Incidence and long-term outcomes of adult patients with diabetic ketoacidosis admitted to intensive care: A retrospective cohort study

Abstract

Aims

Diabetic ketoacidosis is a life-threatening but avoidable complication of diabetes mellitus often managed in intensive care units. The risk of emergency hospital readmission in patients surviving an intensive care unit episode of diabetic ketoacidosis is unknown. We aimed to report the cumulative incidence of emergency hospital readmission and costs in all patients surviving an intensive care unit episode of diabetic ketoacidosis in Scotland.

Methods

We used a national six-year cohort of survivors of first diabetic ketoacidosis admissions to Scottish intensive care units (1 January 2005–31 December 2010) identified in the Scottish Intensive Care Society Audit Group registry linked to acute hospital and death records (follow-up censored 31 December 2010). Diabetic ketoacidosis-related emergency readmissions were identified using International Classification of Disease-10 codes.

Results

During the study period, 386 patients were admitted to intensive care units in Scotland with diabetic ketoacidosis (admission rate 1.5/100,000 Scottish population). Median age was 44 (IQR 29–56); 51% male; 55% required no organ support on admission. Mortality after intensive care unit admission was 8% at 30 days, 18% at one year, and 35% at five years. A total of 349 patients survived their first intensive care unit diabetic ketoacidosis admission [mean (SD) age 42.5 (18.1) years; 50.4% women; 46.1% required ≥1 organ support]. Following hospital discharge, cumulative incidence of 90-day, one-year, and five-year diabetic ketoacidosis readmission (all-cause readmission) was 13.8% (31.8%), 29.7% (58.9%) and 46.4% (82.6%).

Discussion

Diabetic ketoacidosis in patients requiring intensive care unit admission is associated with high risk of long-term mortality and high hospital costs. An understanding of the precipitating causes of diabetic ketoacidosis in patients admitted to intensive care units may allow patients who are at high risk to be targeted, potentially reducing future morbidity and the substantial burden that diabetic ketoacidosis currently places on the healthcare system.

Keywords

Diabetic ketoacidosis intensive care unit long-term outcomes readmission mortality

Introduction

Diabetic ketoacidosis (DKA) is a life-threatening but avoidable complication of diabetes mellitus (DM), and an important cause of morbidity and mortality among patients with diabetes mellitus. As of yet, despite the potentially fatal nature of this disease state, there have been no studies investigating incidence and epidemiology of DKA admissions to intensive care units (ICUs) in Scotland.

DKA represents a state of acute metabolic stress, where the body suffers from an absolute or relative insulin deficiency.¹ The resulting hormone and electrolyte imbalances cause marked hyperglycaemia, acidaemia and ketonaemia.^2,3 On admission, these patients are acutely unwell and often require significant medical interventions;⁴ in a national cohort of patients admitted to two Canadian ICUs with DKA over an eight-year period, 39% required mechanical ventilation (MV), 12% required renal replacement therapy (RRT), and 17% required vasopressor treatment.⁵

The most recent Scottish Diabetes Survey conducted at the end of 2014 reported that 276,430 people recorded on local registers as living with DM in Scotland.⁶ Of these, 11% of registered patients had Type 1 diabetes, while 88% were diagnosed with Type 2 diabetes. Many Type 1 diabetics become symptomatic with severe, life-threatening DKA at diagnosis following a several-week history of polyuria, polydipsia, polyphagia, and weight loss.¹

DKA pathophysiology

Diabetic ketoacidosis is an acute derangement of the physiological mechanisms that normally compensate when the body is in a state of starvation. Despite large amounts of circulating glucose, the lack of insulin in the setting of elevated counter-regulatory hormones (glucagon, catecholamines, cortisol and growth hormone) means that the glucose cannot be used. During this time, ketone production exceeds peripheral utilisation; DKA ultimately develops when the increase in ketoacid production is so great that a metabolic acidosis results.

On admission to hospital, these patients have a distinct clinical presentation: polyuria, polydipsia, nausea or vomiting, diffuse abdominal pain and hyperventilation.⁷ Left untreated, DKA can lead to cerebral oedema (more common in paediatric DKA patients), coma or death.^8,9

Treatment of DKA

Fast and efficient management of DKA is essential in order to restore the patient to their normal metabolic state. The treatment plan for ICU admissions of DKA should be implemented within 6 h of admission: confirm diagnosis, correct hypovolaemia, resolve hyperglycaemia and correct electrolyte imbalances.¹⁰ Crucially, the underlying cause of the ketoacidosis must be identified, and the appropriate steps taken to treat it, as DKA-related mortality is usually a consequence of the underlying illness, rather than the associated metabolic complications.⁴ Common causes of DKA include patients with DM1 diagnosed on their first presentation to the hospital with ketoacidosis, non-compliance with therapy, infections and hospital-acquired DKA.^11–13 Patients who are admitted to hospital with pre-existing diabetes often have very complex insulin requirements that must be managed very carefully, especially in the setting of intercurrent illnesses. Non-compliance and infection have been named as the most common triggers of DKA, with UTIs and pneumonia responsible for the majority of infectious causes.^14,15

Readmission and mortality

Given that DKA is a potentially avoidable complication of DM, mortality among these patients should be minimal with appropriate treatment and diligent follow up care. However, one Canadian study reported one-year mortality rates for patients treated for DKA in ICUs at 9%, while another reported hospital-specific mortality in DKA at 1.5%.^5,13 Factors associated with mortality in DKA patients include increasing age, comorbidities such as cardiovascular disease or sepsis, and a history of substance abuse.^4,16

Readmission of patients with DKA is a well-documented occurrence and represents a serious problem for the patient and a substantial burden on the healthcare system. When considering the readmission of these patients, the economic burden of DKA treatment must also be taken into account. Rehospitalisation among patients admitted to ICU with DKA in Canada has been reported at 36% at one year after the index hospital admission.⁵

Despite a substantial literature base currently available on DKA, epidemiological studies of DKA patients in Scotland represent a gap in current research.

We, therefore, undertook a retrospective cohort study to describe the incidence, long-term mortality and readmission rates of all patients admitted to Scottish ICUs with DKA over a five-year period. The aim of the study was to report mortality rates and readmission rates for patients who were admitted to ICUs in Scotland with a diagnosis of DKA. Mortality and readmission rates were reported for the five years succeeding the initial ICU admission. Furthermore, the study aimed to identify factors associated with mortality and emergency readmission within one year of the index admission.

Methods

Design, setting and population

This was a retrospective cohort study. The ICU cohort used in this study comprised residents aged ≥ 16 years admitted to ICUs in Scotland between 1 January 2005 and 31 December 2010 inclusive. All ICUs involved in the study were either general ICUs or combined High Dependency Units (HDU) and ICUs. Data required for the study was retrieved from the Scottish Intensive Care Society Audit Group (SICSAG) database, the Scottish Morbidity Record of acute hospital admissions (SMR01), and linked death records. The SICSAG database captures all adult general intensive care activity within Scotland. In 2005, mid-year population estimates from the National Records of Scotland¹⁷ indicate that the 24 adult ICUs across Scotland were serving a population of 5.1 million (4.2 million aged ≥16 years). By 2010, this had risen to a population of 5.3 million (4.3 million aged ≥16 years). The cohort for this study was defined as patients admitted to the 24 study ICUs with a primary ICU admission diagnosis of DKA recorded using either the Scottish Intensive Care (SICS) or acute physiology and chronic health evaluation (APACHE) III diagnostic coding systems.

Follow-up period

For mortality outcomes, follow-up commenced on the day of ICU admission and ended at five years or before if censored. For resource use outcomes, follow-up commenced from the day of index hospital discharge and ended at five years or before if censored. The final date for follow-up and censoring was 31 December 2010. Therefore, follow-up for patients in the cohort potentially ranged from one day to five years. Loss to follow-up was accounted for in all analyses.

Outcomes

Thirty-day, one-year and five-year mortality rates were derived from linked death records. Emergency hospital resource use over the five-year study period was obtained from linkage to the SMR01 database and quantified in three ways: mean number of hospital emergency readmissions per year per patient, mean number of emergency inpatient days in hospital per year per patient and cumulative incidence proportion experiencing an emergency readmission. The cumulative incidence proportion of emergency hospital readmission or death was also reported (see below). Elective hospital resource use was not included in analyses. The cause of emergency hospital readmissions was ascertained using the World Health Organisation International Classification of Disease (WHO ICD-10) admission diagnosis code on the hospital record of each readmission. Cause of readmission was classified into three groups, with groups 2 and 3 being a subset of the preceding group: (1) all emergency readmissions (any ICD-10 code); (2) emergency readmissions due to diabetes mellitus (ICD-10 codes E10.X to E14.X); and (3) emergency readmissions due to DKA (ICD-10 codes E10.1, E11.1, E12.1, E13.1, E14.1).

Variables

Demographic variables were: age, sex, social deprivation (as measured by the Scottish Index of Multiple Deprivation) and rurality.¹⁸ Clinical variables were ICU length of stay (LOS), hospital LOS, lowest recorded Glasgow Coma Scale (GCS) score, and degree of organ support on day 1 and during stay. Organ support consisted of mechanical ventilation (MV), renal replacement therapy (RRT) and cardiovascular inotrope support. Laboratory variables were: lowest recorded serum bicarbonate, highest creatinine and bilirubin, and paCO₂ at lowest and highest pO₂ values. The APACHE II score is a severity of illness score that is derived from patient characteristics and physiological parameters recorded within 24 h of patient admission to ICU.¹⁹ Patients are categorised with an integer score between 0 and 71. The score is calculated based on the patients’ age, chronic health status, operative status (operative vs non-operative), emergency or elective status, and a series of clinical and laboratory measures.

The American Diabetes Society has defined the criteria for DKA in degrees of severity (mild, moderate and severe) according to various clinical measures (Table 1).¹⁵ We used serum bicarbonate to derive DKA severity as it was the variable with least missing data.

Table 1.

American Diabetes Association diagnostic criteria for DKA severity.

	Mild	Moderate	Severe
Plasma glucose	>250 mg/dl	>250 mg/dl	>250 mg/dl
Arterial pH	7.25–7.30	7.00 to <7.24	<7.00
Serum bicarbonate (mEq/1)	15–18	10 to <15	<10
Urine ketones	Positive	Positive	Positive
Serum ketones	Positive	Positive	Positive
Anion gap	>10	>12	>12
Mental status	Alert	Alert/drowsy	Stupor/coma

DKA, diabetic ketoacidosis.

Statistical analysis

All analyses were carried out using SPSS 21.0 (Statistical Package for the Social Sciences). Baseline characteristics were reported as follows: normally distributed variables were reported as means with standard deviation (SD); non-normally distributed continuous data were reported as medians with interquartile range (IQR); categorical data were reported as proportions. A p < 0.05 was considered statistically significant for all comparisons and 95% confidence intervals were used. All p-values were two-sided.

Median time to death and median time to emergency hospital readmission were reported if available. Kaplan Meier survival analysis was used to estimate a population survival curve over five years for the whole cohort and to report the cumulative incidence proportion of emergency hospital readmission. Cumulative incidence proportion of the combined outcome emergency hospital readmission or death was also calculated to assess if death was a substantial competing risk for emergency hospital readmission (i.e. to assess if substantial numbers of deaths were occurring without preceding emergency hospital readmissions).

Each measure of resource use (emergency hospital readmission rate, number of days in hospital during emergency readmissions and cumulative incidence proportion of emergency hospital readmission) was stratified by the cause of readmission and plotted against the time after index hospital discharge. The number of emergency hospital readmissions days in hospital was calculated per year per patient alive and under follow-up at three-month intervals (Quarters).

Factors associated with mortality and mortality/emergency readmission

Univariable associations between baseline characteristics and one year mortality and one year emergency readmission or death were assessed using the following tests: normally distributed variables were compared using a Student’s t-test; non-normally distributed continuous data were compared using Mann–Whitney U test; categorical data were compared using χ² or Fisher’s exact test. As the proportional hazards assumption was violated, Cox regression analysis could not be used to identify factors independently associated with outcomes in multivariable analyses. This would have been the best approach to allow use of the whole cohort as time under follow-up varied due to censoring. Instead, we used multivariable binary logistic regression to identify factors associated with one-year mortality. Only those patients with at least one complete year of follow-up were included in these analyses. Variables were entered into regression models if p < 0.1 on univariable association testing.

Approvals

As the study used routinely collected anonymised data, a waiver was granted by the research ethics committee. The Privacy Advisory Committee of Information Services Division, NHS National Services Scotland, granted approval.

Results

From 2005 to 2010, there were a total of 386 patients admitted to the study ICUs with a diagnosis of DKA equating to an incidence of 1.5 per 100,000 of the Scottish population.

Of the whole cohort of 386 patients, 346 (89.6%) had a primary diagnosis of DKA in both APACHE and SICS coding systems. The remaining 40 (10.4%) had a primary diagnosis of DKA in the APACHE coding system, but a non-DKA primary diagnosis in the SICS coding system. Follow-up ranged from 6 to 1826 days from the time of ICU admission (median 1056, 95% CI 997 to 1134 days) and ranged from 7 to 1826 days from the time of hospital discharge in hospital survivors (median 1040, 95% CI 978 to 1129 days).

Median age was 44 years (IQR 29-56), 51% of subjects were male, and 77% were admitted to the ICU from the emergency department. Patients living in areas of socio-economic deprivation were over-represented in the sample, with 55% living in the two most deprived quintiles.

Median ICU LOS was 1.9 days (IQR 1.0-4.1) and median hospital LOS was eight days (IQR 4-18). With respect to interventions on day of admission, 38% of patients required MV, 10% required RRT and 30% required cardiovascular support. Of the whole cohort, 290 patients had recorded serum bicarbonate values, of whom 219 (76%) were classified as ‘severe DKA’ according to ADA classifications.

Table 2 shows the clinical characteristics of patients stratified by primary diagnosis of DKA in both APACHE and SICS coding systems versus DKA in APACHE coding system and another primary diagnosis in the SICS coding system. The alternative primary diagnoses in the SICS coding system are summarised in Figure 1.

Figure 1.

Alternative primary diagnoses in SICS coding system for patients whose primary diagnosis in APACHE and SICS systems was discordant.

Table 2.

Demographic and clinical characteristics of patients stratified by primary diagnosis.

Variable	All patients (n=386)	Primary DKA diagnosis (n=346)	Other primary diagnosis (n=40)	p ^a
Age [median (IQR)]	44 (29–56)	43 (27–55)	51 (41–65)	0.001
Male sex, n (%)	195 (51)	171 (49)	24 (60)	0.205
Socioeconomic status (SIMD quintile), n (%)				0.293
(least deprived) 1	41 (11)	38 (11)	3 (8)	–
2	52 (14)	41 (12)	11 (28)	–
3	78 (20)	70 (20)	8 (20)	–
4	98 (25)	91 (26)	7 (18)	–
(most deprived) 5	116 (30)	105 (30)	11 (28)	–
Urban setting, n (%)	330 (86)	301 (87)	29 (72)	0.012
Original admission to ICU, n (%)				0.002
A&E	298 (77)	274 (79)	24 (60)	–
Hospital ward	40 (10)	36 (10)	4 (10)	–
Other	48 (13)	36 (10)	12 (30)	–
ICU LOS [median (IQR)]	1.9 (1.0–4.1)	1.8 (0.9–3.8)	3.89 (1.2–10.2)	0.003
APACHE II score[mean (SD)]	21 ± 8	21 ± 8	26 ± 9	<0.001
Interventions on day 1, n (%)
MV	146 (38)	115 (33)	31 (78)	<0.001
RRT	39 (10)	32 (9)	7 (18)	0.101
Inotrope therapy	115 (30)	93 (27)	22 (55)	<0.001
Max no. of organs supported on day 1, n (%)				<0.001
0	211 (55)	205 (59)	6 (15)	–
1	71 (18)	58 (17)	13 (33)	–
2 or more	104 (27)	83 (24)	21 (53)	–
Lowest GCS, n (%)
3–7	54 (18)	43 (16)	11 (43)	0.028
8–11	54 (18)	50 (18)	4 (15)	–
12–15	192 (64)	181 (66)	11 (42)	–
DKA severity, n (%)			0.048	0.002
Mild	36 (12)	26 (10)	10 (31)	–
Moderate	35 (12)	31 (12)	4 (13)	–
Severe	219 (76)	201 (78)	18 (56)	–
High Creatinine [median (IQR)]	184 (124–272)	182 (121–264)	244 (137–327)	0.012
High Bilirubin [median (IQR)]	6 (4–10)	6 (4–9)	7 (4–12)	0.510
pCO₂ [mean (SD)]	4.3 ± 1.7	4.2 ± 1.7	5.4 ± 1.9	<0.001

DKA: diabetic ketoacidosis; GCS: Glasgow Coma Scale; LOS: length of stay; MV: mechanical ventilation; RRT: replacement renal therapy; SIMD: Scottish index of multiple deprivation; wbc: white blood cell count.

Univariable analyses: Student’s t-test for continuous parametric data; Mann–Whitney U for continuous non-parametric data; and χ² or Fisher exact tests for comparisons of categorical data

Patients with a primary diagnosis of DKA in both APACHE and SICS coding systems were younger on average, had higher rates of A&E admission, shorter ICU LOS and a lower degree of organ support on their first day of ICU admission.

Mortality

Mortality for the whole cohort at 30 days was 8% (95% CI 5–11%). Mortality at one year was 18% (95% CI 9–27%), and at five years was 35% (95% CI 26–44%). Figure 2 shows a Kaplan–Meier plot for overall survival. Table 3 shows the associated life table for the cohort over the five-year study period. At the end of year five, the cumulative proportion of patients surviving was 0.65.

Figure 2.

Kaplan–Meier plot for overall survival over the five-year study period.

Table 3.

Life table for whole study cohort over five years.

	Year 0–1	Years 1–2	Years 2–3	Years 3–4	Years 4–5
No. entering interval	386	270	207	133	71
No. censored	53	50	62	55	43
No. deaths	63	13	12	7	2
Cum. Proportion surviving	0.82	0.78	0.73	0.68	0.65

Factors associated with one-year mortality

Table 4 shows the demographic and clinical characteristics of patients stratified by mortality at year 1. Analysis of one-year mortality was only performed on patients who remained under follow-up for at least one year from the initial ICU admission. In total, 324 patients had a follow-up period of at least one year from the initial ICU admission.

Table 4.

Demographic and clinical characteristics of patients according to death one year after index DKA admission.

Variable	Dead within one year of index ICU admission (n = 56)	Alive within one year of index ICU admission (n = 268)	p ^a
Age [median (IQR)]	53 (40–66)	42 (25–54)	0.001
Male sex, n (%)	33 (59)	125 (47)	0.107
Socioeconomic status (SIMD quintile), n (%)			0.618
(least deprived) 1	6 (11)	22 (8)	–
2	7 (13)	38 (14)	–
3	10 (18)	56 (21)	–
4	11 (20)	75 (28)	–
(most deprived) 5	22 (39)	77 (29)	–
Urban setting, n (%)	46 (82)	229 (85)	0.540
Original admission to ICU, n (%)			0.663
A&E	42 (75)	208 (78)	–
Hospital ward	7 (13)	31 (12)	–
Other	7 (13)	29 (11)	–
Primary diagnosis of DKA, n (%)	47 (84)	240 (90)	0.248
ICU LOS [median (IQR)]	1.8 (0.9–5.0)	1.9 (1.0–4.0)	0.690
APACHE II score[mean (SD)]	26 ± 9	20 ± 7	<0.001
Interventions on day 1, n (%)
MV	29 (52)	94 (35)	0.023
RRT	6 (11)	29 (11)	1.000
Inotrope therapy	23 (41)	76 (28)	0.079
Max no. of organs supported on day 1, n (%)			0.017
0	21 (38)	155 (58)	–
1	15 (27)	44 (16)	–
2 or more	20 (35)	69 (25)	–
Lowest GCS, n (%)			0.008
3–7	13 (33)	33 (15)	–
8–11	8 (21)	38 (17)	–
12–15	18 (46)	147 (67)	–
DKA severity, n (%)			0.001
Mild	11 (28)	19 (9)	–
Moderate	6 (15)	23 (11)	–
Severe	22 (56)	172 (80)	–
High Creatinine [median (IQR)]	243 (142–307)	177 (122–266)	0.005
High Bilirubin [median (IQR)]	9 (5–12)	6 (4–8)	0.020
pCO₂ [mean (SD)]	5.4 ± 2.2	4.1 ± 1.6	0.001

APACHE: acute physiology and chronic health evaluation; CV support: inotrope therapy; DKA: diabetic ketoacidosis; GCS: Glasgow Coma Scale; LOS: length of stay; MV: mechanical ventilation; RRT: replacement renal therapy; SIMD: Scottish index of multiple deprivation; wbc: white blood cell count.

Univariable analyses: Student’s t-test for continuous parametric data; Mann–Whitney U for continuous non-parametric data; and χ2 and Fisher exact tests for comparisons of categorical data.

Factors associated with mortality at one year from ICU admission were: increasing age, mechanical ventilation on day of ICU admission, lower GCS, elevated serum bicarbonate, elevated creatinine and bilirubin, and elevated pCO₂. From binary logistic regression, the only statistically significant independent predictors of mortality at one year from ICU admission were increasing age (OR 1.031, 95% CI 1.003–1.059, p = 0.029) and severe DKA relative to mild DKA (OR 0.248, 95% CI 0.072–0.857, p = 0.027).

Resource use

Of the whole cohort of 386 patients, 32 died during their index admissions and a further five remained in hospital on the censoring date. The remaining 349 patients were included in analyses relating to resource use.

Mean number of emergency hospital readmissions

The mean number of emergency hospital readmissions in the first year was 1.71 (95% CI 1.38–2.09). Over the five-year period, the mean number of readmissions whilst under follow-up was 3.18 (95% CI 2.56 to 3.88). Figure 3 shows the mean number of emergency hospital readmissions with 95% confidence intervals for all patients surviving to hospital discharge. The mean number of readmissions peaked at 2.17 readmissions per person alive at the beginning of Quarter 2 (year 1), and decreased over time to a low of 0.44 readmissions per person alive at the beginning of Quarters 18–19 (year 5).

Figure 3.

Mean annual number of readmissions for five years after hospital discharge (95% confidence intervals denoted by dashed lines).

Figure 4 shows the mean number of emergency hospital readmissions stratified by the diagnosis on readmission to hospital. For readmissions coded with a ‘Diabetes Mellitus’ diagnosis, the mean number of readmissions peaked at 1.1 readmissions per person alive at the beginning of Quarter 1 (year 1). There was a downward trend in emergency readmissions during the five-year period. Around half of all emergency readmissions were due to a diabetes-related condition according to ICD-10 coding. The majority of these diabetes-related readmissions were due to DKA.

Figure 4.

Mean annual number of readmissions stratified by diagnosis for five years after hospital discharge (95% confidence intervals).

Cumulative incidence proportion of readmission or death

Figure 5 shows the cumulative risk of readmission or death, stratified by the readmission diagnosis. By year 5, the cumulative risk of DM-related emergency hospital readmission or death was 65%. The majority of DM-related readmissions were due to DKA (cumulative risk 59% over five years). The cumulative incidence of readmission or death at five years was 83%.

Figure 5.

Cumulative incidence proportion of readmission or death stratified by readmission diagnosis for five years after index hospital admission.

Discussion

The incidence of DKA admissions to ICUs across Scotland was 1.5 per 100,000 of the population. On admission, these patients often require advanced life support; 38% required MV, 30% required cardiovascular support and 10% required RRT. Mortality in this cohort at 30 days after the index hospital admission was 8%, rising to 18% at one year, and to 35% at five years. Emergency rehospitalisation among patients admitted to the ICU with DKA was common, occurring in 45% at one year, and 61% at five years. Furthermore, the average number of readmissions over the five-year period was 3.2 per patient. Independent predictors of mortality at one year after index ICU admission were elevated serum bicarbonate, creatinine and bilirubin, low GCS score, and requirement for MV on day 1.

Interpretation

Few studies have evaluated the characteristics, long-term outcomes, and resource use of adult patients with DKA admitted to the ICU, with no previous studies conducted of this nature in Scotland. This study flagged several concerning features of patients with DKA admitted to ICUs in Scotland.

First, the high mortality rates, which indicated that 35% of patients with DKA admitted to the study ICUs died within five years of their index hospital admission. Elevated creatinine and bilirubin were identified as independent predictors of mortality one year after ICU admission in univariable analyses. In the setting of DKA, the elevated creatinine may be due to volume depletion caused by polyuria, a false positive result caused by the increased level of acetoacetate in the body, or diabetic nephropathy.²⁰ Our cohort has higher rates of short-term mortality compared with other studies¹³; there are several possible explanations for this. First, patients in Scotland who are admitted to hospital with DKA and who require critical care admission are most likely to be managed in High Dependency Units.²¹ This study focussed on patients admitted to ICUs only, so by default these patients were more unwell, and therefore at greater risk of mortality on admission. This is highlighted by the fact that 38% of this Scottish cohort were mechanically ventilated within 24 h of admission compared with 6.9% in a recent study undertaken in Australian and New Zealand ICUs.¹³ Further, previous papers have focussed specifically on hospital mortality, where we have reported 30-day mortality rates, which may include patients who survive the index hospital admission.¹³

DKA severity defined by serum bicarbonate was identified as an independent predictor of mortality at one year, paradoxically suggesting that those with mild DKA had a higher risk of mortality than those with severe DKA. Decreased mental status and a requirement for mechanical ventilation on the first day of admission would suggest, however, that these patients were acutely unwell, despite the fact that they showed a milder severity of DKA than their counterparts who were alive after one year. The underlying disease in patients with DKA has previously been identified as the primary cause of death in DKA-related mortality, over and above the complications arising from the metabolic derangements.⁴ This would suggest that death in the cohort is likely due to the precipitating cause, or related comorbidities in these patients. The discrepancies in coding systems further demonstrate this; patients with a primary diagnosis of DKA in both APACHE and SICS coding had shorter ICU LOS and required a lower degree of organ support. One potential reason for this pattern is that the patients with the most severe DKA are generally younger patients who have presented with DKA due to non-compliance issues. In contrast, older patients may be more often admitted with mild DKA presenting secondary to other more life threatening comorbidities.¹³

With relevance to severity, it is important also to consider acute kidney injury (AKI) – a frequent complication in patients with severe DKA.²² In critically ill patients, AKI is a significant risk factor for morbidity and mortality. Despite this, current JBDS recommendations for critical care referral for patients with DKA do not include AKI;²¹ larger studies using a cohort of patients selected at the time point of hospital admission, rather than ICU admission, are needed to identify whether this should be a factor in referral of DKA patients to ICU.

With the high incidence of co-existing cardiovascular conditions in patients with diabetes, MI and congestive cardiac failure are common causes of death in patients who present in the ICU with DKA.²³ The Scottish Diabetes Survey reported 4% of patients with Type 1 DM had survived a previous myocardial infarction, 53% had recorded hypertension and 29% had recorded hypercholesterolaemia.⁶ Such trends are well documented in the Scottish population and, among patients with diabetes especially, represent a significant risk of death.²⁴

With good adherence to dietary restriction and insulin therapy, readmission of patients with DKA should be negligible. Yet over the five-year study period, 59% of patients, who survived the index hospital admission, consequently died or were readmitted with a diagnosis of DKA. This finding is of importance nationally and internationally.

The causes of readmission among patients with DKA have previously been stratified into those amenable to correction and those that are more difficult to solve.¹² Correctable problems include: failure to understand or adjust to the disease and the importance of treatment adherence; inadequate discharge plans and follow-up care relevant to the patients’ needs. Problems that are more difficult to solve tend to lie outwith the scope of interventional strategies, such as the natural biologic course of the disease and co-existing illnesses. Socioeconomic, psychosocial and educational disparities have been attributed to varying readmission rates between patients and are also considered to be important predictors of readmission among patients with DM.^12,25

Non-compliance has repeatedly been identified as a major cause of readmission among patients who have presented with DKA in the past.^5,12,26,27 Various demographic characteristics have been identified as contributing towards increased rates of non-compliance: adolescence, ethnic minority status, co-existing mental health problems and low socioeconomic status.^12,26 In recent years, a large body of literature has developed on diabetes education and its efficacy, both among paediatric and adult populations. Among adult patients with diabetes mellitus, the duration of contact time between patient and education has been identified as a predictor of improved glycaemic control, but the effectiveness declines over a period of just 1–3 months after the intervention ceases.²⁸ An improved follow-up program with diabetes specialist nursing teams may be of value in a cohort of non-compliant patients who had an insulin prescription prior to their DKA admission.^29,30 Current research indicates that patients who are not on an established insulin regimen on presentation have poorer outcomes than those who did.¹³

The advent of sodium-glucose cotransporter 2 (SGLT2) inhibitors has been a source of some controversy, regarding the possibility that they place patients at increased risk of developing normoglycaemic DKA.^31,32 Currently indicated for patients with type 2 DM, the exact mechanism by which these oral hypoglycaemic drugs may precipitate a normoglycaemic DKA is unknown. Nonetheless, given this uncertainty, it would be useful to monitor patients’ compliance with SGLT2 medication and any associated admissions to hospital with DKA.

Outwith the natural biologic course of the disease, several other factors may increase infection susceptibility in patients with DM; one such factor of note in the Scottish population is the prevalence of cigarette smoking. The Scottish Diabetes Survey reported smoking status in patients with Type 1 DM as 24%.⁶ Mucous accumulation in the small and large airways interferes with the frequency of ciliary beats, precluding clearance of bacteria and viruses from the respiratory tract.³³ Consequently, a quarter of patients with DM 1 in the Scottish population are left at increased vulnerability to respiratory tract infections such as pneumonia – one of the most common infectious precipitants of DKA.¹⁵

Strengths and limitations

The main strength of this study was the availability of five-year outcome data, which allows analysis of long-term outcomes in way that has never previously been produced for the Scottish population. Other strengths include the use of a complete national cohort of patients and inclusion of all ICU admissions over this five-year period.

The study had a number of limitations to take into consideration. First, the relatively small sample size limited the statistical power of the study thereby increasing the risk of type II error, particularly in relation to multivariable analyses, and multiple hypothesis testing in univariable analyses may have increased the risk of type I error. Second, the data available focused solely on DKA patients admitted to ICU. Therefore, there were no data on DKA patients who were admitted to the emergency room and consequently managed in hospital wards or HDUs where the majority of patients with DKA are treated in Scotland. Third, the serum bicarbonate was the only reliable recorded marker of DKA severity in the dataset and likely an inadequate representation of DKA severity. Fourth, there was no detailed information available as to the causes of emergency readmission. Fifth, the use of data that is somewhat outdated with respect to changes in healthcare since the censoring date of 31 December 2010. Sixth, we were unable to distinguish between hospital-acquired DKA, DKA as a first presentation of diabetes, and other causes.³⁴ This may limit generalisability of our findings to all patients admitted to ICUs with DKA. Finally, much of the analyses were dependent on coding in the dataset, limiting the amount of information available for analysis (such as the precipitating cause of index admission or rehospitalisation).

Implications for health policy

This study highlights the importance of additional quantitative and qualitative research to positively identify the causes of admission, readmission and mortality in these patients. In a population who with the correct treatment should be able to avoid readmission and mortality, the gap in the healthcare system needs to be identified. Evaluation of follow-up and discharge information given to the patient on the index admission and subsequent readmissions may provide a starting point for such investigations.

Our data extract did not identify how many patients admitted with DKA were discharged directly from ICU and which patients were seen by diabetes inpatient specialist nurses (DISNs) or other members of the diabetes team prior to discharge. The high risk of readmission in our cohort suggests that extensive education as well as follow-up by DISNs may provide a way forward in reducing this problem, as recommended by national guidance.²¹ Healthcare resources could then be more effectively directed, which would reflect on the potency of DKA management as a whole.

Conclusion

Diabetic ketoacidosis in adult patients requiring ICU admission is associated with a significant risk of long-term mortality or readmission. Currently, a significant proportion of patients might be slipping through the cracks, leading to high readmission and mortality rates in a population that should have negligible rates.

Further research should investigate the exact precipitants of DKA in patients admitted to the ICU, as this may identify the mechanisms underlying the unfavourable long-term outcomes revealed in this study. This may then allow patients who are at risk of mortality or serial readmission to be targeted, evading significant future stress for the patient and lifting the substantial burden that DKA currently places on the healthcare system.

Glasgow Coma Scale

Footnotes

Acknowledgements

I would like to thank my supervisor, Dr Nazir Lone, for his supervision and excellent statistical guidance throughout. Due to limitations in my statistical training, some analyses, in particular those relating to resource use, were undertaken by my supervisor.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was supported by a scholarship from Medical Research Scotland. We are grateful to SICSAG for supplying data used in these analyses.

References

Silverstein

Klingensmith

Copeland

. Care of children and adolescents with type 1 diabetes. Diab Care 2005, pp. 186–212.

Doshi

Potter

Santos

DDL

. Prospective randomized trial of insulin glargine in acute management of diabetic ketoacidosis in the emergency department: a pilot study. Acad Emerg Med 2015; 22: 657–662.

Maletkovic

Drexler

. Diabetic ketoacidosis and hyperglycemic hyperosmolar state. Endocrinol Metab Clin North Am 2013; 42: 667–695.

Freire

Umpierrez

Afessa

. Predictors of intensive care unit and hospital length of stay in diabetic ketoacidosis. J Crit Care 2002; 17: 207–211.

Azevedo

LCP

Choi

Simmonds

. Incidence and long-term outcomes of critically ill adult patients with moderate-to-severe diabetic ketoacidosis: retrospective matched cohort study. J Crit Care 2014; 29: 971–977.

Scottish Diabetes Survey Monitoring Group. Scottish diabetes survey. Edinburgh: NHS Scotland, 2014, pp.1–82, http://www.diabetesinscotland.org.uk/Publications/SDS2011.pdf (accessed 4 July 2015).

Hardern

Quinn

. Emergency management of diabetic ketoacidosis in adults. Emerg Med J 2003; 20: 210–213.

Holman

Herron

Sinnock

. Epidemiologic characteristics of mortality from diabetes with acidosis or coma, United States, 1970-1978. Am J Public Health 1983; 73: 1169–1173.

Levin

. Cerebral edema in diabetic ketoacidosis. Pediatr Crit Care Med 2008; 9: 320–329.

10.

Kohler

Levy

. Management of diabetic ketoacidosis: a summary of the 2013 Joint British Diabetes Societies Guidelines. J Int Car Soc 2014; 15: 222–225.

11.

Jones

NRV

Fischbacher

Guthrie

. Factors associated with statin treatment for the primary prevention of cardiovascular disease in people within 2 years following diagnosis of diabetes in Scotland, 2006–2008. Diabet Med 2014, pp. 640–646.

12.

Flexner

Weiner

Saudek

. Repeated hospitalization for diabetic ketoacidosis. The game of “Sartoris”. Am J Med 1984; 76: 691–695.

13.

Venkatesh

Pilcher

Prins

. Incidence and outcome of adults with diabetic ketoacidosis admitted to ICUs in Australia and New Zealand. Crit Care 2015; 19: 451.

14.

Dhatariya KK, Nunney I, Higgins K, et al. National survey of the management of Diabetic Ketoacidosis (DKA) in the UK in 2014. Diabet Med 2016; 33: 252–260.

15.

Kitabchi AE, Umpierrez GE and Murphy MB. Diabetic ketoacidosis and hyperosmolar state. In: DeFronzo R, Ferrannini E and Alberti KGMM (eds) International Textbook of Diabetes Mellitus, 2 Volume Set, 4th Edition. Chichester: Wiley-Blackwell, 2015, pp.799–814.

16.

Henriksen

Røder

Prahl

. Diabetic ketoacidosis in Denmark. Incidence and mortality estimated from public health registries. Diabetes Res Clin Pract 2007; 76: 51–56.

17.

National Records of Scotland. Mid-year population estimates. Edinburgh: National Records of Scotland, 2005. http://nationalrecordsofscotland.gov.uk/statistics-and-data/statistics/statistics-by-theme/population/population-estimates/mid-year-population-estimates (accessed 16 October 2015).

18.

The Scottish Government. Scottish Index of Multiple Deprivation 2012: a national statistics publication for Scotland 18 December 2012. Edinburgh: The Scottish Government, 2012, pp.1–117, http://22fa0f74501b902c9f11-8b3fbddfa1e1fab453a8e75cb14f3396.r26.cf3.rackcdn.com/simd_448749_v7_20121217.pdf (accessed 21 July 2015).

19.

Markgraf

Deutschinoff

Pientka

. Performance of the score systems acute physiology and chronic health evaluation II and III at an interdisciplinary intensive care unit, after customization. Crit Care 2001; 5: 31–36.

20.

Kemperman

FAW

Weber

Gorgels

. The influence of ketoacids on plasma creatinine assays in diabetic ketoacidosis. J Intern Med 2000; 248: 511–517.

21.

Joint British Diabetes Societies Inpatient Care Group. The Management of Diabetic Ketoacidosis in Adults. Second edition. London: Joint British Diabetes Societies Inpatient Care Group for NHS Diabetes, 2013. http://www.diabetes.org.uk/Documents/About%20Us/What%20we%20say/Management-of-DKA-241013.pdf.

22.

Ghaddab

Van

Ichai

. Incidence and characteristics of acute kidney injury in severe diabetic ketoacidosis. PLoS One 2014; 9: 1–5.

23.

Slavíková

Kuncova

Topolcan

. Plasma catecholamines and ischemic heart disease. Clin Cardiol 2007; 30: 326–330.

24.

Information Services Division. Publication report heart disease statistics update. Edinburgh: Information Services Division, 2015.

25.

Dungan KM. The effect of diabetes on hospital readmissions. J Diab Sci Technol 2012; 6: 1045–1052.

26.

Wagner DV, Stoeckel M, Tudor ME, et al. Treating the most vulnerable and costly in diabetes. Curr Diab Rep 2015; 606.

27.

Morris

Boyle

McMahon

. Adherence to insulin treatment, glycaemic control, and ketoacidosis in insulin-dependent diabetes mellitus. The DARTS/MEMO Collaboration. Diabetes Audit and Research in Tayside Scotland. Medicines Monitoring Unit. Lancet 1997; 350: 1505–1510.

28.

Norris

Lau

Smith

. Self-management education for adults with type 2 diabetes. Diab Care 2002; 25: 1159–1171.

29.

Crasto

ZINZIN

Htike

Turner

. Management of diabetic ketoacidosis following implementation of the JBDS guidelines: where are we and where should we go? British Journal Diabetes 2015; 15: 11–16.

30.

James

Gosden

Winocour

. Diabetes specialist nurses and role evolvement: a survey by Diabetes UK and ABCD of specialist diabetes services 2007. Diab Med 2009; 26: 560–565.

31.

Ogawa

Sakaguchi

. Euglycemic diabetic ketoacidosis induced by SGLT2 inhibitors: possible mechanism and contributing factors. J Diabetes Investig 2016; 7: 135–138.

32.

Rosenstock

Ferrannini

. Euglycemic diabetic ketoacidosis: a predictable, detectable, and preventable safety concern with SGLT2 Inhibitors. Diab Care 2015; 38: 1638–1642.

33.

Bagaitkar

Demuth

Scott

. Tobacco use increases susceptibility to bacterial infection. Tob Induc Dis 2008; 4: 12.

34.

Hallett

Modi

Levy

. Developments in the management of diabetic ketoacidosis in adults: implications for anaesthetists. BJA Education 2016; 16: 8–14.