bs_bs_banner
Admission blood glucose predicts mortality and length of stay in patients admitted through the emergency department W. G. Martin,1 J. Galligan,2 S. Simpson Jr,3 T. Greenaway4,5 and J. Burgess5,6 1
Department of Medicine, St Vincent’s Hospital Melbourne, Melbourne, Victoria, 2Department of Medicine, 4Clinical Endocrinology, 6Endocrinology Laboratory, Royal Hobart Hospital, 3Menzies Institute for Medical Research and 5School of Medicine, University of Tasmania, Hobart, Tasmania, Australia
Key words diabetes, stress hyperglycaemia, glycaemic variability, length of stay, mortality. Correspondence William G. Martin, Department of Medicine, St Vincent’s Hospital Melbourne, 41 Victoria Parade, Fitzroy, Melbourne, Vic. 3065, Australia. Email: [email protected] Received 6 January 2015; accepted 3 June 2015. doi:10.1111/imj.12841
Abstract Background: Hyperglycaemia has been associated with adverse outcomes in several different hospital populations. Aim: The aim of this study was to investigate the relationship between admission blood glucose level (BGL) and outcomes in all patients admitted through the emergency department. Methods: This study was a retrospective observational cohort study from an Australian tertiary referral hospital. Patients admitted in the first week of each month from April to October 2012 had demographic data, co-morbidities, BGL, intensive care unit admission, length of stay and dates of death recorded. Factors associated with outcomes were assessed by multi-level mixed-effects linear regression. Results: Admission BGL was recorded for 601 admissions with no diagnosis of diabetes and for 219 admissions diagnosed with type 2 diabetes (T2DM). In patients with no diagnosis of diabetes, admission BGL was associated with in-hospital and 90-day mortality (P < 0.001). After multivariate analysis, BGL greater than 11.5 mmol/L was significantly associated with increased mortality at 90 days (P < 0.05). In patients with T2DM increased BGL on admission was not associated with in-hospital or 90-day mortality but was associated with length of hospital stay (β: 0.22 days/mmol/L; 95% confidence interval 0.09–0.35; P < 0.001), although this association was lost on multivariable analysis. In patients with T2DM, increased coefficient of variation of BGL was also positively associated with length of hospital stay in an almost dose-dependent fashion (P < 0.001). Conclusion: Admission BGL was independently associated with increased mortality in patients with no diagnosis of diabetes. Glycaemic variability was associated with increased length of hospital stay in patients with T2DM.
Introduction Hyperglycaemia has been associated with a number of adverse outcomes in certain hospital populations, including patients with acute myocardial infarction,1 community-acquired pneumonia,2 stroke,3 chronic obstructive pulmonary disease (COPD),4 trauma5 and pulmonary embolism.6 The effects of glycaemic control on clinical outcomes have been examined extensively in patients during critical illness.7 However, less work has been done examining outcomes in the general ward population.
Funding: None. Conflict of interest: None.
In a landmark study, Umpierrez et al. demonstrated an almost 10-fold increase in mortality in patients with no diagnosis of diabetes if they had a documented episode of hyperglycaemia during a hospital admission, as well as a similar but more modest increase in mortality in patients with a diagnosis of diabetes.8 Another large prospective study from Australia further showed a stepwise increase in mortality with increasing blood glucose level (BGL) in patients with no diagnosis of diabetes and identified a cut-off for this increase in mortality at 8 mmol/L. This study did not demonstrate an association between hyperglycaemia and increased mortality in patients with type 2 diabetes (T2DM).9 The potential significance of hypoglycaemia was highlighted in the large and wellknown trial by the Normoglycemia in Intensive Care Evaluation–Survival Using Glucose Algorithm Regulation © 2015 Royal Australasian College of Physicians
916
BGL, mortality and length of stay
(NICE-SUGAR) study, which showed intensive glucose control in intensive care unit (ICU) patients to be associated with increased rates of hypoglycaemia and increased mortality.10 An association between hyperglycaemia and length of admission has been shown in several studies.11–15 Tonks et al. found a significantly increased length of stay in all patients, including those without known diabetes who had documented hyperglycaemia during admission, when compared with controls.14 Increased length of stay has also been shown in more specific populations including patients with heart failure11 or those receiving bone marrow transplantation.12 More recently the association between admission BGL and length of stay has been evaluated in general inpatient settings in the UK and Brazil.8,13 These studies combined patients with type 1 diabetes (T1DM) and T2DM when assessing outcomes such as length of stay despite these two groups having many dissimilar clinical characteristics. The neuro-hormonal response to illness may result in elevated glucose levels; this elevation is sometimes referred to as ‘stress hyperglycaemia’ in those individuals with underlying insulin resistance or B cell insufficiency. A number of neuro-hormonal mechanisms contribute to elevated BGL in patients with severe illness, including hypothalamic–pituitary–adrenal axis activation, catecholamine release, insulin resistance and release of inflammatory mediators. Additionally, ill patients may be given exogenous glucocorticoids, vasopressors and total parenteral nutrition, which all contribute to hyperglycaemia.16 Thus hyperglycaemia in hospital is at least in part a manifestation of how sick a patient is.
Aims The aim of our study was to investigate the relationship of dysglycaemia with clinical outcomes including length of stay, ICU admission and death within 90 days in all patients including those with T2DM and no diagnosis of diabetes admitted through the emergency department of an Australian tertiary referral hospital. Our a priori hypothesis was admission hyperglycaemia would be associated with increased mortality and length of hospital stay in patients with and without a diagnosis of diabetes. In patients with T2DM we hypothesised that elevated mean BGL and glycaemic variability would also be associated with adverse outcomes.
Methods Included patients were admitted through the emergency department of the Royal Hobart Hospital, a 490-bed tertiary referral hospital servicing the population of southern
Tasmania. Admissions were during the first week of each month from April to October 2012. Weeks were separated to limit the effect of staffing on patient outcomes. Lists of inpatient episodes were generated by the hospital’s discharge coding system. Two house medical officers manually reviewed records to ascertain clinical information. Exclusion criteria were admission to the emergency department short stay unit (N = 202), patients under the age of 18 (N = 274), a diagnosis of T1DM (N = 20), a new diagnosis of T1DM (N = 1), duplicated coding (N = 1) and inadequate electronic records (N = 1). Medical records were reviewed retrospectively, and demographic variables and major co-morbidities were recorded. Co-morbidities were taken from admission notes and discharge coding; drug charts were also reviewed for presence of oral hypoglycaemic agents or insulin therapy. Where there was a conflict between drug therapy and the progress notes, previous correspondence and admissions were reviewed to clarify the diagnosis. Variables recorded were diabetes status, ischaemic heart disease, congestive cardiac failure, chronic kidney disease (CKD), COPD, stroke or transient ischaemic attack (TIA), active malignancy within the previous 12 months (excluding localised non-melanoma skin cancer) and dementia. BGL were measured as a part of routine care by venous blood gas, finger-prick BGL or serum glucose testing. The first blood glucose reading performed after the patient had been admitted was recorded, and it was documented if this was taken while in the emergency department or within 24 h of admission. All subsequent BGL readings were recorded and analysed for patients with a diagnosis of T2DM. Patients who had greater than two BGL recordings during admission and greater than one BGL per day had total mean BGL calculated. Coefficient of variation (CV) of all BGL readings was calculated by dividing standard deviation (SD) of BGL by mean BGL and was used as a marker of blood glucose variability. Patients admitted more than once were counted as separate episodes for comparison with outcome measure; however, gender and count of co-morbidities were presented as number of absolute patients. Standard measurement of BGL readings in diabetic patients is four times per day (pre-meals and pre-bed). Our local policy for inpatient management of hyperglycaemia in patients with diabetes is basal insulin with supplementary insulin at meal times, although adherence to this policy is variable. Outcomes recorded included length of stay, ICU admission and mortality. Mortality data were obtained from the hospital record, which receives notifications of all deaths within the state. Patients with no diagnosis of diabetes were compared with those with known T2DM. For the purposes of analysis patients newly diagnosed with T2DM were grouped with diagnosis of T2DM. We hypothesised that admission
© 2015 Royal Australasian College of Physicians
917
Martin et al.
hyperglycaemia is associated with increased mortality and length of hospital stay in patients with and without diabetes. Ethics approval was obtained from the Southern Tasmania Human Research Ethics Committee.
Statistical analysis Predictors of BGL, ICU admission and duration of admission were assessed by multi-level mixed-effects linear regression.17 This is appropriate for repeated measure data and allows fitting of random and fixed effects, accounting for some of the variability in parameters measured. As the data were highly skewed, dependent variables were transformed using the optimal transform as identified by Box– Cox regression models18 to reduce heteroskedasticity, although all regression coefficients are reported on the original scale. Where data have been back-transformed to the original scale, these data are reported as the geometric mean, which is a better measure of central tendency than the arithmetic mean in these circumstances. Predictors of death in-hospital and death within 90 days were assessed by log-binomial regression. All analyses were performed using STATA/SE for Windows (Version 12.1; StataCorp LP, College Station, TX, USA).
Results After excluding ineligible patients, 1306 patients admitted through the emergency department were enrolled
with a total of 1394 admissions (Fig. 1). There were 1101 patients with no diagnosis of diabetes. There were 210 patients with a previous diagnosis of T2DM, including 5 patients who were diagnosed with diabetes during their admission. Overall, 134 patients died within 90 days of admission. The largest proportions of admissions were under a general medical unit (45.9%). The mean age of all patients was 59.8 years (SD: 20.5) and 53.7% were male. Table 1 summarises the demographic data of all patients.
Patients with no diagnosis of diabetes There were 1101 patients (1169 admissions) who had no diagnosis of diabetes; of these, there were 107 (9.2%) deaths within 90 days of admission. Of the admissions, 51.4% had a BGL reading performed within 24 h of admission, most of these (n = 546, 90.8%) in the emergency department. Patients who died in hospital had a statistically significantly higher admission BGL when compared with those who survived (7.44 vs 6.22 mmol/L; P < 0.001; Table 1). Admission BGL was significantly associated with mortality at 90 days (P = 0.008), showing evidence of a threshold effect for BGL above 9.4 mmol/L (Fig. 2) (odds ratio (OR): 9.5+ vs ≤9.4 mmol: 6.28, P < 0.001). A total of 50 patients was admitted to ICU (4.3%). Patients without diagnosed diabetes admitted to ICU had a significantly higher admission BGL than those not admitted to ICU (7.03 vs
All admissions N = 1893
Included: Admissions N = 1394 Patients N = 1306
No DM: Admissions N = 1169 Patients N = 1101
BGL in 24 hours: Admissions N = 601
No BGL: Admissions N = 568
Exclusions N = 499 Age 34–61 years >61–77 years >77–97 years ICU admission Death in 90 days n Male sex IHD CCF CKD COPD Stroke/TIA Malignancy Dementia Number of co-morbidities 0 1 2 3 4 5 or more Length of stay (median, IQR) BGL readings per day (Median IQR) Admission BGL (n) Admission BGL (geometric mean, 95% CI) Alive at discharge In-hospital death
1394 349 (25.0) 345 (24.8) 357 (25.6) 343 (24.6) 62 (4.5) 134 (9.6) 1306 607 (46.5) 296 (22.7) 107 (8.2) 95 (7.3) 165 (12.6) 146 (11.2) 133 (10.2) 71 (5.4)
No diabetes n (%)
1169 342 (29.3) 301 (25.8) 267 (22.8) 259 (22.2) 50 (4.3) 107 (9.2) 1101 581 (52.8) 212 (19.3) 74 (6.7) 53 (4.8) 133 (12.1) 104 (9.5) 106 (9.6) 52 (4.7)
619 (47.4) 361 (27.6) 190 (14.6) 91 (7.0) 36 (2.8) 14 (1.1) 4, 2–8 N/A 820 (58.8)
Admitted to ICU Not admitted to ICU Total Average BGL (geometric mean, 95% CI) Alive at discharge In-hospital death Admitted to ICU Not admitted to ICU
T2DM n (%)
225 7 (3.1) 44 (19.6) 90 (40.0) 84 (37.3) 12 (5.3) 27 (12.0) 210 120 (57.1) 86 (41.0) 35 (16.7) 43 (20.5) 32 (15.2) 44 (21.0) 29 (13.8) 19 (9.1)
619 (56.2) 303 (27.5) 120 (10.9) 47 (4.3) 10 (0.9) 2 (0.2) 4, 2–7 N/A 601 (51) No diabetes 6.2 (6.1–6.4) 7.5 (6.9–8.4) P < 0.001 6.2 (6.1–6.4) 7.2 (6.4–7.9) P = 0.0021
0 (0) 58 (27.6) 70 (33.3) 44 (21.0) 26 (12.4) 12 (5.7) 5, 3–9 3.21 (2.50–4.00) 219 (97) T2DM 8.5 (8.1–8.9) 9.1 (7.1–11.2) P = 0.66 8.4 (8.0–8.8) 10.3 (8.0–12.7) P = 0.22
N/A
8.4 (8.1, 8.7) 8.5 (7.0, 9.9) P = 0.94 8.4 (8.0–8.6) 9.4 (7.9, 10.9) P = 0.18
Difference between T2DM and no diabetes
P < 0.001 P = 0.48 P = 0.19 P = 0.14 P < 0.001 P < 0.001 P < 0.001 P = 0.36 P < 0.001 P = 0.055 P = 0.003
P < 0.001
P = 0.14 N/A
P < 0.001 P = 0.16
P < 0.001 P = 0.012
Results for continuous variables assessed by two tailed t-test, and dichotomous variables were assessed by Chi-squared test. Results in bold are statistically significant (P < 0.05) when compared with the reference group. Denominators vary between groups because of patients with multiple admissions, where denominators change a row with the total number of subjects has been included. Variables that did not change between admissions (e.g. sex and co-morbidities) were recorded once. Variables that changed between admissions (e.g. age, admission duration, admission BGL) were recorded for each admission. Co-morbidity count is the count of IHD, CKD, CCF, COPD stroke/TIA, malignancy and dementia per patient. BGL, blood glucose level; CCF, congestive cardiac failure; CI, confidence interval; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; IHD, ischaemic heart disease; IQR, interquartile range; N/A, not applicable; TIA, transient ischaemic attack.
6.23 mmol/L P = 0.006; Table 1). While associated with mortality, admission BGL was not a significant predictor of length of stay among those with no diagnosis of diabetes (P = 0.72).
Patients with T2DM There were 210 patients (225 admissions) with T2DM (16.1% of all patients). Of these, the mean age at admis-
© 2015 Royal Australasian College of Physicians
919
Martin et al.
9 8
6
Admission duration (95% CI)
Risk ratio of death at 90 days (95% CI)
A 7
5 4 3 2
0.19−0.25
>0.25−0.34
>0.34−0.67
Covariance of BGL during admission
BGL at admission
B 7 Risk ratio of death at 90 days (95% CI)
6
3
1 0
Figure 3 Coefficient of variation of blood glucose level (BGL) during admission (mmol/L) versus length of hospital stay in patients with type 2 diabetes (P < 0.001). CI, confidence interval.
6 5 4 3 2 1 0 61–77 years >77–97 years Trend: Male sex
343 (29.2) 304 (25.6) 269 (22.9) 259 (22.0) 558 (47.5)
IHD
222 (18.9)
CCF
79 (6.7)
CKD
60 (5.1)
COPD
144 (12.3)
Stroke/TIA
108 (9.2)
Malignancy
113 (9.7)
Dementia Number of co-morbidities 0 1 2 3 4 5 or more Trend: ICU admission
52 (4.4)
663 (56.4) 323 (27.5) 128 (10.9) 50 (4.3) 9 (0.8) 2 (0.2) 50 (4.3)
Admission BGL (median, IQR)
6.1 mmol/L (5.4–7.4)
Admission BGL range 34–61 years >61–77 years >77–97 years Trend: Male sex
7 (3.2) 41 (18.6) 88 (40.0) 84 (38.2)
1.00 [Reference] 0.88 (0.10–7.92) 1.32 (0.25–7.14) 2.11 (0.41–10.9) P = 0.16 Not included in analysis
1.00 [Reference] 1.63 (0.29–9.25) 5.76 (1.20–27.7) 6.91 (1.30–36.9) 21.2 (3.08–145.9) 26.5 (1.18–594.5) P < 0.001 0.64 (0.08–5.13) P = 0.67 1.01 (0.93–1.10) P = 0.82
1.00 [Reference] 1.54 (0.25–9.57) 5.34 (0.96–29.7) 6.94 (1.12–43.1) 17.4 (2.36–128) 17.8 (0.65–489) P < 0.001 Not included in analysis
1.00 [Reference] 1.64 (0.40–6.66) 0.84 (0.17–4.06) 0.64 (0.10–4.02) 1.76 (0.44–7.07) P = 0.71 0.93 (0.78–1.12) P = 0.46
Not included in analysis
2 (11.1) 3 (16.7) 9 (50.0) 1 (5.6) 3 (16.7)
1.00 [Reference] 0.13 (0.02–1.23) 0.37 (0.06–2.32) 0.08 (0.01–1.11) 0.21 (0.03–1.65) P = 0.50
Not included in analysis
3 (16.7) 5 (27.80 6 (33.3) 4 (22.2)
1.00 [Reference] 1.45 (0.31–6.66) 1.44 (0.34–6.10) 1.26 (0.26–6.09) P = 0.80
Not included in analysis
94 (42.7) 90 (40.9)
CCF
39 (17.7)
CKD
45 (20.5)
COPD
32 (14.6)
Stroke/TIA
47 (21.4)
Malignancy
31 (14.1)
Dementia
21 (9.6)
Admission BGL (median, IQR) Admission BGL 0.25–0.34 >0.34–0.67
Multivariable†,‡
1.00 [Reference] 0.31 (0.02–3.94) 0.68 (0.07–6.34) 1.30 (0.15–11.7) P = 0.067 0.91 (0.40–2.07) P = 0.82 1.97 (0.87–4.43) P = 0.10 2.20 (0.89–5.47) P = 0.090 2.18 (0.91–5.25) P = 0.082 1.40 (0.49–4.00) P = 0.53 0.82 (0.29–2.29) P = 0.70 7.33 (3.00–17.9) P < 0.001 1.80 (0.56–5.82) P = 0.33
IHD
Number of co-morbidities 0 1 2 3 4 5 or more Trend: ICU admission
RR (95% CI) univariable
55 (25.0) 69 (31.4) 56 (25.5) 29 (13.2) 9 (4.1) 2 (0.9) 12 (5.5) 8.5 mmol/L (6.4–11.5)
30 (14.0) 52 (24.3) 47 (22.0) 30 (14.0) 55 (25.7) 8.43 (6.95–9.44)
2.14 (0.81–5.60) P = 0.12 1.60 (0.60–4.28) P = 0.35 2.16 (0.82–5.70) P = 0.12 Not included in analysis Not included in analysis 9.03 (3.16–25.8) P < 0.001 Not included in analysis
Not included in analysis
Not included in analysis
Results in bold are statistically significant (P < 0.05) when compared with the reference group. †Multivariable models adjusted for age, and co-morbidities or co-morbidity count. ‡Multivariable models including co-morbidities are in two fashions: one where individual co-morbidity dichotomous terms are included, and one where the co-morbidity count variable is included, these appropriate to where the analyses are adjusted for individual co-morbidities or co-morbidity count. BGL, blood glucose level; CCF, congestive cardiac failure; CI, confidence interval; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; IHD, ischaemic heart disease; IQR, interquartile range; RR, relative risk; SD, standard deviation; TIA, transient ischaemic attack.
© 2015 Royal Australasian College of Physicians
922
BGL, mortality and length of stay
found for people with a diagnosis of diabetes for death at 90 days (P = 0.71) or death in-hospital (P = 0.53).
Characteristics of patients with no BGL recorded on admission Table S1 compares patients who had a BGL within 24 h of admission with patients who did not have a BGL performed. Patients who had BGL testing were significantly older and had a greater burden of co-morbidities. They were also more likely to be admitted to ICU (6.6% vs 1.5%) and were twice as likely to die within 90 days of admission (12.6% vs 5.3%, P ≤ 0.001).
Discussion Our study’s main finding is that admission blood glucose predicts inpatient mortality in patients with no diagnosis of diabetes, although only when BGL is >11.5 mmol/L was BGL significantly associated with 90-day mortality after correction for age and co-morbidities. The most novel finding of our study is that in patients with a diagnosis of T2DM, CV of all inpatient BGL readings was independently associated with increased length of hospital stay within a heterogeneous population of emergency department admissions. Length of stay was associated with admission BGL and mean BGL, but not after correction for age and co-morbidities. Previous studies examining the relationship between length of stay and BGL in the general inpatient setting have had several limitations. For example, hyperglycaemia has been examined as a dichotomous rather than a continuous variable. Admission BGL is often loosely defined as BGL up to 72 h post-admission.15 In another study of general medical admissions, the authors did not distinguish between patients with and without a diagnosis of diabetes when evaluating the associations of BGL with clinical outcomes,13 despite patients with no diagnosis of diabetes having been shown to have significantly worse outcomes.8,9 Many studies have also evaluated combined cohorts of patients with diabetes rather than evaluating patients with type 1 and type 2 separately, despite these two cohorts having many dissimilar clinical attributes. An association between glycaemic variability and length of stay has only been shown in one other retrospective review of inpatient glycaemic control in a noncritically ill population.17 This study by Mendez et al.19 had a number of factors that limit its generalisability; participants were an older and nearly exclusively male (Veterans’ Affairs) population. The authors included all patients with two BGL readings per day, leaving them with a population that included a relatively large number
of patients without a diagnosis of diabetes (14.3%) whom for reasons that were not stated had an average of more than two BGL readings per day. Glycaemic variability is being increasingly recognised as an important predictor of outcomes in both critically and non-critically ill patients. The mechanism of increased glycaemic variability leading to adverse events is not entirely clear. Work done by El-Osta et al.20 has shown that even very brief episodes of hyperglycaemia can cause pervasive epigenetic changes that are known to lead to diabetic complications.18 It is well documented that hypoglycaemia and hyperglycaemia are independently associated with adverse outcomes and so increased oscillations between the two would be expected to contribute to adverse outcomes. Despite the increasing focus on glycaemic variability, it is still not a readily available or a well-utilised metric by treating clinicians. In time it is possible that increased glycaemic variability will become as important as absolute BGL to treating clinicians. This paper cannot assess the underlying mechanism by which glycaemic variability is associated with length of stay. However, underlying disease severity may influence glucose homeostasis. Cheung et al. performed a large study to identify the point at which a significant elevation in hospital mortality occurs. They found that patients with a blood sugar of 8.0–9.9 mmol/L had increased in-hospital mortality and that as BGL increased above that level, so did the risk of death.9 Our study’s results concur with this work, finding increased mortality with increased BGL in patients with no diagnosis of diabetes, and at a similar threshold of effect (9.4 mmol/L). The lack of association between BGL seen in our study may be due to the already high rates of mortality seen in patients with T2DM or our small sample size. There was no association found between admission BGL and length of hospital stay in patients with no diagnosis of diabetes. One possible explanation for this effect could be the very high rate of in-hospital mortality seen by patients with elevated BGL on admission. As in the paper by Cheung et al., we found no association between admission hyperglycaemia and in-hospital mortality in patients with T2DM.9 However, in a number of studies in specific populations including patients presenting with acute myocardial infarction, community acquired pneumonia, COPD hyperglycaemia on admission was associated with mortality-independent of diabetes status.1–3 This result in our study may have simply been as a result of insufficient sample size. There are several limitations of our study, including its retrospective design, relatively small numbers, the large proportion of patients without BGL recordings and that data gathered relied on admission documentation as well
© 2015 Royal Australasian College of Physicians
923
Martin et al.
as discharge coding. The measurement of BGL was also performed with various diagnostic methods including use of point-of-care finger-prick BGL testing. Most but not all BGL recordings were performed in the emergency department. The method of glycaemic control was not recorded because of the varied methods of documentation. Approximately half of patients with no diagnosis of diabetes did not have an admission BGL recording, potentially introducing bias, as patients who had BGL testing were older, sicker, more likely to be admitted to ICU and to die within 90 days (Table S1).
Conclusion
ity was associated with increased length of hospital stay in patients with T2DM. The relationship between blood glucose variability and length of stay may be in part due to methods of in-hospital glycaemic control, for example use of sliding scale insulin; however, this was not directly addressed in our study. The importance of checking blood glucose on admission cannot be overstated. Admission BGL is a good census point to identify diabetic patients who will go on to have poor glycaemic control and longer admissions. It also provides useful prognostic information in non-diabetic patients. Despite its utility, admission BGL are still only performed in half of the patients admitted who have no diagnosis of diabetes.
Admission BGL was associated with increased mortality in patients with no diagnosis of diabetes. Glycaemic variabil-
References 1 Gholap NN, Mehta RL, Ng L, Davies MJ, Khunti K, Squire IB. Is admission blood glucose concentration a more powerful predictor of mortality after myocardial infarction than diabetes diagnosis? A retrospective cohort study. BMJ Open 2012; 2: e001596. 2 Lepper PM, Ott S, Nuesch E, Von Eynatte M, Schuma C, Pletz MW et al. Serum glucose levels for predicting death in patients admitted to hospital for community acquired pneumonia: prospective cohort study. BMJ 2012; 344: e3397. 3 Williams LS, Rotich J, Qi R, Fineberg N, Espay A, Bruno A et al. Effects of admission hyperglycemia on mortality and costs in acute ischemic stroke. Neurology 2002; 59: 67–71. 4 Baker EH, Janaway CH, Philips BJ, Brennan AL, Baines DL, Wood DM et al. Hyperglycaemia is associated with poor outcomes in patients admitted to hospital with acute exacerbations of chronic obstructive pulmonary disease. Thorax 2006; 61: 284–9. 5 Yendamuri S, Fulda GJ, Tinkoff GH. Admission hyperglycemia as a prognostic indicator in trauma. J Trauma 2003; 55: 33–8. 6 Scherz N, Labarere J, Aujesky D, Mean M. Elevated admission glucose and mortality in patients with acute pulmonary
7
8
9
10
11
12
13
embolism. Diabetes Care 2012; 35: 25–31. van den Berghe G, Wouters P, Weekers F, Verwaest C, Bruyninckx F, Schetz M et al. Intensive insulin therapy in critically ill patients. N Engl J Med 2001; 345: 1359–67. Umpierrez GE, Isaacs SD, Bazargan N, You X, Thaler LM, Kitabchi AE. Hyperglycemia: an independent marker of in-hospital mortality in patients with undiagnosed diabetes. J Clin Endocrinol Metab 2002; 87: 978–82. Cheung NW, Li S, Ma G, Crampton R. The relationship between admission blood glucose levels and hospital mortality. Diabetologia 2008; 51: 952–5. The NICE-SUGAR Study Investigators. Intensive versus conventional glucose control in critically Ill patients. N Eng/ J Med 2009; 360: 1283–97. Adigopula S, Feng Y, Babu V, Parperis K, Amoateng-Adjepong Y, Zarich S. Hyperglycemia is associated with increased length of stay and total cost in patients hospitalized for congestive heart failure. World J Cardiovasc Dis 2013; 3: 245–9. Garg R, Bhutani H, Alyea E, Pendergrass M. Hyperglycemia and length of stay in patients hospitalized for bone marrow transplantation. Diabetes 2007; 30: 993–4. Leite SA, Locatelli SB, Niece SP, Oliveira AR. Tockus D, Tosin T. Impact of hyperglycemia on morbidity and
14
15
16
17
18
19
20
mortality, length of hospitalization and rates of re-hospitalization in a general hospital setting in Brazil. Diabetol Metab Syndr 2010; 2: 49. Tonks KT, Jones GR, McGeechan K, Campbell LV. Hyperglycaemia in hospital inpatients: still a sticky situation. Intern Med J 2010; 40: 521–6. Evans NR, Dhatariya KK. Assessing the relationship between admission glucose levels, subsequent length of hospital stay, readmission and mortality. Clin Med 2012; 12: 137–9. Dungan K, Braithwaite S, Preiser JC. Stress hyperglycaemia. Lancet 2009; 373: 1798–807. Graubard BI, Korn EL. Modelling the sampling design in the analysis of health surveys. Stat Methods Med Res 1996; 5: 263–81. Box GEP, Cox DR. An analysis of transformations. J Royal Stat Soc Series B 1964; 26: 211–52. Mendez CE, Mok KT, Ata A, Tanenberg RJ, Calles-Escandon J, Umpierrez GE. Increased glycemic variability is independently associated with length of stay and mortality in non–critically Ill hospitalized patients. Diabetes Care 2013; 36: 4091–7. El-Osta A, Brasacchio D, Yao D, Pocai A, Jones PL, Roeder RG et al. Transient high glucose causes persistent epigenetic changes and altered gene expression during subsequent normoglycemia. J Exp Med 2008; 205: 2409–17.
Supporting Information Additional Supporting Information may be found in the online version of this article at the publisher’s web-site: Table S1 Comparison of patient characteristics between patients with or without a BGL tested within 24 hours. © 2015 Royal Australasian College of Physicians
924
Admission blood glucose predicts mortality and length of stay in patients admitted through the emergency department W. G. Martin,1 J. Galligan,2 S. Simpson Jr,3 T. Greenaway4,5 and J. Burgess5,6 1
Department of Medicine, St Vincent’s Hospital Melbourne, Melbourne, Victoria, 2Department of Medicine, 4Clinical Endocrinology, 6Endocrinology Laboratory, Royal Hobart Hospital, 3Menzies Institute for Medical Research and 5School of Medicine, University of Tasmania, Hobart, Tasmania, Australia
Key words diabetes, stress hyperglycaemia, glycaemic variability, length of stay, mortality. Correspondence William G. Martin, Department of Medicine, St Vincent’s Hospital Melbourne, 41 Victoria Parade, Fitzroy, Melbourne, Vic. 3065, Australia. Email: [email protected] Received 6 January 2015; accepted 3 June 2015. doi:10.1111/imj.12841
Abstract Background: Hyperglycaemia has been associated with adverse outcomes in several different hospital populations. Aim: The aim of this study was to investigate the relationship between admission blood glucose level (BGL) and outcomes in all patients admitted through the emergency department. Methods: This study was a retrospective observational cohort study from an Australian tertiary referral hospital. Patients admitted in the first week of each month from April to October 2012 had demographic data, co-morbidities, BGL, intensive care unit admission, length of stay and dates of death recorded. Factors associated with outcomes were assessed by multi-level mixed-effects linear regression. Results: Admission BGL was recorded for 601 admissions with no diagnosis of diabetes and for 219 admissions diagnosed with type 2 diabetes (T2DM). In patients with no diagnosis of diabetes, admission BGL was associated with in-hospital and 90-day mortality (P < 0.001). After multivariate analysis, BGL greater than 11.5 mmol/L was significantly associated with increased mortality at 90 days (P < 0.05). In patients with T2DM increased BGL on admission was not associated with in-hospital or 90-day mortality but was associated with length of hospital stay (β: 0.22 days/mmol/L; 95% confidence interval 0.09–0.35; P < 0.001), although this association was lost on multivariable analysis. In patients with T2DM, increased coefficient of variation of BGL was also positively associated with length of hospital stay in an almost dose-dependent fashion (P < 0.001). Conclusion: Admission BGL was independently associated with increased mortality in patients with no diagnosis of diabetes. Glycaemic variability was associated with increased length of hospital stay in patients with T2DM.
Introduction Hyperglycaemia has been associated with a number of adverse outcomes in certain hospital populations, including patients with acute myocardial infarction,1 community-acquired pneumonia,2 stroke,3 chronic obstructive pulmonary disease (COPD),4 trauma5 and pulmonary embolism.6 The effects of glycaemic control on clinical outcomes have been examined extensively in patients during critical illness.7 However, less work has been done examining outcomes in the general ward population.
Funding: None. Conflict of interest: None.
In a landmark study, Umpierrez et al. demonstrated an almost 10-fold increase in mortality in patients with no diagnosis of diabetes if they had a documented episode of hyperglycaemia during a hospital admission, as well as a similar but more modest increase in mortality in patients with a diagnosis of diabetes.8 Another large prospective study from Australia further showed a stepwise increase in mortality with increasing blood glucose level (BGL) in patients with no diagnosis of diabetes and identified a cut-off for this increase in mortality at 8 mmol/L. This study did not demonstrate an association between hyperglycaemia and increased mortality in patients with type 2 diabetes (T2DM).9 The potential significance of hypoglycaemia was highlighted in the large and wellknown trial by the Normoglycemia in Intensive Care Evaluation–Survival Using Glucose Algorithm Regulation © 2015 Royal Australasian College of Physicians
916
BGL, mortality and length of stay
(NICE-SUGAR) study, which showed intensive glucose control in intensive care unit (ICU) patients to be associated with increased rates of hypoglycaemia and increased mortality.10 An association between hyperglycaemia and length of admission has been shown in several studies.11–15 Tonks et al. found a significantly increased length of stay in all patients, including those without known diabetes who had documented hyperglycaemia during admission, when compared with controls.14 Increased length of stay has also been shown in more specific populations including patients with heart failure11 or those receiving bone marrow transplantation.12 More recently the association between admission BGL and length of stay has been evaluated in general inpatient settings in the UK and Brazil.8,13 These studies combined patients with type 1 diabetes (T1DM) and T2DM when assessing outcomes such as length of stay despite these two groups having many dissimilar clinical characteristics. The neuro-hormonal response to illness may result in elevated glucose levels; this elevation is sometimes referred to as ‘stress hyperglycaemia’ in those individuals with underlying insulin resistance or B cell insufficiency. A number of neuro-hormonal mechanisms contribute to elevated BGL in patients with severe illness, including hypothalamic–pituitary–adrenal axis activation, catecholamine release, insulin resistance and release of inflammatory mediators. Additionally, ill patients may be given exogenous glucocorticoids, vasopressors and total parenteral nutrition, which all contribute to hyperglycaemia.16 Thus hyperglycaemia in hospital is at least in part a manifestation of how sick a patient is.
Aims The aim of our study was to investigate the relationship of dysglycaemia with clinical outcomes including length of stay, ICU admission and death within 90 days in all patients including those with T2DM and no diagnosis of diabetes admitted through the emergency department of an Australian tertiary referral hospital. Our a priori hypothesis was admission hyperglycaemia would be associated with increased mortality and length of hospital stay in patients with and without a diagnosis of diabetes. In patients with T2DM we hypothesised that elevated mean BGL and glycaemic variability would also be associated with adverse outcomes.
Methods Included patients were admitted through the emergency department of the Royal Hobart Hospital, a 490-bed tertiary referral hospital servicing the population of southern
Tasmania. Admissions were during the first week of each month from April to October 2012. Weeks were separated to limit the effect of staffing on patient outcomes. Lists of inpatient episodes were generated by the hospital’s discharge coding system. Two house medical officers manually reviewed records to ascertain clinical information. Exclusion criteria were admission to the emergency department short stay unit (N = 202), patients under the age of 18 (N = 274), a diagnosis of T1DM (N = 20), a new diagnosis of T1DM (N = 1), duplicated coding (N = 1) and inadequate electronic records (N = 1). Medical records were reviewed retrospectively, and demographic variables and major co-morbidities were recorded. Co-morbidities were taken from admission notes and discharge coding; drug charts were also reviewed for presence of oral hypoglycaemic agents or insulin therapy. Where there was a conflict between drug therapy and the progress notes, previous correspondence and admissions were reviewed to clarify the diagnosis. Variables recorded were diabetes status, ischaemic heart disease, congestive cardiac failure, chronic kidney disease (CKD), COPD, stroke or transient ischaemic attack (TIA), active malignancy within the previous 12 months (excluding localised non-melanoma skin cancer) and dementia. BGL were measured as a part of routine care by venous blood gas, finger-prick BGL or serum glucose testing. The first blood glucose reading performed after the patient had been admitted was recorded, and it was documented if this was taken while in the emergency department or within 24 h of admission. All subsequent BGL readings were recorded and analysed for patients with a diagnosis of T2DM. Patients who had greater than two BGL recordings during admission and greater than one BGL per day had total mean BGL calculated. Coefficient of variation (CV) of all BGL readings was calculated by dividing standard deviation (SD) of BGL by mean BGL and was used as a marker of blood glucose variability. Patients admitted more than once were counted as separate episodes for comparison with outcome measure; however, gender and count of co-morbidities were presented as number of absolute patients. Standard measurement of BGL readings in diabetic patients is four times per day (pre-meals and pre-bed). Our local policy for inpatient management of hyperglycaemia in patients with diabetes is basal insulin with supplementary insulin at meal times, although adherence to this policy is variable. Outcomes recorded included length of stay, ICU admission and mortality. Mortality data were obtained from the hospital record, which receives notifications of all deaths within the state. Patients with no diagnosis of diabetes were compared with those with known T2DM. For the purposes of analysis patients newly diagnosed with T2DM were grouped with diagnosis of T2DM. We hypothesised that admission
© 2015 Royal Australasian College of Physicians
917
Martin et al.
hyperglycaemia is associated with increased mortality and length of hospital stay in patients with and without diabetes. Ethics approval was obtained from the Southern Tasmania Human Research Ethics Committee.
Statistical analysis Predictors of BGL, ICU admission and duration of admission were assessed by multi-level mixed-effects linear regression.17 This is appropriate for repeated measure data and allows fitting of random and fixed effects, accounting for some of the variability in parameters measured. As the data were highly skewed, dependent variables were transformed using the optimal transform as identified by Box– Cox regression models18 to reduce heteroskedasticity, although all regression coefficients are reported on the original scale. Where data have been back-transformed to the original scale, these data are reported as the geometric mean, which is a better measure of central tendency than the arithmetic mean in these circumstances. Predictors of death in-hospital and death within 90 days were assessed by log-binomial regression. All analyses were performed using STATA/SE for Windows (Version 12.1; StataCorp LP, College Station, TX, USA).
Results After excluding ineligible patients, 1306 patients admitted through the emergency department were enrolled
with a total of 1394 admissions (Fig. 1). There were 1101 patients with no diagnosis of diabetes. There were 210 patients with a previous diagnosis of T2DM, including 5 patients who were diagnosed with diabetes during their admission. Overall, 134 patients died within 90 days of admission. The largest proportions of admissions were under a general medical unit (45.9%). The mean age of all patients was 59.8 years (SD: 20.5) and 53.7% were male. Table 1 summarises the demographic data of all patients.
Patients with no diagnosis of diabetes There were 1101 patients (1169 admissions) who had no diagnosis of diabetes; of these, there were 107 (9.2%) deaths within 90 days of admission. Of the admissions, 51.4% had a BGL reading performed within 24 h of admission, most of these (n = 546, 90.8%) in the emergency department. Patients who died in hospital had a statistically significantly higher admission BGL when compared with those who survived (7.44 vs 6.22 mmol/L; P < 0.001; Table 1). Admission BGL was significantly associated with mortality at 90 days (P = 0.008), showing evidence of a threshold effect for BGL above 9.4 mmol/L (Fig. 2) (odds ratio (OR): 9.5+ vs ≤9.4 mmol: 6.28, P < 0.001). A total of 50 patients was admitted to ICU (4.3%). Patients without diagnosed diabetes admitted to ICU had a significantly higher admission BGL than those not admitted to ICU (7.03 vs
All admissions N = 1893
Included: Admissions N = 1394 Patients N = 1306
No DM: Admissions N = 1169 Patients N = 1101
BGL in 24 hours: Admissions N = 601
No BGL: Admissions N = 568
Exclusions N = 499 Age 34–61 years >61–77 years >77–97 years ICU admission Death in 90 days n Male sex IHD CCF CKD COPD Stroke/TIA Malignancy Dementia Number of co-morbidities 0 1 2 3 4 5 or more Length of stay (median, IQR) BGL readings per day (Median IQR) Admission BGL (n) Admission BGL (geometric mean, 95% CI) Alive at discharge In-hospital death
1394 349 (25.0) 345 (24.8) 357 (25.6) 343 (24.6) 62 (4.5) 134 (9.6) 1306 607 (46.5) 296 (22.7) 107 (8.2) 95 (7.3) 165 (12.6) 146 (11.2) 133 (10.2) 71 (5.4)
No diabetes n (%)
1169 342 (29.3) 301 (25.8) 267 (22.8) 259 (22.2) 50 (4.3) 107 (9.2) 1101 581 (52.8) 212 (19.3) 74 (6.7) 53 (4.8) 133 (12.1) 104 (9.5) 106 (9.6) 52 (4.7)
619 (47.4) 361 (27.6) 190 (14.6) 91 (7.0) 36 (2.8) 14 (1.1) 4, 2–8 N/A 820 (58.8)
Admitted to ICU Not admitted to ICU Total Average BGL (geometric mean, 95% CI) Alive at discharge In-hospital death Admitted to ICU Not admitted to ICU
T2DM n (%)
225 7 (3.1) 44 (19.6) 90 (40.0) 84 (37.3) 12 (5.3) 27 (12.0) 210 120 (57.1) 86 (41.0) 35 (16.7) 43 (20.5) 32 (15.2) 44 (21.0) 29 (13.8) 19 (9.1)
619 (56.2) 303 (27.5) 120 (10.9) 47 (4.3) 10 (0.9) 2 (0.2) 4, 2–7 N/A 601 (51) No diabetes 6.2 (6.1–6.4) 7.5 (6.9–8.4) P < 0.001 6.2 (6.1–6.4) 7.2 (6.4–7.9) P = 0.0021
0 (0) 58 (27.6) 70 (33.3) 44 (21.0) 26 (12.4) 12 (5.7) 5, 3–9 3.21 (2.50–4.00) 219 (97) T2DM 8.5 (8.1–8.9) 9.1 (7.1–11.2) P = 0.66 8.4 (8.0–8.8) 10.3 (8.0–12.7) P = 0.22
N/A
8.4 (8.1, 8.7) 8.5 (7.0, 9.9) P = 0.94 8.4 (8.0–8.6) 9.4 (7.9, 10.9) P = 0.18
Difference between T2DM and no diabetes
P < 0.001 P = 0.48 P = 0.19 P = 0.14 P < 0.001 P < 0.001 P < 0.001 P = 0.36 P < 0.001 P = 0.055 P = 0.003
P < 0.001
P = 0.14 N/A
P < 0.001 P = 0.16
P < 0.001 P = 0.012
Results for continuous variables assessed by two tailed t-test, and dichotomous variables were assessed by Chi-squared test. Results in bold are statistically significant (P < 0.05) when compared with the reference group. Denominators vary between groups because of patients with multiple admissions, where denominators change a row with the total number of subjects has been included. Variables that did not change between admissions (e.g. sex and co-morbidities) were recorded once. Variables that changed between admissions (e.g. age, admission duration, admission BGL) were recorded for each admission. Co-morbidity count is the count of IHD, CKD, CCF, COPD stroke/TIA, malignancy and dementia per patient. BGL, blood glucose level; CCF, congestive cardiac failure; CI, confidence interval; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; IHD, ischaemic heart disease; IQR, interquartile range; N/A, not applicable; TIA, transient ischaemic attack.
6.23 mmol/L P = 0.006; Table 1). While associated with mortality, admission BGL was not a significant predictor of length of stay among those with no diagnosis of diabetes (P = 0.72).
Patients with T2DM There were 210 patients (225 admissions) with T2DM (16.1% of all patients). Of these, the mean age at admis-
© 2015 Royal Australasian College of Physicians
919
Martin et al.
9 8
6
Admission duration (95% CI)
Risk ratio of death at 90 days (95% CI)
A 7
5 4 3 2
0.19−0.25
>0.25−0.34
>0.34−0.67
Covariance of BGL during admission
BGL at admission
B 7 Risk ratio of death at 90 days (95% CI)
6
3
1 0
Figure 3 Coefficient of variation of blood glucose level (BGL) during admission (mmol/L) versus length of hospital stay in patients with type 2 diabetes (P < 0.001). CI, confidence interval.
6 5 4 3 2 1 0 61–77 years >77–97 years Trend: Male sex
343 (29.2) 304 (25.6) 269 (22.9) 259 (22.0) 558 (47.5)
IHD
222 (18.9)
CCF
79 (6.7)
CKD
60 (5.1)
COPD
144 (12.3)
Stroke/TIA
108 (9.2)
Malignancy
113 (9.7)
Dementia Number of co-morbidities 0 1 2 3 4 5 or more Trend: ICU admission
52 (4.4)
663 (56.4) 323 (27.5) 128 (10.9) 50 (4.3) 9 (0.8) 2 (0.2) 50 (4.3)
Admission BGL (median, IQR)
6.1 mmol/L (5.4–7.4)
Admission BGL range 34–61 years >61–77 years >77–97 years Trend: Male sex
7 (3.2) 41 (18.6) 88 (40.0) 84 (38.2)
1.00 [Reference] 0.88 (0.10–7.92) 1.32 (0.25–7.14) 2.11 (0.41–10.9) P = 0.16 Not included in analysis
1.00 [Reference] 1.63 (0.29–9.25) 5.76 (1.20–27.7) 6.91 (1.30–36.9) 21.2 (3.08–145.9) 26.5 (1.18–594.5) P < 0.001 0.64 (0.08–5.13) P = 0.67 1.01 (0.93–1.10) P = 0.82
1.00 [Reference] 1.54 (0.25–9.57) 5.34 (0.96–29.7) 6.94 (1.12–43.1) 17.4 (2.36–128) 17.8 (0.65–489) P < 0.001 Not included in analysis
1.00 [Reference] 1.64 (0.40–6.66) 0.84 (0.17–4.06) 0.64 (0.10–4.02) 1.76 (0.44–7.07) P = 0.71 0.93 (0.78–1.12) P = 0.46
Not included in analysis
2 (11.1) 3 (16.7) 9 (50.0) 1 (5.6) 3 (16.7)
1.00 [Reference] 0.13 (0.02–1.23) 0.37 (0.06–2.32) 0.08 (0.01–1.11) 0.21 (0.03–1.65) P = 0.50
Not included in analysis
3 (16.7) 5 (27.80 6 (33.3) 4 (22.2)
1.00 [Reference] 1.45 (0.31–6.66) 1.44 (0.34–6.10) 1.26 (0.26–6.09) P = 0.80
Not included in analysis
94 (42.7) 90 (40.9)
CCF
39 (17.7)
CKD
45 (20.5)
COPD
32 (14.6)
Stroke/TIA
47 (21.4)
Malignancy
31 (14.1)
Dementia
21 (9.6)
Admission BGL (median, IQR) Admission BGL 0.25–0.34 >0.34–0.67
Multivariable†,‡
1.00 [Reference] 0.31 (0.02–3.94) 0.68 (0.07–6.34) 1.30 (0.15–11.7) P = 0.067 0.91 (0.40–2.07) P = 0.82 1.97 (0.87–4.43) P = 0.10 2.20 (0.89–5.47) P = 0.090 2.18 (0.91–5.25) P = 0.082 1.40 (0.49–4.00) P = 0.53 0.82 (0.29–2.29) P = 0.70 7.33 (3.00–17.9) P < 0.001 1.80 (0.56–5.82) P = 0.33
IHD
Number of co-morbidities 0 1 2 3 4 5 or more Trend: ICU admission
RR (95% CI) univariable
55 (25.0) 69 (31.4) 56 (25.5) 29 (13.2) 9 (4.1) 2 (0.9) 12 (5.5) 8.5 mmol/L (6.4–11.5)
30 (14.0) 52 (24.3) 47 (22.0) 30 (14.0) 55 (25.7) 8.43 (6.95–9.44)
2.14 (0.81–5.60) P = 0.12 1.60 (0.60–4.28) P = 0.35 2.16 (0.82–5.70) P = 0.12 Not included in analysis Not included in analysis 9.03 (3.16–25.8) P < 0.001 Not included in analysis
Not included in analysis
Not included in analysis
Results in bold are statistically significant (P < 0.05) when compared with the reference group. †Multivariable models adjusted for age, and co-morbidities or co-morbidity count. ‡Multivariable models including co-morbidities are in two fashions: one where individual co-morbidity dichotomous terms are included, and one where the co-morbidity count variable is included, these appropriate to where the analyses are adjusted for individual co-morbidities or co-morbidity count. BGL, blood glucose level; CCF, congestive cardiac failure; CI, confidence interval; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; IHD, ischaemic heart disease; IQR, interquartile range; RR, relative risk; SD, standard deviation; TIA, transient ischaemic attack.
© 2015 Royal Australasian College of Physicians
922
BGL, mortality and length of stay
found for people with a diagnosis of diabetes for death at 90 days (P = 0.71) or death in-hospital (P = 0.53).
Characteristics of patients with no BGL recorded on admission Table S1 compares patients who had a BGL within 24 h of admission with patients who did not have a BGL performed. Patients who had BGL testing were significantly older and had a greater burden of co-morbidities. They were also more likely to be admitted to ICU (6.6% vs 1.5%) and were twice as likely to die within 90 days of admission (12.6% vs 5.3%, P ≤ 0.001).
Discussion Our study’s main finding is that admission blood glucose predicts inpatient mortality in patients with no diagnosis of diabetes, although only when BGL is >11.5 mmol/L was BGL significantly associated with 90-day mortality after correction for age and co-morbidities. The most novel finding of our study is that in patients with a diagnosis of T2DM, CV of all inpatient BGL readings was independently associated with increased length of hospital stay within a heterogeneous population of emergency department admissions. Length of stay was associated with admission BGL and mean BGL, but not after correction for age and co-morbidities. Previous studies examining the relationship between length of stay and BGL in the general inpatient setting have had several limitations. For example, hyperglycaemia has been examined as a dichotomous rather than a continuous variable. Admission BGL is often loosely defined as BGL up to 72 h post-admission.15 In another study of general medical admissions, the authors did not distinguish between patients with and without a diagnosis of diabetes when evaluating the associations of BGL with clinical outcomes,13 despite patients with no diagnosis of diabetes having been shown to have significantly worse outcomes.8,9 Many studies have also evaluated combined cohorts of patients with diabetes rather than evaluating patients with type 1 and type 2 separately, despite these two cohorts having many dissimilar clinical attributes. An association between glycaemic variability and length of stay has only been shown in one other retrospective review of inpatient glycaemic control in a noncritically ill population.17 This study by Mendez et al.19 had a number of factors that limit its generalisability; participants were an older and nearly exclusively male (Veterans’ Affairs) population. The authors included all patients with two BGL readings per day, leaving them with a population that included a relatively large number
of patients without a diagnosis of diabetes (14.3%) whom for reasons that were not stated had an average of more than two BGL readings per day. Glycaemic variability is being increasingly recognised as an important predictor of outcomes in both critically and non-critically ill patients. The mechanism of increased glycaemic variability leading to adverse events is not entirely clear. Work done by El-Osta et al.20 has shown that even very brief episodes of hyperglycaemia can cause pervasive epigenetic changes that are known to lead to diabetic complications.18 It is well documented that hypoglycaemia and hyperglycaemia are independently associated with adverse outcomes and so increased oscillations between the two would be expected to contribute to adverse outcomes. Despite the increasing focus on glycaemic variability, it is still not a readily available or a well-utilised metric by treating clinicians. In time it is possible that increased glycaemic variability will become as important as absolute BGL to treating clinicians. This paper cannot assess the underlying mechanism by which glycaemic variability is associated with length of stay. However, underlying disease severity may influence glucose homeostasis. Cheung et al. performed a large study to identify the point at which a significant elevation in hospital mortality occurs. They found that patients with a blood sugar of 8.0–9.9 mmol/L had increased in-hospital mortality and that as BGL increased above that level, so did the risk of death.9 Our study’s results concur with this work, finding increased mortality with increased BGL in patients with no diagnosis of diabetes, and at a similar threshold of effect (9.4 mmol/L). The lack of association between BGL seen in our study may be due to the already high rates of mortality seen in patients with T2DM or our small sample size. There was no association found between admission BGL and length of hospital stay in patients with no diagnosis of diabetes. One possible explanation for this effect could be the very high rate of in-hospital mortality seen by patients with elevated BGL on admission. As in the paper by Cheung et al., we found no association between admission hyperglycaemia and in-hospital mortality in patients with T2DM.9 However, in a number of studies in specific populations including patients presenting with acute myocardial infarction, community acquired pneumonia, COPD hyperglycaemia on admission was associated with mortality-independent of diabetes status.1–3 This result in our study may have simply been as a result of insufficient sample size. There are several limitations of our study, including its retrospective design, relatively small numbers, the large proportion of patients without BGL recordings and that data gathered relied on admission documentation as well
© 2015 Royal Australasian College of Physicians
923
Martin et al.
as discharge coding. The measurement of BGL was also performed with various diagnostic methods including use of point-of-care finger-prick BGL testing. Most but not all BGL recordings were performed in the emergency department. The method of glycaemic control was not recorded because of the varied methods of documentation. Approximately half of patients with no diagnosis of diabetes did not have an admission BGL recording, potentially introducing bias, as patients who had BGL testing were older, sicker, more likely to be admitted to ICU and to die within 90 days (Table S1).
Conclusion
ity was associated with increased length of hospital stay in patients with T2DM. The relationship between blood glucose variability and length of stay may be in part due to methods of in-hospital glycaemic control, for example use of sliding scale insulin; however, this was not directly addressed in our study. The importance of checking blood glucose on admission cannot be overstated. Admission BGL is a good census point to identify diabetic patients who will go on to have poor glycaemic control and longer admissions. It also provides useful prognostic information in non-diabetic patients. Despite its utility, admission BGL are still only performed in half of the patients admitted who have no diagnosis of diabetes.
Admission BGL was associated with increased mortality in patients with no diagnosis of diabetes. Glycaemic variabil-
References 1 Gholap NN, Mehta RL, Ng L, Davies MJ, Khunti K, Squire IB. Is admission blood glucose concentration a more powerful predictor of mortality after myocardial infarction than diabetes diagnosis? A retrospective cohort study. BMJ Open 2012; 2: e001596. 2 Lepper PM, Ott S, Nuesch E, Von Eynatte M, Schuma C, Pletz MW et al. Serum glucose levels for predicting death in patients admitted to hospital for community acquired pneumonia: prospective cohort study. BMJ 2012; 344: e3397. 3 Williams LS, Rotich J, Qi R, Fineberg N, Espay A, Bruno A et al. Effects of admission hyperglycemia on mortality and costs in acute ischemic stroke. Neurology 2002; 59: 67–71. 4 Baker EH, Janaway CH, Philips BJ, Brennan AL, Baines DL, Wood DM et al. Hyperglycaemia is associated with poor outcomes in patients admitted to hospital with acute exacerbations of chronic obstructive pulmonary disease. Thorax 2006; 61: 284–9. 5 Yendamuri S, Fulda GJ, Tinkoff GH. Admission hyperglycemia as a prognostic indicator in trauma. J Trauma 2003; 55: 33–8. 6 Scherz N, Labarere J, Aujesky D, Mean M. Elevated admission glucose and mortality in patients with acute pulmonary
7
8
9
10
11
12
13
embolism. Diabetes Care 2012; 35: 25–31. van den Berghe G, Wouters P, Weekers F, Verwaest C, Bruyninckx F, Schetz M et al. Intensive insulin therapy in critically ill patients. N Engl J Med 2001; 345: 1359–67. Umpierrez GE, Isaacs SD, Bazargan N, You X, Thaler LM, Kitabchi AE. Hyperglycemia: an independent marker of in-hospital mortality in patients with undiagnosed diabetes. J Clin Endocrinol Metab 2002; 87: 978–82. Cheung NW, Li S, Ma G, Crampton R. The relationship between admission blood glucose levels and hospital mortality. Diabetologia 2008; 51: 952–5. The NICE-SUGAR Study Investigators. Intensive versus conventional glucose control in critically Ill patients. N Eng/ J Med 2009; 360: 1283–97. Adigopula S, Feng Y, Babu V, Parperis K, Amoateng-Adjepong Y, Zarich S. Hyperglycemia is associated with increased length of stay and total cost in patients hospitalized for congestive heart failure. World J Cardiovasc Dis 2013; 3: 245–9. Garg R, Bhutani H, Alyea E, Pendergrass M. Hyperglycemia and length of stay in patients hospitalized for bone marrow transplantation. Diabetes 2007; 30: 993–4. Leite SA, Locatelli SB, Niece SP, Oliveira AR. Tockus D, Tosin T. Impact of hyperglycemia on morbidity and
14
15
16
17
18
19
20
mortality, length of hospitalization and rates of re-hospitalization in a general hospital setting in Brazil. Diabetol Metab Syndr 2010; 2: 49. Tonks KT, Jones GR, McGeechan K, Campbell LV. Hyperglycaemia in hospital inpatients: still a sticky situation. Intern Med J 2010; 40: 521–6. Evans NR, Dhatariya KK. Assessing the relationship between admission glucose levels, subsequent length of hospital stay, readmission and mortality. Clin Med 2012; 12: 137–9. Dungan K, Braithwaite S, Preiser JC. Stress hyperglycaemia. Lancet 2009; 373: 1798–807. Graubard BI, Korn EL. Modelling the sampling design in the analysis of health surveys. Stat Methods Med Res 1996; 5: 263–81. Box GEP, Cox DR. An analysis of transformations. J Royal Stat Soc Series B 1964; 26: 211–52. Mendez CE, Mok KT, Ata A, Tanenberg RJ, Calles-Escandon J, Umpierrez GE. Increased glycemic variability is independently associated with length of stay and mortality in non–critically Ill hospitalized patients. Diabetes Care 2013; 36: 4091–7. El-Osta A, Brasacchio D, Yao D, Pocai A, Jones PL, Roeder RG et al. Transient high glucose causes persistent epigenetic changes and altered gene expression during subsequent normoglycemia. J Exp Med 2008; 205: 2409–17.
Supporting Information Additional Supporting Information may be found in the online version of this article at the publisher’s web-site: Table S1 Comparison of patient characteristics between patients with or without a BGL tested within 24 hours. © 2015 Royal Australasian College of Physicians
924
