a b s t r a c t

Background: We investigated independent mortality predictors of hyperglycemic crises and developed a prediction rule for emergency and critical care physicians to classify patients into mortality risk and disposition groups.

Methods: This study was done in a university-affiliated medical center. Consecutive adult patients (N 18 years old) visiting the emergency department (ED) between January 2004 and December 2010 were enrolled when they met the criteria of a hyperglycemic crisis. Data were separated into derivation and validation sets–the former were used to predict the latter. December 31, 2008, was the cutoff date. Thirty-day mortality was the primary endpoint.

Results: We enrolled 295 patients who made 330 visits to the ED: derivation set = 235 visits (25 deaths: 10.6%), validation set = 95 visits (10 deaths: 10.5%). We found 6 independent mortality predictors: Absent tachycardia, Hypotension, Anemia, Severe coma, Cancer history, and Infection (AHA.SCI). After assigning weights to each predictor, we developed a Predicting Hyperglycemic crisis Death (PHD) score that stratifies patients into mortality-risk and disposition groups: low (0%) (95% CI, 0-0.02%): treatment in a general ward or the ED; intermediate (24.5%) (95% CI, 14.8-39.9%): the intensive care unit or a general ward; and high (59.5%) (95% CI, 42.2-74.8%): the intensive care unit. The area under the curve for the rule was 0.946 in the derivation set and 0.925 in the validation set.

Conclusions: The PHD score is a simple and rapid rule for predicting 30-day mortality and classifying mortality risk and disposition in adult patients with hyperglycemic crises.

(C) 2013

Introduction

Hyperglycemic crises present a disease continuum of diabetic emergency. The basic underlying mechanism is the combination of absolute or relative insulin deficiency and an increase in counter- regulatory hormones, such as glucagon, catecholamines, cortisol, and

* Corresponding author. Department of Family Medicine, Chi-Mei Medical Center, Tainan City 710, Taiwan. Tel.: +886 6 251 7844; fax: +886 6 283 2639.

E-mail address: [email protected] (S.-B. Su).

growth hormone [1]. There are three types of hyperglycemic crisis: [a] Diabetic ketoacidosis , [b] hyperosmolar hyperglycemic state (HHS) (two extremes of the same clinical syndrome), and [c] mixed syndrome (both DKA and HHS as a mixed state of acidosis and hyperosmolality) [2-7]. Despite recent improvements, the incidence and the cost of treating hyperglycemic crises is high and continues to rise. The annual DKA incidence rate has been estimated in population- based studies to range from 4.6 to 8 episodes per 1,000 patients with diabetes [3]. More recent epidemiological studies in the US report that the annual DKA incidence rate sharply increased during the past two decades [3]. In 2006, there were about 136,510 hospitalizations for

0735-6757/$ – see front matter (C) 2013 http://dx.doi.org/10.1016/j.ajem.2013.02.010

DKA in the U.S [8]. The average cost per patient per hospitalization was US$13000 [3]. The annual medical expenditure for healthcare providers to patients with DKA might exceed US$1 billion [3]. The incidence of and medical expenditure for HHS care are unknown because there are few population-based studies on HHS, and because many patients with HHS have Multiple comorbidities. The rate of hospital admissions for HHS was estimated more than a decade ago to be 1% of all primary diabetic admissions [9].

The mortality rate for hyperglycemic crises remains high: 1%-9% for DKA, 5%-45% for HHS, and 5%-25% for mixed DKA/HHS [1,4,5]. Among the elderly (>=65 years old), the mortality rate was recently reported to be as high as 71% [10]. Regardless of the observed decrease in the Death rate in the United States, hyperglycemic crises remain a serious heath problem, especially in developing and undeveloped countries. The death rate for hyperglycemic crises is primarily dependent on the treatment experience [10]. The cause of death in patients with hyperglycemic crises is rarely ascribed to the metabolic complications of hyperglycemia or ketoacidosis; rather, it is usually related to the underlying precipitating illness [1]. The most common precipitants are poor compliance, infection, new- onset diabetes, pancreatitis, acute coronary syndrome, stroke, and medications [1].

In the emergency department (ED), treatment for DKA, HHS, and mixed DKA/HHS is similar: correcting dehydration, hyperglycemia, and electrolyte imbalances; identifying comorbid precipitating events; and, above all, frequent monitoring of patients [1]. Two questions are always raised when treating hyperglycemic crises in the ED: [a] What is the mortality risk for this patient? and [b] What is the most appropriate disposition (viz., treatment in the intensive care unit [ICU], a general ward, or only in the ED) for this patient based on the balance of patient safety, cost, and availability of medical resources after ED treatment? As a result, some studies have proposed decision rules based on mortality predictors to help ED physicians make optimum decisions on how best to manage patients with hyperglycemic crises. However, studies to date have reported various predictors, some of which are impractical for use in the ED. For example, MacIsaac et al [5] argued that age was the only independent mortality predictor, Chung et al [4] used mental status rather than age, and Efstathiou et al [11] proposed a prediction model consisting of six independent mortality predictors for DKA. The last prediction model has two major drawbacks. First, the study enrolled only patients with DKA and provides no information about patients with HHS or mixed DKA/HHS. Second, several of the predictors cannot be immediately and easily obtained in the ED; therefore, the model is not useful in a first-line care unit. We thus conceived a research question for developing a feasible and applicable prediction rule for decision making in a

hospital ED.

defined as plasma glucose N 250 mg/dL, a high Anion gap metabolic acidosis (anion gap N 10, serum HCO₃ b 18 mmol/L, and pH b 7.3), and positive urine ketones or Serum ketones; [ii] HHS was defined as plasma glucose N 600 mg/dL, increased effective serum osmolality N 320 mOsm/kg, anion gap b 12, no significant acidosis (HCO₃ N 15 mmol/L or pH N 7.3), small urine ketones or serum ketones, and alteration in mental state; [iii] Mixed syndrome (DKA plus HHS) was defined as acidosis (pH b 7.3, HCO₃ b 18 mmol/L), positive urine ketones or serum ketones, and effective serum osmolality N 320 mOsm/kg. The effective serum osmolality was calculated with the formula: 2[measured Na⁺ (mEq/L)] + [glucose (mg/dL)]/18 [12]. There might be overlaps among the 3 types of hyperglycemic crisis, but since we were dealing with all 3 types of hyperglycemic crises as a whole, the overlaps would not affect our study results.

Data collection and definition of variables

All treatment of hyperglycemic crises in the studied hospital is strictly according to the guidelines suggested by the American Diabetes Association [1,3,12]. Patients were prospectively selected in the ED. Insufficient information was retrospectively collected by reviewers checking medical records after the patients had been discharged from the hospital. The study hospital’s Human Investiga- tion Committee approved the protocol. The reviewers were blinded to the patients’ hospital course and outcomes. Information for a number of variables for each patient was recorded (see Tables 1 and 2). Any variable not present in the patient’s medical history or physical exam was considered absent.

Table 1

Patient characteristics in the derivation and validation sets

Derivation set Validation set (n = 235) (n = 95)

Age (mean +- SD) 59.8 +- 20.9 60.2 +- 22.3

Gender: male (%) 49.4 48.9

Vital signs (mean +- SD)

Glasgow Coma Scale 12.9 +- 3.4 13.0 +- 3.1

Systolic blood pressure (mmHg) 137.1 +- 32.8 135.0 +- 31.0

Heart rate (1/min) 111.0 +- 23.3 112.9 +- 23.1

Respiratory rate (1/min) 20.5 +- 4.6 21.2 +- 5.8

Body temperature (?C) 36.7 +- 1.1 37.0 +- 1.1 Medical history (%)

Diabetes 73.2 79.5

Hypertension 40.4 52.3

Stroke 17.9 19.3

Cancer 8.5 8.0

Laboratory data (mean +- SD)

Our study is organized as follows: [a] identify univariate variables

Blood glucose (mg/dL)

766.9 +- 331.3 650.0 +- 237.3

3

of mortality in adult ED patients with hyperglycemic crises; [b] use multivariate logistic regression analysis to investigate independent mortality predictors; and [c] develop a prediction rule useful for ED physicians, one that allows them to classify patients with hypergly- cemic crises into mortality-risk groups, and that allows them to make an appropriate management decision, viz., treatment in the ICU, a general ward, or only in the ED.

Methods

Study design, setting, population, and selection of participants

This study was done in a 700-bed university-affiliated medical center in Taipei with a 40-bed ED staffed with board-certified emergency physicians who provide emergency care to approximately 55,000 patients per year. Consecutive adult patients (N 18 years old) visiting the ED between January 2004 and December 2010 were enrolled when they met the following criteria [12]: [i] DKA was

White blood cell count (cells/mm ) 12700.0 +- 5870 13400.0 +- 5936

Hemoglobin (g/dL) 13.9 +- 3.1 13.9 +- 2.8

Serum creatinine (mg/dL) 2.2 +- 1.6 1.8 +- 1.1

Effective serum osmolality (mOsm/kg)^a 325.1 +- 30.2 324.2 +- 32.1

Blood pH^b 7.3 +- 0.1 7.3 +- 0.1

Precipitating factors (%)^c

Poor compliance 59.6 61.4

Infection 45.1 40.9

New-onset diabetes 27.2 22.7

Pancreatitis 2.6 4.2

Acute coronary syndrome 3.0 2.1

Stroke 2.1 0

Subgroup diagnosis (%)

DKA 31.6 35.2

HHS 55.3 58.0

Mixed syndrome of DKA/HHS	13.1	6.8
Hospital length of stay (day)	13.9 +- 16.3	11.4 +- 9.5
30-day mortality rate (%)	10.6	10.5

SD, standard deviation.

^a Effective serum osmolality: 2[measured Na⁺ (mEq/L)] + [glucose (mg/dL)]/18.

^b 296 (89.7%) patients had this test.

^c Patients may have multiple precipitating factors.

Table 2

Validation set

(Jan 1, 2009-Dec 31, 2010)

95 visits

Derivation set

(Jan 1, 2004-Dec 31, 2008)

235 visits

38 visits: insufficient data or

received treatment in other hospital

330 visits enrollment

368 visits fitted in with the criteria of hyperglycemic crises

350,000 ED patient visits (Jan 1, 2004-Dec 31, 2010)

Univariate mortality predictors at P b 0.1

Variable Variable present P

Yes

n (% mortality)

No

n (% mortality)

Elderly (>= 65 years old)	114 (17.5)	121 (4.1)	.001
Altered mental status	82 (23.2)	153 (3.9)	b .001
Severe coma (GCS <= 8)	36 (33.3)	199 (6.5)	b .001
Hypotension (SBP b 90 mmHg)	18 (38.9)	217 (8.3)	.001
Absent tachycardia (HR <=100/min)	64 (17.2)	171 (8.2)	.046
Type 2 diabetes history	148 (13.5)	87 (5.7)	.079
Stroke history	42 (19)	193 (8.8)	.092
Bedridden history	21 (23.8)	214 (9.3)	.056

nasogastric tube feeding history 17 (23.5) 218 (9.6) .091

Cancer history 20 (40) 215 (7.9) b .001

Anemia (Hb b10 g/dL or Hct b30%) 26 (26.9) 209 (8.6) .011

Serum creatinine N 2 mg/dL 88 (17) 147 (6.8) .014

Infection as the precipitating factor 106 (22.6) 129 (0.8) b .001 SBP, systolic blood pressure; HR, heart rate; Hb, hemoglobin; Hct, hematocrit.

The categorical variables used are generally acceptable in emergency and critical care. Severe coma was defined as a Glasgow Coma Scale (GCS) <= 8 [13]. Significant anemia was defined as hemoglobin b 10 g/dL or hematocrit b 30% [14]. Hypotension was defined as systolic blood pressure b 90 mmHg [15]. Tachycardia was defined as a heart rate N 100 beats/min. “Absent tachycardia” was defined as a heart rate <=100 beats/min. We used absent tachycardia as a variable because tachycardia is a normal body response to dehydration caused by hyperglycemic crises [1]. Absent tachycardia is an abnormal response found in clinical practice and may indicate a poor prognosis. The definition of infection included lower respiratory tract infection, urinary tract infection, intra-Abdominal infection, skin or soft tissue infection, meningitis, bone/joint infection, perianal abscess, psoas muscle abscess, Infective endocarditis, and sepsis without focus. The Clinical impression of infection was based on the diagnoses of the treating physician’s documentation, laboratory, and image results (such as pneumonia on chest radiograph, pyuria on urinary analysis, abscess on computed tomography, etc).

Overall, 368 ED patient visits met the criteria of hyperglycemic crises. After 38 patients with insufficient data or who had been transferred after receiving treatment in other hospitals had been excluded, 330 patients were enrolled. The enrolled patients were divided into 2 groups according to their 30-day outcome, survival or death. All the study variables were used for comparisons between groups.

Derivation and validation sets

To identify multivariate mortality predictors used to develop a prediction rule, as well as to test the validity of the prediction rule, we separated the sample data into derivation and validation sets according to the ED visit time (Fig. 1). The derivation set enrolled patients and visits between January 1, 2004, and December 31, 2008. The validation set enrolled patients and visits between January 1, 2009, and December 31, 2010, a period after the prediction rule was created.

Definition of endpoint

We used 30 day mortality as the primary endpoint. People who survived at least 30 days whether or not they were still hospitalized were considered “survival” for this analysis.

Data analysis

All analyses were done using SPSS 16.0 for Windows (SPSS Inc, Chicago, IL, USA). Continuous data means +- SD. Comparisons

Test prediction rule

Development of prediction rule

Fig. 1. Patient enrollment and assignment to the derivation or validation set along with the flow of how the prediction rule was developed and validated.

between two groups were made using either an independent-samples t-test (assuming normal distribution) or Mann-Whitney/Wilcoxon tests (assuming non-normality) for the continuous variables. Either a ?² test or a Fisher’s exact test was used for categorical variables.

The significant ? level was set at 0.1 for univariate variables that

are included in a multiple logistic regression analysis of risk for 30-day mortality. Significance was set at P b .05 (2-tailed) to extract variables effective in a model. The area under the receiver operating characteristic curves was used to compare a model’s specifi- cations along with its sensitivity and specificity.

The results of the multivariate stepwise (forward) logistic regression analysis were then used to develop a clinical prediction rule [16]. Weights were assigned to each predictor according to their predicting ? values of multiple logistic regression analysis. Each ? coefficient was divided by 2 and rounded to the nearest integer. A Predicting Hyperglycemic crisis Death (PHD) score was calculated for each patient using each acceptable model. The PHD scores on each different weight model were used to determine their respective cutoff points for risk stratification by ROC curve with the highest sum of sensitivity and specificity.

The prediction rule calculated using the PHD score was then validated for the population of patients in the validation set to classify patients into risk groups according to the cutoff point set by the specific aforementioned model.

Results

We reviewed 330 patient visits (235 derivation set and 95 validation set) by 295 individual patients, about 0.09% of all ED visits in the study period. There were no significant differences between patients in the derivation and validation sets (Table 1). Thirteen univariate mortality predictors were retrieved at the criterion of P b .1 (Table 2). After a multiple logistic regression analysis, 6 independent mortality predictors were retained: Absent tachycardia, Hypotension, Anemia, Severe coma, Cancer history, and Infection as the precipitat- ing factor (AHA.SCI) (Table 3).

The interpretation of infection may be different between physi- cians, however, and the variation may affect the strength of this prediction rule. We thus compared three models using these 6 predictors with different weights on this factor. Their respective cutoff

Table 3

Calculate the PHD score (model 1): AHA.SCI Absent tachycardia x1

Hypotension x1 Anemia x1 Severe coma x1 Cancer history x1 Infection x2

If a patient meets the criteria of hyperglycemic crisis

intermediate risk (Score 3) Mortality (25.5 %)

Low risk (Score 0-2) Mortality (0 %)

ICU admission

High risk (Score ? 4) Mortality (59.5 %)

Independent mortality predictors for derivation set identified using multivariate analysis

Variable Parameters PHD score^a

Models with weights

? OR 95% CI 1 2 3

Intercept -7.629

Absent tachycardia (HR <=100/min)

1.569 4.8 1.4-17.0 1 1 1

Hypotension (SBP b 90 mmHg) 1.881 6.6 1.4-31.2 1 1 1

Anemia (Hb b10 g/dL or 2.43 11.4 2.3-55.0 1 1 1

Hct b30%) Severe coma (GCS <=8)	1.886	6.6	1.8-24.0	1	1	1
Cancer history	2.425	11.3	2.6-48.0	1	1	1
Infection as the precipitating factor	4.192	66.2	4.9-899.4	2	1	0
AUC	0.946				0.935	0.865
Sensitivity	100				100	100

OR, odds ratio; HR, heart rate; SBP, systolic blood pressure; Hb, hemoglobin; Hct, hematocrit; AUC, area under the curve.

Specificity	82.86	77.14	53.81	General ward		ICU admission
Cut-off point (N)b	2	1	0	Or		Or
Cut-off point (N)c	3	2	1	Treatment in ED		General ward admission
Possible scores in a range	0-7	0-6	0-5

^a PHD score: Predicting Hyperglycemic crisis Death score.

^b Low risk vs. intermediate+high risk.

^c Low+intermediate risk vs. high risk.

points for decision-making on risk stratification were determined using the ROC curve with the highest sum of sensitivity and specificity. The PHD score in model 1, with a heavy weight on infection, outperformed models 2 and 3 based on their specificity values (Table 3). We thus suggest that model 1 is the optimal choice. All three models have an acceptable prediction rate for classifying patients into the intermediate and high mortality-risk groups. Therefore, the validation-set PHD score with a high Predictive power (AUC = 0.925, 95% confidence interval [CI]: 0.870-0.979) is acceptable compared with model 1 (AUC = 0.946, 95% CI: 0.917- 0.975) and model 2 (AUC = 0.935, 95% CI: 0.901-0.970) (Table 3). The

number of patients (in both data sets) based on the PHD score in model 1 can be classified into three mortality-risk groups: low = 0% (0/242, 95% CI: 0-0.02%), intermediate = 24.5% (13/51, 95% CI: 14.8-

39.9%), and high = 59.5% (22/37, 95% CI: 42.2-74.8%) (Fig. 2).

Discussion

We developed a novel decision rule to predict 30-day mortality and manage adult ED patients with hyperglycemic crises. ED and ICU physicians can usefully evaluate 6 variables. Patients with a high-risk PHD score should be deemed critically ill and sent to the ICU for advanced treatment such as aggressive fluid resuscitation, strict Intravenous insulin control, detailed investigation and management of the precipitating factors, and careful prevention of possible treatment complications. For patients with an intermediate-risk score, a transfer to the ICU or a general ward depends on the treating physician’s decision based on the each patient’s condition and on available medical resources. For patients with low-risk score, a general ward admission or ED treatment may be sufficient, which would help preserve medical resources for patients in greater need. These logical validated cutoff points are illustrated in Fig. 2. The mortality risk is calculated from the total patients in both data sets.

The predictor “infection as the precipitating factor” is difficult for physicians in some circumstances. For those who are wary of misclassifying patients by using these predictors, model 3 with zero weight on this factor may be an alternative (Table 3).

The multiple logistic regression analysis identified 6 independent correlates of mortality. Infection, with a high ?-value, was the strongest mortality predictor. We found that the most common

Fig. 2. A suggested disposition flowchart based on the PHD score (model 1) for ED patients with hyperglycemic crises. The mortality risk is calculated from the total patients in both data sets.

sources were urinary tract infection (49.3%), lower respiratory tract infection (26.8%), and skin or soft tissue infection (12.0%). Infection should always be suspected in every patient with a hyperglycemic crisis, because not detecting an infection would be a catastrophe [17]. More attention is needed to evaluate elderly patients and patients with long-term diabetes, in whom infection may not be apparent [1]. If test results for infection are equivocal after an evaluation in the ED, immediate blood culture and possibly Empiric antibiotics should be considered [17]. Anemia has been used as a mortality predictor in the APACHE (Acute Physiology and Chronic Health Evaluation) and ODIN (Organ Dysfunctions and/or Infection) scores for critically ill patients [14,18]. In this study, the common cause of anemia was Upper gastrointestinal bleeding, cancer, Chronic renal insufficiency, and Iron deficiency anemia. Metastatic cancer has been used as a mortality predictor in ICU Mortality scores such as MPM (Mortality Probability Model), SAPS (Simplified APACHE Score), and APACHE [19]. In this study, “cancer history” included the presence of any malignancy, whether metastatic or non-metastatic. Altered mental status has been proposed as the only independent mortality predictor of hyperglyce- mic crises [4]. Nevertheless, altered mental status is difficult to definitively define. In this study, severe coma (GCS <=8) instead of altered mental status was identified as an independent mortality predictor, which is easier to quantify in clinical practice. Patients with hyperglycemic crises have a severe water deficit of 100 to 200 ml/kg caused by osmotic diuresis. Tachycardia, which increases cardiac output and preserves organ perfusion, is a normal response to dehydration [1]. Absent tachycardia is also a predictor of a poor prognosis. The common causes impairing the heart rate response to Volume depletion are being elderly with blunted body functions, having long-standing neuropathy, and concomitant drug use (eg, a ?– blocker or a Calcium-channel blocker) [20].

Age has been proposed as a mortality predictor [5]. While we

found that being elderly was a predictor in the univariate analysis, it lost significance in a multivariate analysis. In clinical practice, blood pH and effective serum osmolality are important for evaluating hyperglycemic crises, but interestingly, these two factors did not appear to be significant mortality predictors in our study. In prior studies [1,4,5], the mortality rates for the three subgroup diagnoses of hyperglycemic crises (ie, DKA, HHS, and mixed DKA/HHS syndrome) were different. We found that these three overall mortality rates were 4/104 (3.8%), 23/186 (12.4%), and 8/40 (20%), respectively. The mortality rates for HHS and mixed syndrome were significantly higher than that for DKA (P = .002), which agrees with prior findings [1,4,5].

Thirty of the 35 patients (85.7%) who died within 30 days of their hospital admission succumbed to sepsis, 1 patient (2.8%) to sepsis with an acute coronary syndrome, 1 patient to sepsis with end-stage cancer, 2 patients to hypokalemia, and 1 patient to an acute coronary syndrome. These causes of mortality underscore the importance of infection. Two patients died from hypokalemia related to ventricular arrhythmia; both were young (23 and 26 years old) women with type 1 diabetes. Their initial serum potassium was 2.1 mg/dL and 2.3 mg/ dL, respectively. This reminds us that patients with normal or low serum potassium levels have a severe total-body potassium deficien- cy. Their Cardiac functions must be carefully monitored and their potassium vigorously replaced [1]. The precipitating factor for

8 patients (2.4%) was an acute coronary syndrome, which was a cause of death for 2 patients. Patients with hyperglycemic crises may not have the typical presentation of an acute coronary syndrome. Routine screening is recommended [1].

Conclusions

The PHD score is a simple and rapid rule for predicting 30-day mortality and managing adult patients with hyperglycemic crises in the ED. The 6 factors are easy to memorize and apply in clinical practice. The PHD score can help ED and ICU physicians decide on how best to manage patients with hyperglycemic crises based on the urgency of their clinical condition. In addition, using the ratio of the actual to the expected number of deaths, this prediction rule may be used to evaluate the quality of critical care or of new clinical trials in patients with hyperglycemic crises. Despite the potential benefit of this score, it is well to remember that prognostic estimates are still only estimates. Providing critical medical care to patients always requires experienced clinical judgment and careful integration of objective data with other relevant information, such as the doctor-patient interaction and the patient’s personal intentions and requirements.

This study has several limitations. First, some data were collected from a retrospective chart review. These clinical presentations or records may not have been completely documented. Second, this was a single-center study. Findings from our database may not be generalizable to cohorts in Taiwan or in other nations. Third, the sample size might not be large enough to make conclusions with good statistical power. Additional studies with larger sample sizes are necessary. Fourth, although we have validated the prediction rule in a prospective cohort, external validation in other populations is necessary. Fifth, the interpretation of infection in an early stage may be different between physicians. In a busy ED with a short patient stay, suspected infection by the treating physician along with laboratory and image result is more practical than confirmed infection. If this is unacceptable to the treating physician, we suggest adopting model 3 as an alternative. Sixth, there might be overlaps among the 3 types of hyperglycemic crisis, but since we were dealing with all 3 types of hyperglycemic crises as a whole, the overlaps would not affect our study results. Finally, using the rule to predict the outcome for an individual patient is always different from using it to predict outcomes for a group [21].

Acknowledgments

H.C.C. and S.S.B. conceived the study concept and design, acquired data, did statistical analysis, analyzed and interpreted the data, wrote the manuscript, and reviewed and edited the manuscript. K.S.C. and

C.T.W. did statistical analysis, analyzed and interpreted data, and reviewed and edited the manuscript. L.H.J. and G.H.R reviewed and edited the manuscript. C.W.L., C.J.W., and C.S.H. acquired, analyzed, and interpreted data. H.C.C. takes responsibility for the paper as a whole. All authors read and approved the final manuscript.

References

1. Kitabchi AE, Umpierrez GE, Miles JM, et al. Hyperglycemic crises in adult patients with diabetes. Diabetes Care 2009;32:1335-43.
2. Wachtel TJ, Tetu-Mouradjian LM, Goldman DL, et al. Hyperosmolarity and acidosis in diabetes mellitus: a three-year experience in Rhode Island. J Gen Intern Med 1991;6:495-502.
3. Kitabchi AE, Umpierrez GE, Murphy MB, et al. Hyperglycemic crises in adult patients with diabetes: a consensus statement from the American Diabetes Association. Diabetes Care 2006;29:2739-48.
4. Chung ST, Perue GG, Johnson A, et al. Predictors of hyperglycaemic crises and their associated mortality in Jamaica. Diabetes Res Clin Pract 2006;73:184-90.
5. MacIsaac RJ, Lee LY, McNeil KJ, et al. Influence of age on the presentation and outcome of acidotic and hyperosmolar diabetic emergencies. Intern Med J 2002;32:379-85.
6. Magee MF, Bhatt BA. Management of decompensated diabetes. Diabetic ketoacidosis and hyperglycemic hyperosmolar syndrome. Crit Care Clin 2001;17:75-106.
7. Carroll P, Matz R. uncontrolled diabetes mellitus in adults: experience in treating diabetic ketoacidosis and hyperosmolar nonketotic coma with low-dose insulin and a uniform Treatment regimen. Diabetes Care 1983;6:579-85.
8. National Center for Health Statistics. National hospital discharge and ambulatory surgery data [article online]. http://www.cdc.gov/nchs/about/major/hdasd/nhds. htm. Accessed 24 January 2009.
9. Fishbein HA, Palumbo PJ. Acute metabolic complications in diabetes. In: National Diabetes Data Group. Diabetes in America. Bethesda (MD): National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases; 1995.

   p. 283-92.

   Wang J, Williams DE, Narayan KM, et al. Declining death rates from hyperglycemic crisis among adults with diabetes, U.S., 1985-2002. Diabetes Care 2006;29: 2018-22.

10. Efstathiou SP, Tsiakou AG, Tsioulos DI, et al. A Mortality prediction model in diabetic ketoacidosis. Clin Endocrinol (Oxf) 2002;57:595-601.
11. Kitabchi AE, Umpierrez GE, Murphy MB, et al. Management of hyperglycemic crises in patients with diabetes. Diabetes Care 2001;24:131-53.
12. Teasdale G, Jennett B. Assessment of coma and impaired consciousness. A practical

    scale. Lancet 1974;2:81-4.

    Knaus WA, Draper EA, Wagner DP, et al. APACHE II: a Severity of disease classification system. Crit Care Med 1985;13:818-29.

13. Bone RC, Sibbald WJ, Sprung CL. The ACCPSCCM consensus conference on sepsis and organ failure. Chest 1992;101:1481-3.
14. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modifications of methodological standards. JAMA 1997;277:488-94.
15. Tintinalli JD, Kelen GD, Stapczynski JS, editors. Emergency medicine: a comprehensive study guide. 6th ed. New York: McGraw-Hill; 2004. p. 1294-304.
16. Fagon JY, Chastre J, Novara A, et al. Characterization of intensive care unit patients using a model based on the presence or absence of organ dysfunctions and/or infection: the ODIN model. Intensive Care Med 1993;19:137-44.
17. Lemeshow S, Le Gall JR. Modeling the severity of illness of ICU patients. JAMA 1994;272:1049-55.
18. Gaglia JL, Wyckoff J, Abrahamson MJ. Acute hyperglycemic crisis in the elderly. Med Clin North Am 2004;88:1063-84.
19. Wasson JH, Sox HC, Neff RK, et al. clinical prediction rules. Applications and methodological standards. N Engl J Med 1985;313:793-9.