Diabetic Ketoacidosis at Onset of Type 1 Diabetes and Long-term HbA_1c in 7,961 Children and Young Adults in the Australasian Diabetes Data Network

Diabetic ketoacidosis (DKA) is a life-threatening complication that occurs frequently at the onset of type 1 diabetes in children and adolescents, causing physical and psychological stress for the individual and their family and placing considerable burden on the health system. The rate of DKA at diagnosis of type 1 diabetes in children remains high, with reported rates of 19–44% in Europe, North America, Australia, and New Zealand (1).

In addition to the acute metabolic derangement of DKA at diagnosis, long-term effects are described. DKA is associated with later recurrent DKA episodes (2) and with neurocognitive deficits in executive function and speed of information processing (3,4) that may be accompanied by distinct morphological and functional changes on MRI (5). Evidence for the relationship between DKA at diagnosis and long-term glycated hemoglobin A_1c (HbA_1c) is less conclusive. While some studies are limited by small numbers and short periods of follow-up (6,7), a Danish study of ∼3,000 children followed for a median of 6 years reported higher mean HbA_1c in association with moderate or severe DKA at type 1 diabetes onset (8). In Colorado (U.S.), a study of similar size reported a stepwise and sustained relationship between DKA severity at onset and HbA_1c over a median 7 years of follow-up (9), and a further U.S. study showed a trajectory of worsening glycemic control over time after DKA at onset (10). By contrast HbA_1c at 3 years in a large international study was more closely related to HbA_1c than to DKA at onset (11), and a recent population-based study of ∼2,100 Western Australian children followed for a median of 5 years found no consistent relationship between moderate/severe DKA at diagnosis and long-term HbA_1c (12).

Improved glycemic control in association with increasing use of continuous glucose monitoring (CGM) was recently reported by the Australasian (Australia and New Zealand) Diabetes Data Network (ADDN) (13). Other studies report improved glycemic control associated with the use of continuous subcutaneous insulin infusion (CSII) (14,15), particularly in those with a high baseline HbA_1c (16). Modern intensive management with use of more sophisticated technology may therefore potentially ameliorate the adverse effects on glycemic control associated with DKA (8).

We aimed, firstly, to characterize the association of DKA and its severity at onset of type 1 diabetes with long-term HbA_1c. Secondly, we aimed to identify other independent determinants of HbA_1c in a large cohort of children, adolescents, and young adults diagnosed with type 1 diabetes at or before 30 years of age and followed in ADDN from 2000 to 2020.

ADDN, established in 2012, includes deidentified, prospectively collected patient data from participating pediatric and adult centers across Australia and New Zealand. Data are collected using a common data dictionary and transferred every 6 months from the clinical databases or electronic medical record systems of participating centers to a web-based server hosted by the University of Melbourne (17). Data for this study were downloaded from ADDN in October 2021 and included visits up until 31 December 2020.

Participants attended type 1 diabetes multidisciplinary clinics in 14 pediatric (11 in Australia; 3 in New Zealand) and 9 adult (all in Australia) tertiary care diabetes centers. Type 1 diabetes was diagnosed according to International Society for Pediatric and Adolescent Diabetes/American Diabetes Association criteria, and the date of the first insulin injection was recorded as the date of diagnosis.

Inclusion criteria were 1) age ≤30 years at diagnosis of type 1 diabetes; 2) diagnosed with type 1 diabetes between 1 January 2000 and 31 December 2019, 3) type 1 diabetes duration ≥1 year at the last visit; and 4) at least one HbA_1c measurement (excluding HbA_1c at diagnosis, which was defined as within 1 week of diagnosis).

Exclusion criteria were 1) other forms of diabetes, including type 2 diabetes, maturity onset diabetes of the young, and cystic fibrosis-related diabetes; and 2) missing values for coding of the presence or absence of DKA at diagnosis.

Data collected in ADDN for each participant included DKA at diagnosis (yes/no), pH at diagnosis, bicarbonate at diagnosis, date of birth, sex, date of diagnosis, indigenous status, main language spoken, country of birth, family history of type 1 diabetes, residential postcode at the time of diagnosis, and CGM start date.

DKA at diagnosis was defined as the presence of hyperglycemia, ketonuria or ketonemia, and pH <7.3 or bicarbonate <15 mmol/L. DKA severity at diagnosis was categorized according to standard criteria as mild (7.2 ≤pH <7.3 or 10 mmol/L ≤bicarbonate <15 mmol/L), moderate (7.1 ≤pH <7.2 or 5 mmol/L ≤bicarbonate <10 mmol/L) or severe (pH <7.1 or bicarbonate <5 mmol/L) (18).

Data from the Australian Bureau of Statistics 2016 census provided the level of socioeconomic disadvantage at the postcode level at that time in Australian participants (19). Index of relative socioeconomic disadvantage (IRSD) was determined from the residential postcode at diagnosis and used as a surrogate indicator of the area-level socioeconomic status. A low IRSD score indicated a high proportion of relatively disadvantaged people in the corresponding postal area. IRSD was categorized into quintiles (Table 1). New Zealand has a different measure of socioeconomic status, the New Zealand index of deprivation, which cannot be combined in analysis with the IRSD.

Australia is divided into classes of remoteness based on relative access to health and other services (20). The residential postcode at diagnosis was used to determine this measure.

At each clinic visit, height was measured using a Harpenden stadiometer and weight using a floor scale, and these measurements were used to calculate BMI for adults >18 years of age. For children aged 2–18 years, BMI SD or z scores, measures of relative weight adjusted for child age and sex, were calculated from height and weight using the Centers for Disease Control and Prevention 2000 reference scale (21). The insulin delivery system (CSII, multiple daily injection [MDI] or twice-daily insulin injections [BD]) was recorded at each visit. HbA_1c was measured at diagnosis and at each visit using standardized techniques, most commonly a Vantage analyzer (Siemens Diagnostics, Camberley, U.K.) or a Variant analyzer (Bio-Rad Laboratories, Hercules, CA). All laboratories participated in an ongoing quality assurance program for HbA_1c with accreditation according to international standard International Organization for Standardization (ISO) 15189 Medical laboratories, which mandates that all analytes in a laboratory's test menu be subject to the Royal College of Pathologists of Australasia (Australia and New Zealand) Quality Assurance Programs (22).

The ANOVA model was used to compare mean values of covariates by DKA status. The Kruskal-Wallis test was used instead when data were skewed. For categorical covariates, the χ² test was used to compare across DKA categories. A linear mixed-effects model was used to study the relationship between HbA_1c (outcome) and the other variables. Random intercept terms were specified for patient ID and the treating site as well as a random slope for years from diagnosis. The main explanatory variable in the final model was severity of DKA. The effect of moderate DKA was not significantly different from mild DKA (P = 0.22), and the two categories were combined, giving three groups for DKA severity: no DKA, mild or moderate DKA, and severe DKA. In order to compare the slope between the three groups, an interaction term between DKA severity and duration of diabetes was specified in the main model. We also evaluated the interaction effect between DKA severity and 1) insulin regimen and 2) country of the treating site. Years from diagnosis was centered to aid interpretation of the coefficient, and a quadratic term was included to capture the nonlinear trend. Potential confounders (see below) were selected and included using the likelihood ratio test. Starting from the most significant variable identified in the univariate model, we sequentially included the next most significant covariate and evaluated whether there was an improvement in model fit.

Potential confounders were age at visit, age at diagnosis, average number of visits per year, indigenous status, insulin delivery method (CSII, MDI, BD), English as the main language spoken at home, Australian/New Zealand born, CGM ever used, treating ADDN center (included as random intercept term), and BMI z score (grouped into quartiles). Missing data were kept as a separate category in the model. For the final multivariate model, sensitivity analyses, including IRSD and remoteness and HbA_1c at diagnosis, were undertaken. The margins command was used to plot the marginal mean values of HbA_1c against duration of diabetes, stratified by DKA severity, with other covariate values fixed at their average. Data analysis was performed in Stata 17 software (StataCorp, College Station, TX), and level of significance was set at 5%.

Ethics approval was obtained through the Human Research Ethics Committee for each of the participating centers in Australia and in New Zealand. Informed written consent was obtained from parents of children <18 years of age and when required by the Human Research Ethics Committee, assent was obtained from children aged 10–17.9 years. Nonidentified data from adults who transitioned their care to adult centers aged ≥18 years was collected under a waiver.

Of the 11,530 individuals diagnosed with type 1 diabetes by age 30 years between 1 January 2000 and 31 December 2019 and followed for at least 12 months within ADDN, 7,961 participants fulfilled all eligibility criteria and were included in the analysis (Fig. 1).

Demographic and clinical characteristics of the 7,961 participants are reported in Table 1. They were followed for a median (interquartile range [IQR]) of 5.6 (3.2, 9.4) years. Of these 7,961 participants, 2,647 (33.2%) had DKA at diagnosis. Participants with DKA at diagnosis were younger, more likely to be female, and had a slightly shorter duration of type 1 diabetes than those without DKA (Table 1). They also had a higher HbA_1c at diagnosis than those without DKA and were less likely to subsequently use CGM. Participants diagnosed with DKA in Australia were more likely to reside in areas of greater socioeconomic disadvantage (IRSD quintiles 1 and 2) than those without DKA (Table 1). Among those presenting with DKA, 775 of 2,647 (29.3%) had severe DKA.

ADDN participants who were excluded due to missing data for presence or absence of DKA at diagnosis (n = 2,778) (Fig. 1) were similar to the study cohort in sex distribution but were older (mean age 11.3 [SD 6.8] years) with a longer duration of diabetes (median 10.2 [IQR 6.4, 13.7] years) and they were less likely to have used CGM (15%) or to be using CSII at the time of their last visit (37%).

On multivariate analysis, DKA at diagnosis was associated with a mean HbA_1c that was 0.11% higher (95% CI 0.05, 0.16; [1.2 mmol/mol, 95% CI 0.6, 1.8]; P < 0.001) over the duration of follow-up than for participants without DKA at diagnosis. Inclusion of DKA severity at diagnosis in the model revealed that this increase in HbA_1c was entirely attributable to the group with severe DKA at diagnosis (Table 2). Compared with participants with no DKA, those with severe DKA at diagnosis had a mean HbA_1c that was 0.23% higher (95% CI 0.13, 0.33; [2.5 mmol/mol, 95% CI 1.4, 3.6]; P < 0.001) over the duration of follow-up, whereas those with mild or moderate DKA had no significant difference in HbA_1c (0.04%, 95% CI −0.03, 0.11; [0.4 mmol/mol, 95% CI: −0.4, 1.2]; P = 0.29). Unadjusted estimates of HbA_1c at 60 days postdiagnosis were no DKA: 7.9% (86.4 mmol/mol); mild/moderate DKA: 7.9% (86.4 mmol/mol); and severe DKA: 8.1% (88.5 mmol/mol).

There was no interaction between DKA severity and duration of diabetes, so that predicted HbA_1c levels over time followed a similar trajectory for each DKA group (Table 2). As shown in Fig. 2, participants presenting with severe DKA at diagnosis had a marginally higher predicted mean HbA_1c at 2 years postdiagnosis than the groups with mild/moderate and no DKA, and this difference remained unchanged for the next 18 years. Addition of HbA_1c at diagnosis, IRSD, and remoteness to the multivariate model did not alter the results.

Indigenous status, language other than English as the main language at home, MDI regimen rather than CSII, no CGM use, longer duration of diabetes, older age at diagnosis, higher number of visits per annum, and lower BMI z score were all independently associated with higher HbA_1c over time (Table 2).

Use of CSII was associated with a mean HbA_1c that was 0.28% lower over time (95% CI −0.31, −0.25; [−3.1 mmol/mol, 95% CI −3.4, −2.8]; P < 0.001) compared with MDI, and use of CGM with a mean HbA_1c that was 0.13% lower over time (95% CI −0.19, −0.07; [−1.4 mmol/mol, 95% CI −2.0, −0.8]; P < 0.001) compared with no use of CGM (Table 2). An interaction between insulin regimen and DKA severity moderated the association between DKA severity and overall HbA_1c, so that CSII interacted with severe DKA to lower the predicted HbA_1c over time (Table 2). There was no overall significant interaction effect between country of the treating site (Australia or New Zealand) and DKA severity on HbA_1c (P = 0.86).

IRSD, but not remoteness to access to health services, was also independently associated with HbA_1c over time. Increasing disadvantage was associated with a progressively higher overall HbA_1c; those participants residing in postcodes in the most disadvantaged quintile had an overall HbA_1c that was 0.43% higher (95% CI 0.34, 0.52; [4.7 mmol/mol, 95% CI 3.4, 5.6]; P < 0.001) than those residing in the least disadvantaged quintile.

We report that presentation with severe DKA at diagnosis of type 1 diabetes was associated with marginally higher HbA_1c over a median follow-up duration of 5.6 years in 7,961 children and young adults in Australia and New Zealand. Moderate or mild DKA at diagnosis had no effect on HbA_1c. Other factors that were independently associated with more marked increases in HbA_1c included older age at diagnosis, indigenous status in both countries, English not being the main language spoken at home, and residing in postcodes of most socioeconomic disadvantage. Socioeconomic disadvantage was also associated with presentation with DKA at diagnosis. Importantly, use of CSII was associated with more favorable glycemic control, and there was an interaction between insulin regimen and DKA severity so that CSII use reduced HbA_1c over time in the group with severe DKA at diagnosis.

The 0.2% (2.5 mmol/mol) higher HbA_1c associated with severe DKA at diagnosis, while independent and of statistical significance, is of doubtful clinical significance. American Diabetes Association guidelines have suggested that an increment in HbA_1c of ∼0.5% (5.5 mmol/mol) is needed for clinically meaningful change (23) and that higher increments closer to a 1% (10.9 mmol/mol) increase in HbA_1c have been associated with increased cardiovascular risk (24).

Our results differ from the U.S. studies (9,10) in some important aspects. Our participant number was more than double the size of the larger of these two studies (9) in which there was a stepwise progression of less favorable HbA_1c levels from no to mild/moderate to severe DKA at diagnosis of type 1 diabetes in children <18 years of age. Strikingly, the effect of severe DKA on HbA_1c was sevenfold greater than our findings. The SEARCH for Diabetes in Youth (SEARCH) cohort study (10) showed a worsening trajectory of HbA_1c levels after DKA at onset, but we showed no such trajectory after any category of DKA. The age range of our participants was wider, but mean age at diagnosis was comparable. Other determinants of glycemic control reported in these studies, namely, use of CSII and older age at diagnosis, were of similar effect size to those reported in the current study. Of note, HbA_1c levels in their cohort over time were only slightly higher than those in our cohort, but their participants used CSII and likely CGM (this latter was not documented) less frequently, as would be expected during a study time period of some 8 years earlier. It seems plausible that the increasing use, sophistication, and benefits of diabetes technology may contribute to our different findings.

Our results are more consistent with the Danish study (8), which showed a similar effect size of moderate/severe DKA on HbA_1c in ∼3,000 children with type 1 diabetes, but a greater increment (0.5% [5.5 mmol/mol]) when the insulin dose-adjusted HbA_1c was calculated as a surrogate for β-cell function. As in our study, the insulin regimen moderated the association between DKA and HbA_1c, with the Danish study reporting that during the first year of CSII, the effect of moderate/severe DKA on HbA_1c disappeared. Indeed, in a very small study of ∼90 individuals with type 1 diabetes who had high use of CSII (90%), there was no association between DKA at diagnosis and HbA_1c 5 years later (25).

The recent Western Australian study (12) followed a group of ∼2,000 children with type 1 diabetes and also found no consistent or clinically relevant difference in HbA_1c over time according to DKA status at diagnosis. These and our results highlight questions about the broader range of factors—sociodemographic, psychological, and physiological—that may relate to both the likelihood of developing DKA at diagnosis and long-term glycemic outcomes in children and young adults with type 1 diabetes. The association that we found between IRSD quintile, as a measure of socioeconomic status, and both DKA at diagnosis and HbA_1c, supports this concept, and further research in this area to better tease out the critical components of socioeconomic status is warranted. It remains important to continue efforts to prevent DKA at diagnosis in those belonging to all socioeconomic groups.

Indigenous peoples with type 1 diabetes in both Australia (Aboriginal) and New Zealand (Māori) had strikingly higher HbA_1c levels. This and their increased cardiovascular risk that we have recently documented (26) are of considerable concern. The weak relationship between more visits and higher overall HbA_1c is likely explained by the clinical practice to increase the frequency of follow-up in patients with less favorable glycemic control. Despite the increasing emphasis on tailoring insulin doses to appetite and food intake, a relationship between increasing BMI z score quartile and lower HbA_1c was still detected.

Strengths of our study first include its relatively large size compared with other reports. Second, approximately half of the participants used CGM and a little under half used CSII, so that we could investigate the impact of DKA in the presence of contemporary diabetes technology. Third, Australia and New Zealand provide universal access to health care across intensive care units, emergency departments, and diabetes centers, so that differences in standards of care and health care access are minimized. ADDN cares for ∼40% of children and young adults with type 1 diabetes in Australia and New Zealand, and centers deliver both regional and metropolitan clinics by way of well-established outreach services, so our data are representative of the ethnicity, urban, and regional demographic, and health care systems in both countries.

Our study also has several important limitations. We did not have C-peptide data or sufficiently accurate insulin dose data in order to generate insulin dose-adjusted HbA_1c, so we could not explore whether the observed effect of severe DKA on HbA_1c is explained by less endogenous insulin production over follow-up. HbA_1c at diagnosis was marginally but significantly higher in those with DKA, and while its inclusion in the multivariate analysis did not alter the results, this variable in particular had a high proportion of missing values, as it is not routinely measured at diagnosis of type 1 diabetes in Australia and New Zealand. IRSD and remoteness data were available in Australian but not New Zealand participants; the Australian participants represented 87% of the total cohort. The 2,778 participants who were excluded on account of missing data for DKA at diagnosis differed from the study cohort in age, duration, and CGM and CSII use, all of which can influence HbA_1c levels. This limitation reflected lower rates of completeness in data collected for ADDN adults.

In conclusion, severe DKA at diagnosis was associated with a marginally higher HbA_1c, likely of limited clinical relevance, over follow-up in a large prospective cohort of Australian and New Zealand children and young adults with type 1 diabetes. Use of CSII was associated with improved glycemic control, and its use reduced the predicted HbA_1c in participants with severe DKA at diagnosis, suggesting that increasing benefits of modern diabetes technology may account at least in part for the differences in findings between this study and earlier studies. Independent of these effects, markedly less favorable control was detected in Australian and New Zealand indigenous peoples and in those residing in areas of greater socioeconomic disadvantage. These findings further emphasize the need for equity of access to health services and modern diabetes technology for all patients with type 1 diabetes.

Acknowledgments. This research was conducted as part of the ADDN. The authors are grateful to JDRF Australia, the Australian Research Council, and to the children and young people with diabetes and their families who provided the data. The authors acknowledge the work of the Melbourne eResearch Group at the University of Melbourne.

Funding. This research was supported by JDRF Australia (4-SRA-2016-169-M-B), the recipient of the Australian Research Council Special Research Initiative in Type 1 Juvenile Diabetes, and the Channel 7 Research Foundation of South Australia.

The study funders had no role in study design, data collection, analysis, interpretation, or the manuscript preparation.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. H.F.C., A.E., P.G.C., E.A.D., C.J., and J.J.C. conceived and designed the study. H.F.C., A.E., P.G.C., E.A.D., C.J., and J.J.C. wrote the manuscript. H.F.C., P.G.C., E.A.D., C.J., K.A., P.B., M.d.B., K.-T.K., P.G.F., D.J.H.-W., S.J., B.R.K., K.N., A.S.P.V., B.J.W., A.Z., M.E.C., and J.J.C. collected data required for the study and oversaw implementation of ethical practice and the ADDN protocol at the clinical sites. A.E. performed the statistical analyses. M.C., M.T.M., and R.S. supervised the storage, maintenance, and extraction of all data for the study. All authors provided critical revision of the manuscript and approved the final version. A.E. and J.J.C. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the International Society for Pediatric and Adolescent Diabetes (ISPAD) 48th Annual Conference, Abu Dhabi, United Arab Emirates, 13–16 October 2022.