant change over time after adjustment for age, sex, and race (Appendix Table 7). However, a growing proportion of T2D cases were identified by routine health visits (26.4% in 2002 vs. 45.2% in 2018) with a corresponding decline in cases presenting with symptoms (68.9% vs 51.8%) (p<0.0001).

Peak incidence occurred at age 10 years for T1D among 0–19 year-olds and age 16 years for T2D among 10–19 year-olds (Table 2 and Figure 2). For T1D, males had a later peak (12 years) than females (10 years), and Hispanic and Asian Pacific Islander youth (9 years) had earlier peaks than other racial and ethnic groups (10–11 years). For T2D, males and females, and all racial and ethnic minority groups (except non-Hispanic Black youth) had the same peak age at 16 years. Non-Hispanic Black youth had an earlier peak at 13 years. Peak age did not differ between the two periods (2002–2009 and 2010–2018) for either T1D or T2D. Appendix Figure 1 shows the age distributions by race, ethnicity, and sex, and suggest that for T2D in non-Hispanic Black youth the peak occurs earlier in girls than in boys.

The statistical test for seasonality was significant for both T1D (p=0.006) and T2D (p=0.0005) indicating that there was a repeating yearly pattern of diagnoses over the 17 year period. For T1D, the visual inspection of the polar seasonal plots shows a peak of diagnosis in January and a dearth of diagnoses in June (Figure 3). (Two periods are shown to demonstrate the consistency). Smaller sample sizes by study site led to more variability in these patterns (Appendix Figure 2), although the January and June inflections were still apparent. For T2D, the visual inspection of the polar seasonal plots shows a peak of diagnosis in August which was consistent in both periods (Figure 3). Site-specific data show considerable variability owing to smaller sample sizes of T2D and further stratification by site (Appendix Figure 2).

Examination of case presentation by month indicates a common pattern in August that favors routine health visits or community screening with a concomitant modest reduction in cases presenting with symptoms. For T1D, this pattern of increased routine health visits or community screening was most common in August (7%), and for T2D, this presentation pattern was most common in December (42%) and August (41%) (Appendix Table 8).

DISCUSSION

The incidence of T1D and T2D in youth continues to rise in the US. Over the 17 years of the SEARCH study, the incidence of T1D increased annually by 2·02% among 0–19 year-olds, and T2D increased annually by 5·31% among 10–19 year-olds. For both T1D and T2D, the rates of increase were generally higher among racial and ethnic minority youth than those among the non-Hispanic White youth. Specifically, annual percent increases for T1D and T2D were highest for Asian/Pacific Islander, Hispanic, and non-Hispanic Black youth. These findings extend our previous reports, confirming a continuing rise in the incidence4,5 of both T1D and T2D in youth, and giving evidence for a rise in prevalence.12

These findings are consistent with those from Europe and other regions of the world that show a rise in incidence in youth with T1D over the past two to three decades.13,14 Data from other countries (and the US) on the change in the incidence of T2D are limited to a few small studies15–17 yet show a rise in the incidence of T2D, with China, India, and the US showing the highest absolute incidence rates of T2D.18 An analysis of EURODIAB registry data from over 84,000 children with T1D in 22 countries conducted for the period 1989–2013 concluded that T1D has increased by 3% per year, with a possible slowing of this increase between 2004–2008.14 Importantly, the present SEARCH results of increasing incidence over the past several decades for both T1D and T2D agree with the findings of other studies around the globe.

However, the rise in incidence of youth-onset diabetes is in sharp contrast to a declining (or stable) incidence of diabetes in adults. In the US, the National Health Interview Survey documented a decrease in incidence of diagnosed diabetes in adults of 3·1% (annual percent change) between 2007 and 2017.19 This report, using cross-sectional survey data, followed a period of nearly two decades of increasing incidence. The decline was driven predominantly by a decrease in incidence among non-Hispanic White individuals. Incidence rates in racial and ethnic minority groups were stable during this period. Additionally, a recent report from a multicountry analysis with 5 billion person-years of follow-up of adults over the previous decade shows stabilizing or declining incidence of diabetes.20 Further exploration of the deviating trends in incidence in youth (rising) and adults (declining or stable) is needed.

There are several possible explanations for the rise in T2D incidence. First, the incidence is rising in parallel with increasing rates of obesity in youth, a known risk factor for T2D. Between 1999 and 2016, in youth aged 2–19 years, the prevalence of obesity increased from 13.9% to 18.5% in the US, with the highest rates among non-Hispanic black and Hispanic youth.21 Alternatively, some of the rise may be artifactual, reflecting either increases in screening for T2D in at-risk youth or a more accurate diagnosis of T2D. Our data suggest that the mode of presentation for T2D has changed over time, with a growing proportion of cases identified by routine health visits with a corresponding decline in presentation with symptoms. This may lead to an earlier diagnosis of T2D, which may have a (short-term) impact on incidence rates. Additionally, previous analyses from SEARCH5 shows that there has been an improvement over time in the accuracy of diagnosis of T2D. The proportion of T2D cases that meet a gold standard criterion (insulin resistance in the absence of diabetes autoantibodies) rose from 73% in 2002 to 84% in 2016 (p=0.05). Together, these observations are unlikely to fully explain the increasing incidence of T2D.

We make several observations about the age at which diagnosis peaks among youth. Regarding T1D, our work is consistent with the known peak occurring between ages 10 and 14. However, to our knowledge, no reports have narrowed the peak age to a single year. We found that the peak occurred at age 10 overall, with the peak occurring earlier in girls (10 years) than in boys (12 years) and somewhat earlier in Hispanic and Asian Pacific Islander youth (9 years) than in other race/ethnic groups (10–11 years). And, despite an increasing incidence, peak age had not shifted over time.

Regarding T2D, our data are among the first to identify a peak age at diagnosis of youth-onset T2D. First, across sex, race, and ethnicity, peak age was 16 years, except for non-Hispanic Black youth for which peak age occurred at 13 years, and possibly earlier among non-Hispanic Black girls than boys. This pattern coincides with the timing of menarche, which occurs earlier in non-Hispanic Black girls and Hispanic girls than in non-Hispanic White girls, an observation that is partially attributable to disparities in socioeconomic status.22 A second observation is a steep rise between ages 10 to 13 with a peak at age 16 years, followed by a steep decline through age 19 years. The rise appears to initiate at age 9; we observed a large number of cases at that age from among the few youth diagnosed before age 10, but not included in this study. Our capture-recapture analysis shows modest under-ascertainment among those aged 18 and 19, although not enough to fully explain the drop in incidence after its peak. Another possible explanation for the steep decline after age 16 is reduced contact with the health system and reduced screening opportunities. If real, the shape of the distribution may suggest a relatively narrow period of susceptibility in youth, which returns to a lower-risk period until rising again in middle age, between ages 45 and 64 years.23 These observations have important implications for identifying a period ripe for intervention to reduce the risk of T2D in youth. Finally, peak age has not shifted over time, despite an increasing incidence of diabetes.

Seasonal variation in the onset of T1D is well established, with most studies observing a peak in the winter.24 This pattern is observed in the SEARCH data, with a peak in January. Possible explanations for seasonality include fluctuation in hours of sunlight and thus vitamin D levels25 and viral infections.26 We were unable to examine these potential etiologies in SEARCH. Seasonal variation in the onset of T2D in youth has not been previously examined due to the limited youth-onset T2D incidence data. A significant seasonal pattern was observed (p=0.0005). Visually, this pattern is most apparent with an increase in diagnoses occurring in August. A possible explanation for an August peak is an increase in physical exams for school athletic programs, at which time asymptomatic hyperglycemia may be detected. Our data show that the month of August had the highest (T1D) or second highest (T2D) proportion of cases presenting at routine health visits rather than with symptoms, providing some evidence in support of this hypothesis. In conclusion, seasonal patterns were observed for both types of diabetes although seasonality in diagnosis of T2D may indicate a pattern of increased health care contact rather than periods of exposure to viral or environmental triggers as for T1D.

Strengths of the SEARCH study include a rigorous approach to identifying all incident cases of diabetes in youth with a high completeness rate over a 17-year period. This is particularly unique for T2D in youth, for which no other surveillance studies have been conducted in the US. SEARCH is also the only US-based registry of both T1D and T2D in youth with representation of all major racial and ethnic groups. Peak age at diagnosis and seasonality of diagnosis are both relatively well-known for T1D in youth, but, to our knowledge, SEARCH is the only incidence study that can examine peak age and season at diagnosis for T2D. Limitations include a modestly lower ascertainment rate for ages 18 and 19, which only partially explains the drop in incidence rate following peak incidence; and, the use of a descriptive statistical approach to identify the age at which the maximum incidence rate is observed. A more formal analytic approach would be to derive the probability distribution of the maximum incidence as a function of age.27,28 However, this is a more methodologic approach and out of scope for this paper.

The SEARCH study29 has been a national resource to explore the epidemiology of youth-onset diabetes in the US and the consequences of these chronic conditions. Our data show a rising incidence of diabetes in youth in the US. Peak age of diagnosis occurs at 10 years for T1D and 16 years for T2D. Seasonal patterns were observed for both types of diabetes. These data document an expanding population of young adults at risk for developing early complications of diabetes2, which appear to be more devastating when diabetes develops at a young age.30