Trends in Hospital Admission for Diabetic Ketoacidosis in Adults With Type 1 and Type 2 Diabetes in England, 1998–2013: A Retrospective Cohort Study

Hospital admission for diabetic ketoacidosis (DKA) is a potentially life-threatening but largely preventable complication in diabetes (1–3) and is associated with significant health care utilization and expenditure (4,5). Previous studies focused on describing DKA trends in children (6–9); however, >80% of patients with DKA are adults (1), among whom data for long-term trends are limited. The available studies in adults from different countries reported increased (10,11), stable (12), or slightly declined (13) rates of DKA admission to the hospital, emergency department, or intensive care unit. Other studies in adults found declined hospitalization rates for hyperglycemia as a whole (14,15). However, none of these studies assessed DKA trends by diabetes type. Type 2 diabetes mellitus (T2DM) is the predominant diabetes type in adults, whereas DKA risk is substantially higher in type 1 diabetes mellitus (T1DM) than in T2DM (1). Thus, direct application of DKA trends identified from all people with diabetes to each specific diabetes type is misleading. Using primary care data from the Clinical Practice Research Datalink (CPRD) and linked secondary care data from the Hospital Episode Statistics (HES), we aimed to determine DKA trends in incidence, length of hospital stay, 30-day all-cause readmission rate, and 30-day and 1-year all-cause mortality rates, from 1998 to 2013, in adults with T1DM or T2DM in England, respectively.

The CPRD, one of the largest electronic medical record databases in the world, collects anonymous longitudinal patient records from 684 general practices from England, Scotland, Wales, and Northern Ireland as of January 2015. Patients within the CPRD are broadly representative of the population in the U.K. in age, sex, and ethnicity (16). The CPRD contains rich primary care information including, but not limited to, basic demographics, prescribed medications, clinical diagnoses, and laboratory test results. The CPRD is linked to a secondary care database via a trusted third party, the HES, which houses patient-level information on all hospital admissions to National Health Service hospitals in England. The study population was from 398 of the 684 CPRD practices that agreed to be linked to the HES between 1 April 1997 and 31 March 2014. The linked CPRD-HES population is representative of the entire CPRD population (17). Hospital admissions for DKA were identified from the HES, and all other data were extracted from the CPRD. The study protocol (No. 15_259RA) was approved by the Independent Scientific Advisory Committee for Medicines & Healthcare Products Regulatory Agency in London, U.K. A minor modification was made to the approved protocol, which did not need an amendment submission. We included two additional variables, length of hospital stay and mortality, that measure quality of care (18,19).

The definition for T1DM and T2DM was previously published (20). Specifically, among individuals with one or more diabetes-related diagnosis code, we defined T1DM as having any one of the following criteria: 1) one or more T1DM codes and insulin use only; 2) one or more T1DM codes and insulin use only on the diagnosis date and noninsulin glucose-lowering drug (NIGLD), if any, was introduced 6 months later; 3) two or more insulin prescriptions only. T2DM was defined as having any one of the following criteria: 1) two or more T2DM codes and no T1DM code, regardless of drug use; 2) one or more T2DM codes and no T1DM code and NIGLD use only; 3) one or more T2DM codes and no T1DM code and being treated with both NIGLD and insulin, but NIGLD prescribed no later than insulin; 4) two or more classes of NIGLD; 5) two or more prescriptions of a noninsulin nonmetformin glucose-lowering drug only. NIGLDs included metformin, sulfonylureas, glinide, thiazolidinediones, inhibitors of dipeptidyl peptidase 4, inhibitors of sodium-glucose cotransporter 2, glucagon-like peptide 1 receptor agonists, and acarbose. Patients were excluded if they had nonresearch quality data determined by the CPRD team based on the prespecified data quality metrics or had a record of secondary diabetes, maturity onset diabetes of young, latent autoimmune diabetes in adults, or malnutrition-related diabetes.

The full-year HES data were available between 1998 and 2013. The follow-up started at the latest of the following dates: 1 January 1998, first diabetes visit, patient registration, up to standard date (16), or 18 years old. Follow-up ended at the earliest of the following dates: 31 December 2013, death, transfer out, or last data collection for the practice. DKA (E10.1, E11.1, E12.1, E13.1, and E14.1) listed as primary diagnosis for hospital admission during follow-up was identified. We included all DKA admissions for analyses. Because DKA could be the first presentation of diabetes (1,21), we also included DKA admissions with hospital admission date within 28 days before the first diabetes visit date documented in the CPRD.

All analyses were performed separately by diabetes type. The incidence rate of hospital admission for DKA was calculated by dividing the total number of DKA admissions by the total accumulated person-years with diabetes within each year between 1998 and 2013. We computed the age standardized incidence rate using the entire study sample as the reference population, according to these age categories in T1DM (18–24, 25–34, 35–49, ≥50 years) and T2DM (18–49, 50–64, ≥65 years). For T1DM, age standardized incidence rates were also given by sex (male, female), duration of recorded diabetes history (as an alternative for diabetes duration; <1, 1–9, 10–19, and ≥20 years), Charlson comorbidity score (1, 2, 3, and ≥4) (22), and region (North East, North West, Yorkshire and The Humber, East Midlands, West Midlands, East of England, South West, South Central, London, and South East Coast). For T2DM, age standardized incidence rates were given by sex, duration of recorded diabetes history (<1, 1–4, 5–9, and ≥10 years), Charlson comorbidity score (1, 2, 3, and ≥4), current use of glucose-lowering drugs (sulfonylureas with or without NIGLD, insulin only, insulin and NIGLD, and other), and region.

Trends in the incidence for hospital admission for DKA were estimated using joinpoint regression models (23). Annual percentage change and 95% CI were quantified. Length of hospital stay in days was calculated for adults with T1DM or T2DM and was then categorized into four categories (0–2 days, 3–6 days, ≥1 but <2 weeks, and ≥2 weeks). Also calculated were the 30-day all-cause readmission rate and 30-day and 1-year all-cause mortality rates. Trends in the length of hospital stay were determined by negative binomial regression models. Trends in the four categories of hospital stay were estimated using ordinal logistic regression models. Trends in the 30-day all-cause readmission rate and 30-day and 1-year all-cause mortality rates were examined using logistic regression models. All models included calendar year, age, sex, duration of recorded diabetes history, Charlson comorbidity score, and current use of glucose-lowering drugs (for T2DM only). Nonlinearity was assessed by adding a quadratic term for year. Differences between T1DM and T2DM (reference group) were also compared. Incidence rate ratios (IRRs) or odds ratios (ORs) and 95% CIs were presented. To account for DKA as a recurrent event, we applied robust error estimate for correlated data in all models implemented by using the PROC GENMOD procedure with the REPEATED statement (24,25) in SAS 9.4 software (SAS Institute). Statistical significance was indicated by a two-sided P value <0.05.

Sensitivity analyses were conducted to evaluate the effect of potential misclassification of diabetes type on DKA trends. Previous investigation revealed that the aforementioned criteria 2 and 3 for T1DM and criterion 3 for T2DM were mostly likely to misclassify diabetes type from clinical perspectives (20). A decent proportion of adults with T1DM (11.86%) and a small proportion of adults with T2DM (0.33%) may have been incorrectly classified. We reestimated DKA trends by removing these patients. We also evaluated whether DKA trends were materially influenced by any single year data.

Our primary goal was to estimate the burden of hospital admission for DKA over time in England, and thus, our primary analyses focused on all available DKA hospital admissions from each included patient. Secondary analyses were conducted to determine whether the observed DKA trends were driven by a group of patients who had multiple DKA hospital admissions. We reassessed the trends by 1) analyzing only the first DKA hospital admission from each patient during the entire study period and 2) analyzing only the first DKA hospital admission in each year from each patient. Characteristics of patients were also described according to the number of DKA hospital admissions during the entire study period (none, one, two, three or more for T1DM and none, one, and two or more for T2DM). Further, the length of hospital stay and 30-day and 1-year all-cause mortality were compared between patients with different numbers of DKA hospital admissions (one vs. two or three or more for T1DM and one vs. two or more for T2DM), using the aforementioned modeling approaches. In addition, among patients with at least one DKA hospital admission, we assessed the 30-day all-cause readmission rate by including only the first 30-day hospital readmission for DKA during the entire study period for each patient or the first 30-day readmission in each year for each patient.

Among 23,246 adults with T1DM, 4,346 hospital admissions for DKA were identified during 121,258 person-years of follow-up (Supplementary Table 1). The incidence of hospital admission for DKA was 35.84 per 1,000 person-years during the entire study period. Among 241,441 adults with T2DM, 1,148 DKA hospital admissions were found during 1,342,896 person-years of follow-up, with an incidence of 0.85 per 1,000 person-years (Supplementary Table 2). Every one in five DKA hospital admissions was attributed to T2DM. The mean (SD) of age was 46.66 (17.67) years in T1DM and 66.41 (13.19) years in T2DM.

In adults with T1DM, the incidence for hospital admission for DKA did not have any significant change between 1998 and 2004 and then substantially increased by 14.10% (95% CI 2.34–27.22, P = 0.02) annually from 2004 to 2007 (Fig. 1A). The incidence then remained largely stable. The incidence was much higher in 2013 than in 1998 (IRR 1.53 [1.09–2.16]). Trends in DKA hospital admission increased during the entire study period among adults with T1DM aged ≥35 years (Supplementary Fig. 1A and Supplementary Table 3), who were male (Supplementary Fig. 1B), with <1 year or ≥20 years of recorded diabetes history (Supplementary Fig. 1C), and few comorbidities (Supplementary Fig. 1D). A nonlinear trend was only seen in the South West region (Supplementary Fig. 1E). The incidence of hospital admission for DKA was the highest among adults with T1DM with <1 year of recorded diabetes history and young adults aged 18–24 years.

In adults with T2DM, the incidence of hospital admission for DKA increased from 0.70 to 0.98 per 1,000 person-years (Supplementary Table 2). The annual increase was 4.24% (95% CI 2.82–5.69, P < 0.0001) (Fig. 1B). The incidence increased among those aged <50 or ≥65 (Supplementary Fig. 2A and Supplementary Table 3), for both sexes (Supplementary Fig. 2B), with ≥10 years of recorded diabetes history (Supplementary Fig. 2C), and current insulin users (Supplementary Fig. 2E). The trend was stable among sulfonylureas users. A decreasing trend was observed among adults with T2DM who were not currently being treated with insulin and sulfonylureas. No nonlinear trend was found by region (Supplementary Fig. 2F). Of note, the incidence of hospital admission for DKA was the highest among insulin users and also high in young adults and those with <1 year or ≥10 years of recorded diabetes history.

The median length of hospital stay in days decreased over time for both T1DM (P < 0.0001) and T2DM (P = 0.0004) (Fig. 2A). No difference was found between two diabetes types (rate ratio 0.93 [95% CI 0.81–1.07]). When hospital stay was examined as a categorical variable (Fig. 2B and C), adults with T1DM were more likely to be discharged within 2 days (OR 1.28 [1.07–1.53]) compared with adults with T2DM.

The 30-day all-cause readmission rate did not change over time for T1DM (P = 0.94) and T2DM (P = 0.13), but the readmission rate was higher in T1DM than in T2DM (OR 1.61 [95% CI 1.04–2.50]) (Fig. 3). Similarly, 30-day and 1-year all-cause mortality rates both remained unchanged for both diabetes types (P > 0.3) (Fig. 4A and B). The mortality rate did not differ by diabetes type.

Sensitivity analyses revealed that after excluding likely-to-be-misclassified adults with T1DM, the incidence of hospital admission for DKA increased 7.41% (95% CI 5.23–9.65, P < 0.0001) annually between 1998 and 2009, and no significant change was observed after then. After excluding likely-to-be-misclassified adults with T2DM, the incidence increased 4.35% (2.80–5.92, P < 0.0001) annually between 1998 and 2013. For T1DM using the entire sample, if we excluded the year 2004 data, an annual increase of 5.21% (2.67–6.77, P < 0.0001) was seen between 1998 and 2009, and then the incidence remained unchanged (Supplementary Fig. 3).

At least two hospital admissions for DKA occurred in 3.17–4.42% of adults with T1DM and 0.05–0.13% of adults with T2DM in three selected years (Supplementary Table 4). Compared with those without DKA hospital admission or with only one admission, adults with T1DM who had recurrent DKA admissions were younger (P < 0.0001), and adults with T2DM who had recurrent DKA admissions were younger, less likely to be male, had longer duration of recorded diabetes history, and were more likely to be receiving insulin therapy (P < 0.0001). The number of patients included in each year and according to the number of hospital admission for DKA were also shown for T1DM (Supplementary Table 5) and T2DM (Supplementary Table 6). Further analysis revealed that 65.72% and 35.28% of the total DKA hospital admissions were recurrent for T1DM and T2DM, respectively.

For T1DM, the incidence of hospital admission for DKA in 2013 was 21.44, 29.40, and 39.13 per 1,000 person-years, respectively, when including only the first DKA admission during the entire study period (Supplementary Fig. 4A), the first DKA admission in each year (Supplementary Fig. 5A), and all DKA admissions (Fig. 1A) from each patient. The DKA trends increased 1.63% (95% CI 0.32–2.95, P = 0.02) and 3.10% (1.93–4.28, P = 0.0001) annually between 1998 and 2013, respectively, when analyzing only the first DKA admission during the entire study period and the first DKA admission in each year. For T2DM, the incidence of hospital admission for DKA in 2013 was 0.68, 0.84, and 0.98 per 1,000 person-years, respectively, when including only the first DKA admission during the entire study period (Supplementary Fig. 4B), the first DKA admission in each year (Supplementary Fig. 5B), and all DKA admissions (Fig. 1B) from each patient. The DKA trends increased 2.02% (0.27–3.81, P = 0.03) and 3.40% (1.82–5.00, P = 0.0004) annually between 1998 and 2013, respectively, when analyzing only the first DKA admission during the entire study period and the first DKA admission in each year.

During the entire study period, the average 30-day all-cause readmission rate decreased from 12.67 to 6.05% for T1DM and from 6.18 to 3.74% for T2DM when comparing the analysis with including all DKA hospital admissions to including only the first DKA readmission for each patient (Supplementary Table 7). Similar to the primary analysis with all DKA admissions included (Fig. 3), the trends in the 30-day all-cause readmission rate remained stable over time for both T1DM and T2DM (P > 0.1) (Supplementary Fig. 6A and B). Adults with T1DM with three or more DKA admissions (OR 1.90 [95% CI 1.15–3.13]) and adults with T2DM with two or more DKA admissions (2.05 [1.16–3.62]) were associated with higher 1-year all-cause mortality compared with those adults with T1DM or T2DM with only one DKA admission (Supplementary Table 8). No difference was found for 30-day all-cause mortality and length of hospital stay for both T1DM and T2DM.

Among a large representative cohort of adults with T1DM or T2DM in England, from 1998 to 2013, we found an increased incidence of hospital admission for DKA before 2007 and no change was seen after then for adults with T1DM. The incidence of hospital admission for DKA increased steadily among adults with T2DM during the entire period. About 20% of all DKA admissions were from adults with T2DM. The length of hospital stay for adults with T1DM or T2DM decreased over time, but the 30-day all-cause readmission rate and 30-day and 1-year all-cause mortality rates remained unchanged. Nonetheless, adults with T1DM were more likely to be discharged within 2 days and also had higher readmission rate than adults with T2DM. These findings provide informative data to understand the long-term DKA burden and associated health care performance in England and to inform future practice-level or policy-level intervention for reducing DKA burden.

DKA is a well-recognized hallmark of T1DM, but our findings suggest that managing DKA in adults with T1DM has not been successful in England. We found that the incidence of hospital admission for DKA was 53% higher in 2013 than in 1998, although the incidence stopped increasing after 2007. Given the substantially higher DKA risk in T1DM than in T2DM, an elevated DKA trend in T1DM would contribute considerable health care and patient burden. Importantly, the increasing trend was the most prominent among adults with T1DM with <1 year of recorded diabetes history who also had the highest risk for DKA hospitalization. This subgroup of patients requires particular attention. For T2DM, most of the DKA admissions were from insulin users, and the increase in DKA incidence among insulin users was also the greatest among all subgroups of adults with T2DM, suggesting that appropriate use of and adherence to insulin therapy have been challenging.

Our data suggest that the incidence of hospital admission for DKA was >40-fold higher in T1DM than in T2DM, but every one in five DKA admissions was attributed to T2DM. These concur with the evidence from Wales, U.K. (11), the U.S. (26), and Nordic countries (27,28). The contribution of DKA burden from T2DM is more significant than traditionally recognized when DKA was considered a key clinical feature for T1DM (2). In fact, we revealed a steady increase in the incidence of hospital admission for DKA in T2DM. The underlying causes are unclear, but rising T2DM prevalence and incidence may be one (29). The risk of hospital admission for DKA was higher among those with <1 year or ≥10 years of recorded diabetes history. In our study population, the proportion of adults with T2DM with <1 year of recorded diabetes history decreased over time (Supplementary Fig. 7), but more adults had ≥10 years of recorded diabetes history. Therefore, the aging of population with T2DM may play a more important role. Although speculative, other reasons may be increased insulin use and increased prevalence of infection and ketosis-prone T2DM over time. Infection and noncompliance to insulin therapy are two of the most common precipitating factors for DKA (3). Whether there is an increased trend of infection in T2DM patients paralleling increased DKA trends needs future investigation. Our data indicate that insulin users had the greatest risk for DKA hospital admission in T2DM. A previous study reported that the prevalence of insulin use increased in earlier years before 2005 and then remained in that high level (20). DKA is commonly seen in ketosis-prone T2DM, which usually occurs in nonwhite patients (2), but correctly phenotyping this subtype is difficult.

Length of hospital stay and 30-day all-cause readmission and mortality rates are commonly used outcome measures for health care utilization and performance (18,19). The declining trend in DKA-related length of hospital stay found in our study may imply reduced hospital costs per DKA admission and improved treatment and management. Discrepancy existed by diabetes type. Adults with T1DM were much more likely to be discharged within 2 days (65% vs. 29% in 2013), which may be partly related to adults with T2DM being ∼20 years older than adults with T1DM. Although speculative, the shortened hospital stay may partly account for the unchanged 30-day all-cause readmission rate and 30-day and 1-year all-cause mortality rates, because it may be likely that patients were discharged too quickly before they were fully treated to reduce health care costs. However, the temporal relationship between length of hospital stay, short-term readmission and mortality rates, and hospital costs requires more investigations.

Approximately two-thirds of all DKA hospital admissions in T1DM and one-third in T2DM (mostly insulin users) were recurrent cases in our study. Therefore, restricting analyses to include only the first DKA hospital admission and the first DKA readmission considerably reduced the incidence rate and 30-day readmission rate for DKA, respectively. Nonetheless, we still observed an elevation of trends in hospital admission for DKA for both T1DM and T2DM during the entire study period after applying these restrictions. Other than previously hypothesized reasons for the rising DKA trends, the diagnostic criteria of DKA from the American Diabetes Association in 2001 recommended a classification of the severity of DKA (mild, moderate, and severe) to help with patient disposition and selection of therapy (30). The specific inclusion of DKA with mild metabolic acidosis may have resulted in an increased number of DKA cases, although patients with mild DKA can be treated and observed in the emergency department and may not be admitted for hospitalization.

Reducing DKA risk in adults has been challenging. Possibly, managing chronic complications, such as retinopathy, nephropathy, neuropathy, and amputation and cardiovascular disease risk, are the priorities, and cardiovascular disease is the primary cause of death and source of health care costs in people with T1DM or T2DM (31–33). However, managing acute complications such as DKA deserves more attention. Our study and most of the published studies across different countries reported increased/stable DKA rates or an increased number of hospital admissions, or both (10–12). Ideal management for DKA should prevent both the first and recurrent DKA and improve care during hospitalization and immediately after discharge. Unlike hypoglycemia, which is totally preventable (34), some DKA cases may not have an obvious precipitating cause, particularly in ketosis-prone patients with T2DM (1–3). Nonetheless, most DKA cases are preventable through implementing structured educational programs emphasizing the importance of continuation of and adherence to insulin therapy, improving sick day management (e.g., increase insulin doses if unwell, test ketones), having medications available for treating fever or infection, accessing medical care easily, and enhancing communication with a health care provider during an intercurrent illness (1–3).

Our study had several important limitations. First, we did not have data on DKA events admitted to the emergency department but not resulting in hospitalization. Second, misclassification of diabetes type was likely. Sensitivity analyses suggested that the potential misclassification would not change our conclusions. However, we could not rule out the possibility of misdiagnosing ketosis-prone T2DM as T1DM, and thus, its effect on the DKA trends in T1DM and T2DM was unknown (2,35). Third, the identification of DKA admissions was solely based on the ICD-10 diagnosis code and was not verified by laboratory criteria such as glucose level, pH level, serum bicarbonate level, anion gap, and serum/urine ketones level. Accordingly, DKA could have been miscoded or misdiagnosed. Fourth, the introduction of the Quality and Outcomes Framework (QOF) in 2004 may improve data completeness and facilitate identification of patients with diabetes in the CPRD (36). Coincidentally, the greatest increase was observed between 2004 and 2007 for adults with T1DM, although the incidence of hospital admission for DKA already went up before 2003. Our sensitivity analyses indicated that the trend before 2007 was largely influenced by the year 2004 data; dropping this year yielded a significant increase between 1998 and 2007. Also, we did not observe any unusual trends in 2004–2007 in the following variables in T1DM adults, including the number of included adults with T1DM, the number of DKA admissions, person-years with diabetes, percentage of adults with T1DM with duration of recorded diabetes history <1 year, percentage male, and percentage of patients aged 18–24 (Supplementary Table 9). Taken together, it is unlikely that our conclusion regarding the trends in hospital admission for DKA in T1DM adults was biased. Fifth, extending current DKA trends beyond 2013 should be done cautiously. Several changes were made in December 2015 and May 2017 to update the 2009 diabetes guideline from the National Institute for Health and Care Excellence in the U.K. (37,38). For example, sodium-glucose cotransporter 2 inhibitors have been added to the glucose-lowering therapy algorithm, and the recommended HbA_1c target level is 6.5% (48 mmol/mol) and 7.0% (53 mmol/mol) if more than one glucose-lowering drug is prescribed. It remains unknown whether these changes have influenced the DKA trends beyond 2013, although we do not anticipate any dramatic alteration in the DKA trends between 2014 and the time of the publication of this study. Sixth, DKA trends from our study may not be applicable to health care systems beyond the U.K. Seventh, we did not correct for multiple comparisons, and thus, we could not rule out that some findings may be due to chance. Lastly, we underestimated duration of diabetes, because we used the first diabetes visit date as an approximate for the diabetes diagnosis date. To avoid confusion, we used duration of recorded diabetes history throughout instead of diabetes duration.

In summary, reducing DKA hospital admission is urgent for adults with T1DM or T2DM in England, given the high incidence of hospital admission for DKA in T1DM and the increasing trend in T2DM. Adults with T2DM contribute 20% of DKA admissions, although the admission rate is 40-fold higher in adults with T1DM. In addition to preventing DKA hospital admission, successful management also needs to reduce short-term all-cause readmission and mortality rates for which stable trends have been seen in England. Future investigations are required to identify factors that contribute to the increased DKA risk and unchanged readmission and mortality rates for DKA.

Acknowledgments. The authors sincerely thank Meera Kumar and Ling Zhang, from Sanofi US (Bridgewater, NJ), for extracting the CPRD and HES data.

Funding. This study was supported by the Sanofi Global Nutrition Scholars program at the University of North Carolina at Chapel Hill.

Duality of Interest. V.W.Z. received financial support from the Sanofi Global Nutrition Scholars program, and J.J. is an employee of Sanofi US, Bridgewater, NJ. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. V.W.Z. analyzed the data and wrote and revised the manuscript. V.W.Z. and J.J. acquired the data. V.W.Z. and E.J.M.-D. designed the study. All authors reviewed and edited the manuscript and contributed to discussion. V.W.Z. and E.J.M.-D. 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.