Rising Rates and Widening Socioeconomic Disparities in Diabetic Ketoacidosis in Type 1 Diabetes in Scotland: A Nationwide Retrospective Cohort Observational Study

Diabetic ketoacidosis (DKA) is a significant contributor to mortality in type 1 diabetes, and we recently showed that this acute complication is second only to cardiovascular disease as the most frequent cause of death <50 years of age in these individuals (1). DKA events are often precipitated by infection, acute comorbidities, insulin omission, or substance use disorders (2–4) and occur as a consequence of insulin deficiency in the context of these precipitating factors. Further established risk factors associated with higher rates of DKA include insulin pump failure, higher HbA_1c, disordered eating, lower socioeconomic status, and depression (5–10).

DKA is a preventable cause of death, and reducing DKA mortality would lower the excess mortality associated with type 1 diabetes (1,11). Recent studies have suggested that DKA admission rates and the proportion of people with DKA at diagnosis of type 1 diabetes are increasing (12,13). In addition to this, we recently reported that DKA mortality rates have not significantly improved in Scotland in recent years for people <50 years of age (11). Improving early diagnosis of type 1 diabetes, identification of DKA, and the reduction of the impact of deprivation on diabetes care and outcomes are key priorities in the Scottish Government’s Diabetes Improvement Plan (14).

Accurate quantification of DKA event rates and identification of associated risk factors are required to inform future DKA prevention strategies and health care delivery models. Few studies have quantified DKA event rates over time in cohorts representative of large populations (3,5), and even fewer have attempted to estimate the association of socioeconomic status with these rates. Consequently, identifying the characteristics of individuals at elevated risk of DKA admission or mortality remains a priority. This current study aimed to quantify trends in DKA events in a comprehensive nationwide cohort of individuals with type 1 diabetes over a 14-year period using electronic health care records, and to identify the association of socioeconomic status with DKA event rates and their associated time trends.

Scottish Care Information–Diabetes (SCI-Diabetes) is the national disease management system of individuals diagnosed with any form of diabetes in Scotland, achieving >99% coverage. This register can be linked to hospital admissions and death data provided by the Information Services Division of National Health Service Scotland and National Records of Scotland. We identified a cohort of all people with a clinical diagnosis of type 1 diabetes who were alive at any point between 1 January 2004 and 31 December 2018. Clinician-assigned diabetes type was accepted unless there was contradictory prescription history data. Cohort members contributed data from 1 January 2004 or the date of their diagnosis with type 1 diabetes, whichever was later, up to date of death during the study period, leaving the jurisdiction, or the end of the study period. Periods of observability were defined using routine observation and prescription data. The number of hospital admissions for DKA for each individual was collected from Scottish Morbidity Record (SMR-01) data, and DKA mortality data were taken from the Medical Certificate of Cause of Death (MCCD) for each individual who died during the study period, as provided by National Records of Scotland. The presence of an ICD-10 code for DKA (i.e., E10.1, E11.1, E12.1, E13.1, E14.1) anywhere on an admissions record or MCCD was considered to be a hospital admission or a death with DKA, respectively. We combined DKA hospital admissions and DKA deaths into a composite count of DKA events experienced for each individual. This count included both events at DKA diabetes diagnosis and any subsequent events.

Directly standardized rates of DKA were calculated using the 2011 population structure of people with type 1 diabetes in Scotland as the standard. Bayesian Poisson generalized linear regression analyses were performed to estimate the association of a number of covariates with DKA event rates. Longitudinal electronic health care record data for each individual were time sliced to produce a survival table with each time slice being a single calendar year in length. To summarize the results from Bayesian analysis, we report median posterior estimates, 95% credibility intervals (95% CrIs), and a measure that is equivalent to a P value assessed at the posterior median.

To investigate the association of covariates with rates of recurrent DKA, we performed a Bayesian Poisson regression analysis in which the outcome was a binary measure of whether or not an individual experienced more than one DKA event in a given time slice. To investigate DKA case-fatality, we used a Bayesian logistic regression model in which each DKA event was an observation and the outcome was the fatality status of the DKA event. If an individual died during a hospital admission with an ICD-10 code for DKA mentioned at any point on their MCCD, we determined this to be an instance of case-fatality. Similarly, any out-of-hospital death for which DKA was identified on the MCCD was also considered to be an incidence of case-fatality.

We explored several factors previously reported as being associated with increased risk of DKA (6,9,15,16). To test the basic association of social deprivation with each DKA outcome, we performed a set of minimally adjusted analyses in which the adjustment covariates were age, sex, quintile of social deprivation, and diabetes duration. Area-based social deprivation was measured using the Scottish Index of Multiple Deprivation (SIMD), with each individual assigned to a quintile of SIMD at baseline. SIMD is constructed from several domains of area-based deprivation, including income, education, housing, and crime (17). Consequently, SIMD provides a high-level overview of social deprivation but cannot be used to assess associations between individual aspects of deprivation and health outcomes. Further adjustment was performed to account for potential risk factors that had been observed to vary with SIMD quintile (18) (Supplementary Fig. 1). Subsequent fully adjusted models included annualized median HbA_1c in previously used categories (18), SIMD quintile, use of continuous subcutaneous insulin infusion (CSII), completion of any level 3 (i.e., delivered in a group setting) structured education course, the prescription of methadone (Anatomical Therapeutic Chemical classification N07BC02), and the prescription of an antidepressant (Anatomical Therapeutic Chemical classification N06A, N06CA). All covariates were time updated except for quintile of social deprivation, which was only available at baseline. The risk factors for DKA events caused by delayed diagnosis of diabetes may differ from those for DKA events after diagnosis. Consequently, we performed DKA event rate regression analyses excluding DKA at diagnosis as sensitivity analysis. We also performed DKA case-fatality regression sensitivity analyses restricted to consider only the deaths in which DKA was assigned as the underlying cause on the associated MCCD.

Age-band–stratified subgroup analyses were also performed to identify age-specific associations between covariates and DKA outcomes, with broad age-bands of <30, 30–59, and >59 years used for this purpose. We also calculated the annual percentage of individuals presenting with DKA at diagnosis of type 1 diabetes over the study period and used a univariate Bayesian linear model to investigate the effect of calendar time on the annual percentage change in this outcome. Further details of the statistical methods outlined here are presented in the Supplementary Methods.

Analyzed data were provided deidentified, with approval from the Public Benefit and Privacy Panel (PBPP refs. 1617-0147), originally set up under Privacy Advisory Committee (PAC) 33/11, with approval from the Scotland A Research Ethics Committee (ref. 11/AL/0225). National Health Service data governance rules do not permit us to secondarily share the analyzed data directly. However, bona fide researchers can apply to the Scottish Public Benefits Protection Committee for access to these data.

We identified 37,939 individuals with type 1 diabetes alive in Scotland at any point between 1 January 2004 and 31 December 2018. These individuals contributed 393,223 person-years to the study, and 96.84% of all possible person-time was observable. Over the study period, there were 30,427 hospital admissions for DKA (which occurred in 10,397 individuals), of which 1,490 admissions at the time of diagnosis with type 1 diabetes. There were 472 deaths in which DKA was present anywhere on an MCCD and 413 where DKA was the underlying cause of death. These event counts correspond to a crude mortality rate of 120 and a crude DKA event rate of 7,858 per 100,000 person-years. During the study period, 72.6% of the cohort never had a hospital admission with DKA, 15.1% had one admission, and 12.3% had multiple admissions. Of all deaths for which DKA was a cause, 83.47% occurred outside of a hospital admission. A detailed characterization of the cohort at the time of entry is presented in Supplementary Table 1, grouped by number of DKA events experienced over the study period.

There was a gradual increase in the standardized rate of DKA events across the study period, starting at 6,182 events per 100,000 person-years in 2004 and finishing at 8,261 per 100,000 person-years in 2018 (Supplementary Fig. 2). This pattern was seen in both sexes and across SIMD quintiles (Fig. 1), with the event rate in the most deprived quintile increasing the most over the study period.

A nonlinear relationship between the standardized event rate and age was observed (Fig. 2). Across all age-bands, the general trend over the study period has been toward increasing standardized DKA event rates. The exception to this pattern was the reduction in event rate in the final years of the study period for 10- to 19-year-olds. The association of calendar time with DKA event rates varied across broad age-bands (Supplementary Table 2), with DKA event rates increasing the least in individuals <30 years of age (incidence rate ratio [IRR] 1.044 [95% CrI 1.039–1.050]) and increasing most rapidly in individuals aged ≥60 (IRR 1.096 [1.081–1.110]).

Rates of recurrent DKA increased over the study period (Supplementary Table 3). The rate of DKA at diagnosis was 13.35%, and this increased over the study period from 8.93% in 2004 to 17.68% in 2018 (β = 0.57 [95% CrI 0.39–0.74]) (Supplementary Fig. 3).

Directly standardized DKA mortality rates were 95 deaths per 100,000 person-years in the first year of the study period and 204 deaths per 100,000 person-years in the final year. The annual directly standardized mortality rate is presented in Fig. 3, and the age-band–specific directly standardized mortality rate is presented in Supplementary Fig. 4. The case-fatality rate over the entire study period was 1.53% (odds ratio for year 1.03 [95% CrI 1–1.05]). In age-band–stratified analyses, evidence of an annual increase in case fatality was identified for individuals aged ≥30 only.

Age had a nonlinear effect on DKA event rates (Table 1) and DKA recurrence rates (Supplementary Table 3), with rates elevated in children, teenagers, and younger adults and in individuals aged ≥70 years, as seen in the IRR for each age-band. Conversely, the effect of age on case fatality was linear (Supplementary Table 4).

Male sex was associated with elevated DKA case fatality, but a lower DKA event rate and recurrence rate (IRR 0.814 [95% CrI 0.776–0.855]. Age-band–stratified analyses showed that the association of male sex with elevated DKA case fatality was not present in the ≥60 age-band (Supplementary Table 5).

Minimally adjusted analyses showed a clear association between level of social deprivation and the DKA event and recurrence rate, although the association of increasing levels of social deprivation with case fatality was less consistent (Supplementary Table 6). The association of social deprivation with each DKA outcome in minimally adjusted analyses was maintained after adjustment for further covariates. Higher HbA_1c was associated with elevated rates of DKA events and DKA recurrence, but no association between the highest HbA_1c category and increased DKA case fatality was observed. Age-band–stratified analyses did not support an association between HbA_1c category and case fatality (Supplementary Table 5). Prescription for an antidepressant and prescription for methadone were both associated with elevated rates of each analyzed DKA outcome. Age-band–stratified analyses supported the association of elevated DKA case fatality with antidepressant prescription in all age-bands apart from the 0–29 age-band.

The use of CSII therapy or completion of a structured education course were both associated with reduced rates of DKA events and reduced DKA recurrence, but an association between these covariates and DKA case fatality was not supported in these analyses. Age-band–stratified analyses did not support the association of CSII with reduced DKA event rates in the 30–59 age-band. Similarly, age-band–stratified analyses did not support the association of CSII with reduced rates of recurrent DKA after the age of 29 (Supplementary Table 7), and they did not support the association of completion of structured education with reduced rates of DKA below the age of 30.

Sensitivity analyses showed that apart from lowering the event rate in the youngest age-band, excluding DKA at diagnosis events from the DKA event rate regression analysis did not meaningfully alter the association between each covariate and DKA event rates (Supplementary Table 8). Excluding deaths for which DKA appeared on the MCCD, but not as the underlying cause of death, resulted in no association between HbA_1c category and case fatality being established (Supplementary Table 9).

Through the analysis of a national cohort of individuals with type 1 diabetes, we have identified patterns in rates of DKA and estimated the association of several risk factors with DKA outcomes. We identified a gradual increase in DKA event rates, mortality rates, and DKA recurrence rates over the study period. Increases in DKA event rates over time were identified across all age-bands, excluding individuals aged 10–19, with rates for individuals aged 10–19 now similar to those of aged 20–29 years. The largest relative increase in event rates was identified in individuals aged ≥60. Age-band–stratified regression analyses showed that the greatest elevation in event rates over the study period were for individuals aged ≥60 years. These results suggest that global increases in DKA admission rates could be partially driven by increases in DKA rates in older individuals. The association of age with DKA case fatality appears to be similar across all age-bands, suggesting that this association is linear and that DKA events are more likely to be fatal for older individuals. The observation that older individuals are experiencing increasing rates of DKA, coupled with DKA events being more likely to result in death for this age-group, is concerning.

Elevated HbA_1c was associated with increased DKA event rates, and prescription for methadone and antidepressants were both associated with considerably elevated DKA event and case-fatality rates. For some covariates, we found differences in the estimated effect on DKA event rates and case-fatality rates, suggesting that the factors that elevate the risk of the occurrence of a DKA event may be different from the factors that elevate the risk of an event being fatal. Completion of a structured education course and the use of CSII were associated with reduced DKA event rates, but there was little evidence that completion of a structured education course or CSII are associated with reduced case fatality. We found that females experienced DKA events at a higher rate than males, but a given DKA event was more likely to be fatal for males. A potential contributing factor in higher DKA rates for females in Scotland is elevated HbA_1c, which has been observed in this subgroup in a previous study (18).

Associations between social deprivation and mortality or morbidity in type 1 diabetes have previously been identified in Scotland (18–20). The association of social deprivation with DKA outcomes is supported by the results presented here. Adjustment for several covariates suggests that this association occurs independently of HbA_1c, which is higher on average in more socially deprived areas (18), and CSII or completion of structured education, which are less prevalent in more deprived areas (Supplementary Fig. 1). This suggests that factors other than those included in these models contribute to the elevated DKA rates observed in more socially deprived areas.

Increases in DKA admission rates have been identified in England, Wales, Australia and New Zealand, Denmark, and the U.S. (21–25), although reductions in DKA admission rates have been observed in Italy and Taiwan (26,27). A lower threshold for hospital admission, prevalence of basal-bolus insulin regimens, and alterations to DKA diagnostic criteria have all been proposed as potential contributors to increasing DKA rates (23,25). The reduction in DKA event rates for 10- to 19-year-olds in the final years of our study period represents a deviation from the general trend of increasing DKA rates. While the cause of this reduction has not been established by these analyses, across a similar time period, these younger individuals experienced the greatest improvement in HbA_1c and the widest initiation of CSII among all individuals with type 1 diabetes in Scotland (18).

We found that the rate at which individuals presented with DKA at the time of diagnosis increased over the study period, but the absolute rate did appear to be comparatively low. A recent study found that the prevalence of DKA at diagnosis in children with type 1 diabetes ranged from 18.4 to 53.2% across a number of different nations (13), whereas the annual prevalence reported here, among both children and adults, never exceeded 17.68%.

Few published studies have focused on DKA mortality rates in large cohorts (3), making it difficult to compare the mortality trends identified here to other populations. Another study using the Scottish diabetes register found that DKA mortality rates in individuals aged <50 had not significantly reduced in the period 2004–2017 (11). The results presented here demonstrate that this trend is also observed when this population is extended to include individuals aged ≥50.

One previous study found that CSII therapy was not associated with a significant decrease in DKA admission rates (6). Conversely, the results presented here and by Jeyam et al. (28) suggest that CSII therapy is associated with reduced DKA event rates. Other studies have demonstrated that structured education courses are associated with lower rates of DKA admission (15,16), and our results provide further observational evidence to support this association. Similarly, our results support previously observed associations between recurrent DKA and antidepressant use, social deprivation, or higher HbA_1c (9).

A strength of these analyses was the use of a national-level cohort with almost complete coverage of all individuals with type 1 diabetes. The follow-up of these individuals was extensive, with only a small percentage of possible person-time being unobserved. One potential weakness in the construction of outcomes in these analyses was the lack of confirmation of DKA through biochemical tests. However, an evaluation of DKA discharge coding for a Scottish cohort (2002–2009) found that the sensitivity of SMR data is adequate for epidemiological study (29). It is possible that changes in DKA diagnosing and discharge coding practices over the study period have contributed to increasing rates of DKA, although a simultaneous increase in DKA mortality rate suggests that any such change is unlikely to be driving the observed DKA event rate increases. Generalization of the results presented here to other countries is likely to be limited by the fact that Scotland has universal health care that is free at the point of use, with free access to prescribed medications, including insulin. There is also potential allocation bias associated with CSII referral and initiation in Scotland (30), meaning that a causal relationship between CSII and DKA outcomes cannot be established in these analyses.

While we included prescription for antidepressants and methadone as covariates in these analyses, their accuracy as markers for depression and opioid dependence is limited by certain antidepressants being indicated for the treatment of neuropathy and methadone being indicated for the treatment of severe pain. A further potential weakness of using methadone prescription as a marker for opioid dependence is that this approach will not capture individuals with opioid dependence who did not receive a methadone prescription. Despite this potential weakness, we still observed a consistent association between methadone prescription and elevated DKA event rates and mortality.

The significant contribution of DKA to excess mortality means that prevention of this acute complication should remain a priority, particularly as DKA rates are rising. The association of social deprivation with DKA outcomes suggests that more work is required to reduce rates of DKA in people living in more deprived areas. Mair et al. (18) identified higher HbA_1c in people living in more socially deprived areas in Scotland, and here we demonstrate the consistent association of high HbA_1c with DKA event and recurrence rates. Similarly, while CSII therapy and structured education completion are associated with lower DKA event and recurrence rates, they have lower cumulative incidence among people living in more deprived areas (Supplementary Fig. 1). Consequently, improved glycemic control, more widespread use of CSII, and completion of structured education may improve DKA outcomes for people living in the most deprived areas. The association of social deprivation with the DKA event rate is still significant after adjustment for several confounders, and this suggests that socioeconomic factors not included in these models play a key role in DKA event rates for people living in more socially deprived areas. Future work must identify these factors and establish effective interventions to reduce health inequalities. Furthermore, as the prevalence of CSII increases, future work must assess rates of DKA as a consequence of CSII failure, particularly after initiation of treatment (31).

Drug-related deaths in the general population in Scotland have risen over the previous 25 years (32). While the association of methadone prescription with elevated DKA rates has been established here, methadone is only implicated in ∼44% of all drug-related deaths in Scotland, with 94% of all drug-related deaths involving multiple substances. Consequently, more work is required to investigate the broader effect of drug misuse on DKA admission and mortality in Scotland, particularly as DKA mortality rates have increased in the same age-groups in which drug-related deaths in the general population have increased the most.

More studies using electronic health care records are required to accurately quantify the incidence of DKA in other countries and to identify the underlying causes of the association of each risk factor with DKA outcomes (5,33). Of particular concern are the causes of the association of social deprivation with DKA and whether the higher rate of DKA in young women is influenced by the prevalence of disordered eating in this group (10,34).

Funding. This study was supported by funding from Diabetes UK (17/0005627). T.M.C. received the Diabetes UK Sir George Alberti Clinical Research Training Fellowship (18/0005786).

Duality of Interest. J.R.P. reports personal fees from Merck KGaA, Novo Nordisk, Boehringer Ingelheim, and Biocon and nonfinancial support from AstraZeneca. R.M. reports personal fees from Novo Nordisk and Sanofi. H.M.C. reports grants and personal fees from Eli Lilly and Novo Nordisk; grants from AstraZeneca LP, Regeneron, and Pfizer Inc.; and other from Novartis, Sanofi, and Roche Pharmaceuticals. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.E.O. performed all analyses. J.E.O., A.J., T.M.C., J.M., A.H., P.M.M., and H.M.C. created the first draft of the manuscript. J.E.O., P.M.M., and H.M.C. created the concept and design of the analysis. S.J.M. and L.A.K.B. constructed the analyzed data set. R.M., S.H.W., J.R.P., J.A.M., B.K., J.C., S.P., G.L., R.S.L., N.S., and F.W.G. contributed to data acquisition. All authors contributed to the writing and revision of the final manuscript. J.E.O. and H.M.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.

Scottish Diabetes Research Network Epidemiology Group: J. Chalmers (Diabetes Centre, Victoria Hospital, Kirkcaldy, U.K.), C. Fischbacher (Information Services Division, National Health Service National Services Scotland, Edinburgh, U.K.), B. Kennon (Queen Elizabeth University Hospital, Glasgow, U.K.), G. Leese (Ninewells, Hospital, Dundee, U.K.), R. Lindsay (British Heart Foundation Glasgow Cardiovascular Research Centre, University of Glasgow, Glasgow, U.K.), J. McKnight (Western General Hospital, National Health Service, Edinburgh, U.K.), J. Petrie and N. Sattar (Institute of Cardiovascular & Medical Sciences, University of Glasgow, Glasgow, U.K.), R. McCrimmon (Divison of Molecular and Clinical Medicine, University of Dundee, Dundee, U.K.), S. Philip (Grampian Diabetes Research Unit, Diabetes Centre, Aberdeen Royal Infirmary, Aberdeen, U.K.), D. McAllister (Institute of Health & Wellbeing, University of Glasgow, Glasgow, U.K.), E. Pearson (Population Health and Genomics, School of Medicine, University of Dundee, Dundee, U.K.), and S. Wild (Usher Institute, University of Edinburgh, Edinburgh, U.K.).