Improved Metabolic Control in Children and Adolescents With Type 1 Diabetes: A trend analysis using prospective multicenter data from Germany and Austria

The current study is based on prospective data from the German and Austrian DPV documentation system (Diabetessoftware für Prospektive Verlaufsbeobachtung) (12,13). Within the framework of the quality assurance and scientific research project DPV, German and Austrian diabetes centers (hospitals and practices) document treatment and outcome of routine diabetes care prospectively using the DPV software. Twice a year, locally collected anonymous longitudinal data are transmitted for central plausibility checks and analyses. Inconsistent data are reported back to participating centers for validation and correction. Overall, 326 German and 9 Austrian diabetes centers participated in the DPV project up to 2010.

For the present analysis, data from patients with type 1 diabetes were selected from the available DPV database in September 2010. Patients and medical data were eligible for inclusion in the analysis when meeting the following criteria: documented data within 1995–2009, age at follow-up visit <20 years, and disease duration at follow-up visit >2 years. Applying the described criteria and then selecting individual patient data of the most recent year of follow-up resulted in a study sample of 30,708 patients from 305 diabetes centers (296 German and 9 Austrian centers) with 86,914 documented medical visits. Finally, individual patient data of the most recent year were aggregated for each patient to set up the final analysis dataset.

Demographic data included age at onset and at follow-up, sex, duration of diabetes, year of follow-up, and migration background. Patients with at least one parent born outside of Germany or Austria were considered to have a migration background.

Clinical data were weight, height, modalities of insulin treatment (number of injections or pump therapy), daily dose of insulin (units per kilogram body weight), HbA_1c, and the occurrence of hypoglycemic events and diabetic ketoacidosis (DKA). Acute complications were assessed at each follow-up visit for the period since the past visit (but at most for a 1-year period), giving an observation period of at most 2 years. Information on BMI (kg/m²), insulin dose, and hypoglycemic events was not available for 687, 404, and 2,748 patients, respectively.

In total, 86,914 HbA_1c measurements in 30,708 patients were available for analysis. The number of measurements per year increased from 675 (in 332 patients) in 1995 to 7,207 (in 2,286 patients) in 2008 and added up to 31,921 measurements (in 13,557 patients) in 2009. During 1996–2008, the average number of HbA_1c measurements per patient and year ranged between 2.8 (1996) and 3.4 (2004). The overall mean (SD) of the annual averages was 3.1 (0.4). In order to adjust for different laboratory methods, local HbA_1c values were mathematically standardized to the DCCT reference range (4.05–6.05%) using the “multiple of the mean” transformation method (1). BMI values were transformed to standard deviation scores (BMI-SDS) based on German reference values by applying the LMS method (17,18).

For each patient, clinical data (BMI-SDS, HbA_1c, number of injections, and daily insulin dose) from the most recent year of follow-up were averaged, and numbers of hypoglycemic and DKA events were added up. According to current recommendations of the International Society for Pediatric and Adolescent Diabetes (ISPAD) (19), HbA_1c values <7.5, 7.5–9.0, and >9.0% were considered as good, moderate, and poor metabolic control, respectively.

For analysis of metabolic control and acute complications, age at visit and duration of diabetes were categorized as >2–5, >5–10, >10–15, and >15–20 years and >2–5, >5–10, and >10 years, respectively. BMI-SDS and insulin dosage were classified into terciles.

Insulin therapy was categorized as conventional therapy (CT) (one to three daily injection time points), MDI (greater than or equal to four daily injection time points), and CSII. Patients were classified into therapy groups according to the treatment that they received for the most part during the most recent year. Treatment centers were grouped into large or small centers (according to >100 or ≤100 patients treated annually during 2005–2009, respectively) and general care or rehabilitation facilities.

Hypoglycemic events were classified, according to ISPAD guidelines (20), as severe hypoglycemia when requiring assistance (ISPAD grade 2 and 3) and as hypoglycemic coma (ISPAD grade 3) when loss of consciousness or seizures occurred. An event of DKA was defined as hyperglycemia with pH value <7.3 and/or hospital admission due to DKA.

For descriptive analysis, mean, SD, or SE were calculated for continuous variables and percentages for categorical variables. Rates of ketoacidosis and hypoglycemia were estimated by the person-years (PYs) method assuming Poisson distribution of events and given as incident number of events per 100 PYs ± SE.

Multiple generalized linear mixed regression models were used to assess the effect of potentially influencing factors on metabolic control (linear regression for HbA_1c and logistic regression for HbA_1c >7.5% and for HbA_1c >9.0%) and rates of hypoglycemia and DKA (Poisson regression) in order to account for confounding effects. Age and diabetes duration at follow-up, sex, BMI-SDS (terciles), year of treatment, migration background, type of insulin treatment, insulin dose (terciles), and size and type of center were modeled as independent fixed effects. In order to account for variation between diabetes centers, treatment center was modeled as random effect. In Poisson regression models, overdispersion of rates was taken into account. Results of regression analyses are presented as multiple adjusted means, multiple adjusted rates, odds ratios (ORs), or relative risks (RRs) including 95% CIs. Within the regression approach, F tests were used to test for differences between groups.

P < 0.05 was considered statistically significant. All analyses were performed with SAS for Windows version 9.2 (SAS Institute, Cary, NC).

Mean age of the cohort (n = 30,708) was 14.6 ± 3.7 years. Mean age at onset and mean diabetes duration were 7.9 ± 4.0 and 6.7 ± 3.6 years, respectively; 52.1% (n = 16,014) of patients were male. Migration background was present in 12.0% (n = 3,683) of patients.

Among the 305 diabetes centers, 211 (96.2%) were pediatric centers, 155 (50.8%) were classified as large, and 293 (96.1%) were general care facilities; 73.1% (n = 22,462) of patients were treated in large centers, 95.8% (n = 29,433) in pediatric centers, and 94.5% (n = 29,020) in general care facilities.

Overall, 10.1% (n = 3,100) of the patients were treated with CT, 65.0% (n = 19,962) with MDI, and 24.9% (n = 7,646) with CSII. The overall average number of daily injections of patients on injection therapy, i.e., on CT or MDI, was 4.5 ± 1.1. Mean daily insulin dose (n = 30,304) was 0.90 ± 0.29 units/kg body weight.

Average HbA_1c was 8.4 ± 1.7%. 33.7% (n = 10,341) of patients achieved HbA_1c values below the recommended target of 7.5%. 38.1% (n = 11,695) of patients had values between 7.5 and 9.0%, and 28.2% (n = 8,672) of patients were in poor metabolic control with HbA_1c >9.0%.

Eight point nine percent (n = 2,477) of patients had at least one severe hypoglycemic event and 2.4% (n = 683) at least one hypoglycemic coma. During a total of 22,633.12 PYs, 4,315 events of severe hypoglycemia and 903 of hypoglycemic coma were observed in the cohort, corresponding to crude event rates of 19.1 ± 0.29 and 4.0 ± 0.13 per 100 PYs. Among patients with episodes of severe hypoglycemia, average HbA_1c was lower (8.1 ± 1.5 vs. 8.4 ± 1.7%, P < 0.001) and consistently the proportion of patients with HbA_1c <7.5% was higher (38.0 vs. 33.7%, P < 0.001) compared with those without severe hypoglycemia. Multiple adjustment for confounders affected estimates only slightly.

At least one DKA event occurred in 4.1% (n = 1,259) of patients. A total of 1,476 DKA episodes (total PYs, 24,917.50) were observed, giving a crude DKA incidence rate of 5.9 ± 0.15 per 100 PYs.

Mode of insulin therapy changed importantly during the study period (Supplementary Fig. 1). In 1995, 37.7% of patients were treated with CT. This proportion decreased to 7.1% in 2009. The rate of MDI therapy increased from 61.4% in 1995 to 78.1% in 2003 and decreased thereafter to 56.3% in 2009. Contemporaneously, the portion of patients with CSII rose continuously from 0.9% in 1995 to 36.6% in 2009.

Unadjusted mean HbA_1c decreased from 8.7 ± 1.8% in 1995 to 8.1 ± 1.5% in 2009, with an average absolute annual decrease in HbA_1c of 0.054% (95% CI 0.048–0.059%, P < 0.001). The annual decrease in HbA_1c differed significantly between treatment groups (P = 0.005). The unadjusted decrease was greater in the CSII group (0.075%; 0.059–0.091%) than in the MDI (0.049%; 0.043–0.055%) and CT groups (0.045%; 0.032–0.057%).

Likewise, the unadjusted proportion of patients with HbA_1c >7.5% decreased steadily from 75.6 ± 2.4% (±SE) in 1995 to 61.9 ± 0.4% in 2009. The respective average OR per year was 0.961 (95% CI 0.954–0.968, P < 0.001). The unadjusted proportion of patients with poor metabolic control (HbA_1c >9.0%) decreased even more steeply from 39.8 ± 2.7% in 1995 to 20.6 ± 0.3% in 2009, with a corresponding OR of 0.931 (0.925–0.938, P < 0.001). The decreasing trend in proportion of patients with HbA_1c >7.5 or >9.0% differed significantly between treatment regimens (P < 0.001). The unadjusted ORs for HbA_1c >7.5% were 0.934 (0.913–0.955), 0.958 (0.950–0.966), and 0.986 (0.969–1.002) for the CSII, MDI, and CT regimen, respectively. Corresponding ORs for HbA_1c >9.0% were 0.897 (0.878–0.917), 0.941 (0.933–0.948), and 0.940 (0.924–0.957).

After multiple adjustment for confounders, the decline in mean HbA_1c over the study period remained significant in the whole cohort, with an estimated average absolute annual decrease of 0.038% (95% CI 0.032–0.043%, P < 0.001) (Fig. 1A and Supplementary Table 1). However, differences in the annual decrease between treatment regimens disappeared (P = 0.703); the adjusted annual decreases were 0.032% (0.016–0.048%), 0.039% (0.032–0.045%), and 0.036% (0.024–0.048%) for the CSII, MDI, and the CT regimen, respectively.

The decreasing trends in proportions of patients with HbA_1c >7.5 or >9.0% were also slightly attenuated after multiple adjustment but remained significant (Fig. 1B and C and Supplementary Table 1). The overall adjusted average ORs per year for HbA_1c >7.5 or >9.0% were 0.969 (95% CI 0.961–0.977, P < 0.001) and 0.948 (0.941–0.956, P < 0.001), respectively. This corresponds to an annual decrease in the odds for HbA_1c >7.5 or >9.0% by 3.1% (95% CI 2.3–3.9%) and 5.2% (4.4–5.9%). Trend differences in the proportion of patients with HbA_1c >7.5% between insulin regimens were diminished after multiple adjustments but remained significant (P = 0.036), in contrast trend differences in the proportion of patients with HbA_1c >9.0% dissolved (P = 0.697). The adjusted OR for HbA_1c >7.5% was still lower for CSII (0.970; 95% CI 0.947–0.993) and MDI (0.964; 0.955–0.973) compared with CT (0.989; 0.972–1.007). Respective ORs for HbA_1c >9.0% were 0.940 (0.919–0.962), 0.950 (0.941–0.958), and 0.946 (0.929–0.964).

In multiple regression analysis age, sex, diabetes duration, migration background, BMI-SDS, and daily insulin dose were significant predictors of metabolic control (assessed as HbA_1c, proportion HbA_1c >7.5 or >9.0%) (Table 1). Older and female patients, patients with longer diabetes duration, higher insulin dose, or a migration background, and patients in the upper BMI-SDS tercile had poorer metabolic control. Mode of insulin therapy was significantly associated only with the proportion of patients having HbA_1c >9.0%; patients with more intensive insulin therapy were less frequent in poor metabolic control.

In multiple regression, size of diabetes center was significantly associated with mean HbA_1c but not with moderate or poor metabolic control. Adjusted mean HbA_1c was higher among patients treated in large centers. Metabolic outcomes did not significantly differ between patients treated in general care or rehabilitation facilities, but rehabilitation patients tended to have poorer metabolic control.

Unadjusted rate of severe hypoglycemia decreased steadily from 54.1 ± 5.8 per 100 PYs in 1995 to 15.1 ± 0.4 per 100 PYs in 2009. According to the estimated RR for a 1-year period (0.950; 95% CI 0.920–0.980, P = 0.001), the rate of severe hypoglycemia dropped on average by 5.0% (95% CI 2.0–8.0%) per year. The unadjusted rate of hypoglycemic coma decreased even more during the study period from 15.4 ± 3.1 per 100 PYs in 1995 to 2.3 ± 0.2 per 100 PYs in 2009. The respective average RR estimate per 1-year period was 0.926 (0.910–0.943, P < 0.001).

The unadjusted rate of DKA varied significantly over the study period between 2.0 ± 0.9 per 100 PYs in 1995 and 8.8 ± 0.7 per 100 PYs in 2007 (P < 0.001). However, the DKA rate showed no significant log-linear time trend according to the estimated annual RR of 1.017 (95% CI 0.989–1.045, P = 0.241).

Adjustment for confounders affected the decreasing trends for hypoglycemic events only slightly and the trends remained significant (P < 0.001) (Fig. 1D and E and Supplementary Table 2). On average, the rate of severe hypoglycemia decreased by 5.2% (95% CI 2.1–8.2%) per year and the rate of hypoglycemic coma by 8.3% (5.0–11.5%) per year. The overall variation of the DKA rate was not significant after multiple adjustment (P = 0.409). The adjusted RR per year for DKA was 1.023 (95% CI 0.986–1.060, P = 0.225), thus indicating no significant log-linear trend (Fig. 1F and Supplementary Table 2).

This large multicenter study showed a significant improvement in metabolic control in German and Austrian children and adolescents with type 1 diabetes during 1995–2009. Both average HbA_1c and proportion of patients with poor metabolic control decreased over time independent of confounders. After multiple adjustment, improvement in average HbA_1c over time was not different between treatment regimens. Besides treatment year, the main influencing factors of metabolic control were age, sex, diabetes duration, migration background, BMI-SDS, and daily insulin dose. A simultaneous decrease of hypoglycemia rate was observed, whereas the incidence rate of hospitalized DKA remained almost stable.

Although being not so distinct, the observed improvement in metabolic control over the past 15 years is in agreement with reports from several recent studies from other countries. A decline of average HbA_1c has been observed in cohorts from Western Australia, Denmark, and Norway (4,10,11). Concordantly, the proportion of children with an HbA_1c level <8% increased in Norway and the U.S. (11,5). However, the Hvidoere studies did not find an improvement in glycemic control over the period 1995–2005 (7,8).

The observed improvement in glycemic control in the current study was not explained by the considerable intensification of insulin treatment during the past decade or other confounders. Further, in the current study, improvement in average HbA_1c over time was comparable for different treatment regimens. This supports the view that other factors, such as the development of multidisciplinary diabetes care teams and improvements in structural quality of diabetes care and patient education, may have accounted for the observed trend (8,21,22).

Several studies have demonstrated that intensive treatment and lower HbA_1c levels are associated with an increased risk of severe hypoglycemia (2,4,6,7,14), but other studies did not support such a relationship (15,16). Although patients with severe hypoglycemia had lower average HbA_1c than those without severe hypoglycemia, the current study provides evidence, in concordance with other reports (5,9,10), that metabolic control can be improved on average in the diabetic population in routine care without an increased risk of hypoglycemia. The DKA rate showed a slight, but nonsignificant, increase over time in the current study, in concordance with previous reports (9,11).

The overall mean HbA_1c level (8.4%) and the proportion of patients with good (HbA_1c <7.5%, 33.7%) or poor metabolic control (HbA_1c >9.0%, 28.2%) in the study cohort were within the ranges found in several previous studies (5–8,13–15,21,23–25).

Older age, female sex, and longer diabetes duration were significantly associated with worse metabolic control, affirming previously reported findings (4,6–8,10,13–16,21,23–25). The varying quality of metabolic control may in part be attributable to differences in insulin sensitivity, treatment compliance, or insulin needs related to these factors.

In the current study, migration background was also a significant predictor of poor metabolic control, as reported previously (13,15,24). This association may partly be related to language difficulties (8) and lower socioeconomic status (24).

Previous findings on the association between metabolic control and BMI are quite inconsistent. Several studies found no significant association (8,15,16,24,25), but positive (14,23) and inverse association has also been reported (21). The present results indicate that patients with a high BMI-SDS (upper tercile) may be prone to poorer metabolic control. On the other hand, long-standing poor glycemic control could lead to higher insulin dose and hence to weight gain. Because of its cross-sectional design, the current study cannot clear up the causality of the association.

The type of insulin regimen is supposed to have an impact on metabolic control. Although several studies did not indicate an association between insulin regimen/number of daily insulin injections and HbA_1c levels (6,14,16), in other studies, patients with an increased number of daily insulin doses exhibited poorer metabolic control (12,15,21,23,25). The insulin regimen found to be associated with best HbA_1c level varied between studies, ranging from conventional therapy and twice-daily regimen (8,12) to thrice-daily injections and pump therapy (10,21,25). In the current study, insulin regimen did not significantly affect the average HbA_1c level, but the proportion of patients with HbA_1c >9% among those using CT was significantly higher than among those with intensified therapy (MDI or CSII). With respect to the inconclusive data, it has to be noted that these results are from observational studies and not randomized trials. Therefore, HbA_1c levels are influenced by many factors beyond insulin regimen. Most importantly, in routine care, the choice of insulin regimen depends in particular on metabolic control. Patients with poor metabolic control are often transferred to intensified regimens. Thus, intensified treatment is likely to be a consequence of poor control rather than a cause.

In accordance with the results of the majority of studies (7,8,10,14,16,21,23), we found a positive association between daily insulin dose and HbA_1c level. The cross-sectional design of most studies, however, limits conclusions that can be drawn on the direction of the observed association. Further, it has to be considered that insulin data usually represent the recommended insulin dose, which possibly differs from the actual dose applied.

Patients treated in larger centers showed on average a higher HbA_1c level than patients of smaller centers. This may be attributed to the fact that large centers possibly care for a greater portion of patients with situations difficult to manage. Otherwise, large centers may not be able to attend to single patients as it might be possible in smaller centers. However, the size of the center did not affect the proportion of patients with HbA_1c >7.5 or 9% in the current study. In the Hvidoere studies (8), center resources were thought to account for significant differences in metabolic control between centers, but it was concluded that a motivated and well-organized diabetes care team is an important determinant of metabolic outcomes rather than the mere size of staff (8,22).

In our study, the proportion of patients with poor metabolic control (HbA_1c >9.0%) tended to be higher in rehabilitation compared with general care centers. This is most likely due to the fact that poor glycemic control is a major indication for the transfer to rehabilitation.

Some strength and limitations of the current study have to be noted. This large study shares the shortcomings of all observational studies compared with randomized trials. Because of the observational design, the study could not control for all influencing factors on metabolic control and acute complications, and thus observed associations could not be proven to be unbiased or to reflect causal effects. Another shortcoming is that HbA_1c levels were not measured centrally. However, HbA_1c values were mathematically standardized to the DCCT normal range in order to reduce between-laboratory variation. Therefore, the observed trend in glycemic control can be assumed to be valid. Hypoglycemic events could have been underestimated because of self-reporting. However, as differential reporting over the study period is unlikely, the trend estimates can be assumed to reflect a real trend.

The main strength of the study is that it comprises a large cohort of children and adolescents and data from small and large secondary level and university diabetes care centers. The study mirrors real-life data of daily pediatric diabetes care almost on a population basis.

In summary, this large multicenter study showed a significant improvement in metabolic control among diabetic children and adolescents during the past decade and a simultaneous decrease in the rate of severe hypoglycemic events. The improvement in glycemic control was not fully explained by demographic factors and changes in the mode of insulin treatment. Thus, other factors, such as improvement in resources, organization and attitudes of diabetes care teams, and patient education, also may have accounted for the observed trend.

This study was supported by the Competence Network for Diabetes Mellitus funded by the Federal Ministry of Education and Research (FKZ 01GI0802 and 01GI0859). The German Diabetes Center is institutionally funded by the German Ministry of Health and the Ministry of Innovation, Sciences, and Research of the Federal State of North Rhine-Westphalia. The DPV software development was supported by Novo Nordisk Pharma GmbH (Germany), the Dr. Bürger-Büsing Foundation, the German Diabetes Foundation, and the German Ministry of Health. No other potential conflicts of interest relevant to this article were reported.

J.R. researched data, performed the statistical analysis, and wrote the manuscript. A.D. and B.K. contributed to discussion and reviewed the manuscript. A.H. contributed to the development of the DPV documentation software and data handling. A.S., C.B., E.M.G., C.K., and S.E.H. contributed to discussion and reviewed the manuscript. R.W.H. researched data, contributed to discussion, reviewed the manuscript, and is principal investigator of the DPV Initiative.

The authors thank all German and Austrian diabetes centers participating in the DPV initiative.