Secondary Logo

Journal Logo

Original Clinical Report

Independent Association of Glucose Variability With Hospital Mortality in Adult Intensive Care Patients: Results From the Australia and New Zealand Intensive Care Society Centre for Outcome and Resource Evaluation Binational Registry

Kulkarni, Hemant MD1; Bihari, Shailesh MD, FCICM, PhD2; Prakash, Shivesh MD, MPH, EDIC, FCICM2; Huckson, Sue BAppSci (Health Promotion)3; Chavan, Shaila MSPH3; Mamtani, Manju MD1; Pilcher, David MRCP, FRACP, FCICM3–5

Author Information
Critical Care Explorations: August 2019 - Volume 1 - Issue 8 - p e0025
doi: 10.1097/CCE.0000000000000025

Abstract

The putative contribution of glucose variability with adverse outcome in critically ill patients is far from established. Demonstrating this association is challenging for several reasons. First, association of higher glucose variability with mortality may be confounded by hyperglycemia (1) or hypoglycemia (2,3). Second, the measurement of glucose variability is neither straightforward nor consistent across studies. Rodbard (4,5) has elegantly reviewed important current roadblocks to measurement of glycemic variation including the novelty (and therefore the immaturity) of the field. Third, the mechanistic basis of why glucose variability would influence hospital mortality is unclear despite observed correlation with oxidative stress (6). Fourth, in addition to the factors mentioned above the observational studies that form the basis of the putative association tend to be influenced by reporting bias as demonstrated by Eslami et al (7) in a review of 12 cohort studies published around the world in nondiabetic ICU patients with stress hyperglycemia. Considering these challenges and powered by the large, binational repository of ICU patients in Australia and New Zealand, we tested the hypothesis that glucose variability is independently associated with hospital mortality in nonhyperglycemic ICU patients.

MATERIALS AND METHODS

We used the Australia and New Zealand Intensive Care Society (ANZICS) Adult Patient Database (APD, https://www.anzics.com.au/wp-content/uploads/2018/08/ANZICS-CORE-APD-Activity-Report-2016-17.pdf), one of the largest such datasets in the world with over 2 million ICU admissions (https://www.anzics.com.au/adult-patient-database-apd/). The registry has information on the sociodemographic variables, severity, comorbidity, biochemistry, and outcomes on all ICU admissions in 181 ICUs across Australia and New Zealand. This study was approved by the Institutional Review Board, University of Texas Rio Grande Valley, Brownsville, Texas, and by the ANZICS Centre for Outcome and Resource Evaluation (CORE) Management Committee.

Inclusion Criteria

Reporting diabetes (especially type 1 diabetes) was made mandatory by the ANZICS CORE Committee in 2007. Therefore, we constrained our dataset to years 2007–2016 (n = 983,555). From this, we included all the patients on whom the following data was available: lowest blood glucose level (BGL), highest BGL, hospital death, severity of illness (SOI) score baseline risk, and a glycemic status. As shown in the detailed inclusion protocol is shown in Figure 1, majority of the patients were excluded since at least one of their glucose measurements was outside the nonhyperglycemic range. Euglycemia was defined as highest BGL less than 7.78 mmol/L and lowest BGL value greater than or equal to 3.33 mmol/L, respectively whereas hypoglycemia was defined as any BGL value less than 3.33 mmol/L.

F1
Figure 1.:
Inclusion protocol. The figure shows the inclusion protocol for the final sample size of 290,066 patients. The percentages shown in the boxes on the right-hand side use the previous box on the left as the denominator. For example, for missing data on death, the percentage (0.5%) is calculated as patients with missing death data (n = 4,027) from the 796,460 patients on whom glucose measurements were available. ANZROD = Australia and New Zealand Risk of Death.

Outcomes and Predictors

The outcome of interest in this study was hospital death. In the ANZICS CORE database, the BGL measurements in the first 24 hours of admission are reported as the highest value and lowest value—entire set of measured BGL values are not available. Therefore, glucose variability was captured using glucose width which was defined as the difference between the highest and lowest BGL values. Average BGL was captured as the midpoint of the range from lowest to highest BGL values and referred to here as midpoint BGL (MBGL). The appropriateness of these two measures in the context of glucose variability was determined in a publicly available dataset of continuous glucose monitoring in 70 diabetic patients (https://archive.ics.uci.edu/ml/datasets/diabetes). Details of the dataset, the methods and results of these proof-of-concept studies are provided in Supplementary Note 1 and Supplementary Figure 1 (Supplemental Digital Content 1, https://links.lww.com/CCX/A73). SOI was estimated using Australia and New Zealand Risk of Death (ANZROD) (8), an adaptation of the Acute Physiology and Chronic Health Evaluation (APACHE) III scoring system, derived and calibrated for the Australian and New Zealand population. This model accounts for age, chronic health status, acute physiology, admission diagnosis, and additional locally available variables such as the presence of treatment limitations at admission to the ICU. To avoid confounding from glucose information already included within the overall predicted mortality risk, BGL values were regressed out of the ANZROD model. This corrected ANZROD mortality prediction (referred to as the SOI score) was then used in analyses.

Statistical Analysis

Because the dataset is contributed to by many ICUs with differing case mix and local population characteristics, all association analyses were conducted under the framework of hierarchical, mixed effects models. Specifically, we ran a series of mixed-effects logistic regression analyses wherein the ICU identifier was used as a random-effects variable. Thus, all the results are adjusted for potential inter-ICU variation. Additionally, these models permitted us to estimate the median odds ratio (MOR) and its 95% credible interval to quantify and statistically test the existence of inter-ICU variation (9). Robustness of associations was examined using sensitivity analyses for unmeasured and measured confounding. All statistical analyses were conducted using the Stata 12.0 (Stata Corp, College Station, TX) software package. Statistical significance was tested at a type I error rate of 0.05.

RESULTS

We included 290,966 nonhyperglycemic patients from 176 ICUs of whom 8% died during index hospitalization. Clinical characteristics of these patients are detailed in Table 1. Briefly, majority of the patients were 60 years old or more years, were female, generally nonobese, normotensive and included ~8% of the indigenous population with a prevalence of hypoglycemia at 2.6%. The Glasgow Coma Scale (GCS) indicated a mild affliction if any with average GCS score of 13.46 and the patients had a relatively healthy blood profile as indicated by hemoglobin concentration and blood cell counts. In general, patients who died as compared those who survived were older, less likely to have been admitted for elective surgery, more likely to have been admitted for intensive rather than high dependency care, lower GCS scores and higher white cell count (Table 1). Notably, the patients who died had very high SOI at admission as well as strikingly high APACHE III scores (Table 1).

T1
TABLE 1.:
Characteristics of the Patients Included in the Study

The average glucose width was 0.51 mmol/L (sd 0.46 mmol/L) which was higher in those who died (0.68 mmol/L) as compared with those who survived (0.49 mmol/L; Supplementary Fig. 2, Supplemental Digital Content 1, https://links.lww.com/CCX/A73). We, next, ran four hierarchical mixed effects models. Within each model, we compared the association of each quartile of glucose width with hospital death using the lowest quartile of glucose width as the reference category. The first model (column labeled Unadjusted in Table 2) shows unadjusted results. There was a stepwise increase in the odds ratio (OR) for each quartile of the glucose width from 1.13 for the second, to 1.35 for the third to 1.96 for the fourth quartile, all of which were strongly significant.

T2
TABLE 2.:
Independent Association of Glucose Concentration Variability With Hospital Mortality

Next, even though hypoglycemia was a significant predictor of death (OR, 4.44; 95% CI, 4.27–4.62; p < 1.0 × 10–317), the association of glucose width with hospital death remained significant after adjustment for hypoglycemia. Even after addition of the corrected SOI score as a covariate, the association between glucose width and hospital death remained significant—the ORs for the second, third, and fourth quartiles were 1.07, 1.19, and 1.43, respectively.

As shown in Supplementary Table 1 (Supplemental Digital Content 1, https://links.lww.com/CCX/A73), the MBGL levels varied significantly and concordantly with the quartiles of glucose width. Thus, we also corrected the association for MBGL values. We observed (last column, Table 2) that after accounting for the MBGL levels, the association of glucose width quartiles with hospital death was stronger and was similar in strength. Notably, the final model showed a significant variation in the association across ICUs. The MOR was 1.46 (95% credible interval, 1.39–1.55) indicating that the high propensity ICUs are 46% more likely to find an association than a low propensity ICU for a clinically identical patient profile.

Finally, we determined the specificity and independence of the observed association between glucose width quartiles and hospital mortality. For this, we conducted additional mixed effects modeling analyses wherein widths (difference between highest and lowest value within first 24 hr) for the following six variables were added to the last model in Table 2: hemoglobin concentration, hematocrit, systolic blood pressure, diastolic blood pressure, white cell count, and platelet count. In these analyses, we tested the hypothesis that the quartiles of glucose width continue to remain statistically significantly associated with hospital mortality with these additional covariates. As shown in Supplementary Table 2 (Supplemental Digital Content 1, https://links.lww.com/CCX/A73), our analyses indicated that the glucose width quartiles were consistently and significantly associated with hospital mortality even in the face of these covariates. Furthermore, with the exception of diastolic blood pressure variability, all other covariates were not significantly associated with hospital mortality indicating that the observed association between glucose width and hospital mortality was both independent and specific.

DISCUSSION

This is the largest study of nonhyperglycemic ICU patients that clearly demonstrates the independent and specific association of glucose variability with hospital. By design, the study eliminated a possible confounding by hyperglycemia and by way of analysis it accounted for the potential confounding by hypoglycemia. Despite additionally adjusting for a complex baseline SOI score that relies on many patient characteristics (10), for the risk of hypoglycemia, for MBGL and for inter-ICU variation, the association remained significant. These results strengthen the view that glucose variability, even in euglycemia, may be a third, orthogonal dimension of hospital mortality in ICU patients. This finding supports the increasing interest in the last decade on the importance of glucose variability in critical care (11–14). Of note, there was a significant inter-ICU variability in the observed associations which can be conceptually be explained by variability in glucose measurement methods, case-mix, treatment protocols, and annual volume of patients.

Some limitations of this study need to be recognized. First, our study presents another candidate measure of glucose variability (glucose width) which is simple and reasonably well correlated with currently used measures (as shown in Supplementary Note 1, Supplemental Digital Content 1, https://links.lww.com/CCX/A73). However, the value of this simple measure in clinical settings will need to be robustly investigated. In the ANZICS CORE database, we were constrained by the nonavailability of all glucose measurements within the first 24 hours, and therefore it was not possible to directly compare the validity of this measure against other accepted measures (15) of glucose variability. However, our proof-of-principle studies indirectly support the use of glucose width as a simple and reasonably accurate measure of glucose variability. Second, we demonstrated a significant variation across ICUs of the glucose variability → hospital mortality nexus. The factors that can contribute to this variation are currently unknown and need to be evaluated in future studies. Third, the observational nature of this study entails a possibility of measured and unmeasured confounding that can influence the interpretations. We conducted extensive sensitivity analyses to address this limitation. Our results (Supplementary Note 2 and Supplementary Tables 3–5, Supplemental Digital Content 1, https://links.lww.com/CCX/A73) indicate that the influence of confounding by factors other than hypoglycemia is likely to be minimal. The unmeasured confounding factor will need to be very strongly associated with hospital mortality (OR > 4) and highly prevalent to be able to sway the association of glucose width with hospital mortality. Furthermore, a comparison of the patients in the highest and lowest quartiles of glucose width (Supplementary Table 6, Supplemental Digital Content 1, https://links.lww.com/CCX/A73) demonstrated very similar clinical profiles except for baseline SOI which was high in the highest quartile patients. We have therefore adjusted for this confounder in the final model. Last, the ANZICS database does not record information on all the drugs (dopamine, acetaminophen, mannitol, etc.) that have been shown (16) to influence glucose measurement. Although many of these drugs are very rarely used in the ICUs in Australia and New Zealand, the information on these drugs remains unmeasured in our study.

The mechanisms contributing to glucose variability need to be investigated in future studies. Similarly, whether glucose variability over longer duration (than 24 hr of admission studied herein) will improve prediction of hospital mortality also needs to be investigated. Finally, longitudinal studies are needed before the clinical implications of our results can be translated into practice. Nevertheless, this study demonstrates an independent association of glucose variability with hospital mortality in nonhyperglycemic ICU patients.

ACKNOWLEDGMENTS

We and the Australia and New Zealand Intensive Care Society Centre for Outcome and Resource Evaluation management committee would like to thank clinicians, data collectors and researchers at the following contributing ICUs: Albury Base Hospital, Alfred Hospital, Alice Springs Hospital, Allamanda Private Hospital, Armadale Health Service, Ashford Community Hospital, Auckland City Hospital CV, Auckland City Hospital DCCM, Austin Hospital, Ballarat Health Services, Bankstown-Lidcombe Hospital, Bathurst Base Hospital, Bendigo Health Care Group, Blacktown Hospital, Box Hill Hospital, Brisbane Private Hospital, Brisbane Waters Private Hospital, Buderim Private Hospital, Bunbury Regional Hospital, Bundaberg Base Hospital, Caboolture Hospital, Cabrini Hospital, Cairns Hospital, Calvary Hospital (Canberra, ACT, Australia), Calvary Hospital (Lenah Valley, TAS, Australia), Calvary John James Hospital, Calvary Mater Newcastle, Calvary North Adelaide Hospital, Calvary Wakefield Hospital (Adelaide, SA, Australia), Campbelltown Hospital, Canberra Hospital, Central Gippsland Health Service, Christchurch Hospital, Coffs Harbour Health Campus, Concord Hospital (Sydney, NSW, Australia), Dandenong Hospital, Dubbo Base Hospital, Dunedin Hospital, Epworth Eastern Private Hospital, Epworth Freemasons Hospital, Epworth Hospital (Richmond, VIC, Australia), Figtree Private Hospital, Fiona Stanley Hospital, Flinders Medical Centre, Flinders Private Hospital, Footscray Hospital, Frankston Hospital, Fremantle Hospital, Gold Coast Private Hospital, Gold Coast University Hospital, Gosford Hospital, Gosford Private Hospital, Goulburn Base Hospital, Goulburn Valley Health, Grafton Base Hospital, Greenslopes Private Hospital, Griffith Base Hospital, Hawkes Bay Hospital, Hervey Bay Hospital, Hollywood Private Hospital, Holy Spirit Northside Hospital, Hornsby Ku-ring-gai Hospital, Hutt Hospital, Ipswich Hospital, John Fawkner Hospital, John Flynn Private Hospital, John Hunter Hospital, Joondalup Health Campus, Kareena Private Hospital, Knox Private Hospital, Latrobe Regional Hospital, Launceston General Hospital, Lismore Base Hospital, Liverpool Hospital, Logan Hospital, Lyell McEwin Hospital, Mackay Base Hospital, Macquarie University Private Hospital, Manly Hospital & Community Health, Manning Rural Referral Hospital, Maroondah Hospital, Mater Adults Hospital (Brisbane, QLD, Australia), Mater Health Services North Queensland, Mater Private Hospital (Brisbane, QLD, Australia), Mater Private Hospital (Sydney, NSW, Australia), Melbourne Private Hospital, Mersey Community Hospital, Middlemore Hospital, Mildura Base Hospital, Modbury Public Hospital, Monash Medical Centre-Clayton Campus, Mount Hospital, Mount Isa Hospital, Nambour General Hospital, Nambour Selangor Private Hospital, National Capital Private Hospital, Nelson Hospital, Nepean Hospital, Newcastle Private Hospital, Noosa Hospital, North Shore Hospital, North Shore Private Hospital, North West Regional Hospital (Burnie, TAS, Australia), Northeast Health Wangaratta, Norwest Private Hospital, Orange Base Hospital, Peninsula Private Hospital, Peter MacCallum Cancer Institute, Pindara Private Hospital, Port Macquarie Base Hospital, Prince of Wales Hospital (Sydney, NSW, Australia), Prince of Wales Private Hospital (Sydney, NSW, Australia), Princess Alexandra Hospital, Queen Elizabeth II Jubilee Hospital, Redcliffe Hospital, Repatriation General Hospital (Adelaide, SA, Australia), Robina Hospital, Rockhampton Hospital, Rockingham General Hospital, Rotorua Hospital, Royal Adelaide Hospital, Royal Brisbane and Women’s Hospital, Royal Darwin Hospital, Royal Hobart Hospital, Royal Melbourne Hospital, Royal North Shore Hospital, Royal Perth Hospital, Royal Prince Alfred Hospital, Shoalhaven Hospital, Sir Charles Gairdner Hospital, South West Healthcare (Warrnambool, VIC, Australia), Southern Cross Hospital (Hamilton, New Zealand), Southern Cross Hospital (Wellington, New Zealand), St Andrew’s Hospital (Adelaide, SA, Australia), St Andrew’s Hospital Toowoomba, St Andrew’s War Memorial Hospital, St George Hospital (Sydney, NSW, Australia) CICU, St George Hospital (Sydney, NSW, Australia), St George Hospital (Sydney, NSW, Australia) 2, St George Private Hospital (Sydney, NSW, Australia), St John Of God Health Care (Subiaco, WA, Australia), St John Of God Hospital (Ballarat, VIC, Australia), St John of God Hospital (Bendigo, VIC, Australia), St John Of God Hospital (Geelong, VIC, Australia), St John Of God Hospital (Murdoch, WA, Australia), St Vincent’s Hospital (Melbourne, VIC, Australia), St Vincent’s Hospital (Sydney, NSW, Australia), St Vincent’s Hospital (Toowoomba, QLD, Australia), St Vincent’s Private Hospital (Sydney, NSW, Australia), St Vincent’s Private Hospital Fitzroy, Sunnybank Hospital, Sunshine Hospital, Sutherland Hospital & Community Health Services, Sydney Adventist Hospital, Sydney Southwest Private Hospital, Tamworth Base Hospital, Taranaki Health, Tauranga Hospital, The Memorial Hospital (Adelaide, SA, Australia), The Northern Hospital, The Prince Charles Hospital, The Queen Elizabeth (Adelaide, SA, Australia), The Townsville Hospital, The Valley Private Hospital, The Wesley Hospital, Timaru Hospital, Toowoomba Hospital, Tweed Heads District Hospital, University Hospital Geelong, Wagga Wagga Base Hospital & District Health, Waikato Hospital, Warringal Private Hospital, Wellington Hospital, Western District Health Service (Hamilton, New Zealand), Westmead Hospital, Westmead Private Hospital, Whangarei Area Hospital, Northland Health Ltd, Wimmera Health Care Group (Horsham, VIC, Australia), Wollongong Hospital, Wollongong Private Hospital, Wyong Hospital.

REFERENCES

1. DeVries JH. Glucose variability: Where it is important and how to measure it.Diabetes2013621405–1408
2. Inzucchi SE, Umpierrez G, DiGenio A, et al. How well do glucose variability measures predict patient glycaemic outcomes during treatment intensification in type 2 diabetes?Diabetes Res Clin Pract2015108179–186
3. Kilpatrick ES, Rigby AS, Goode K, et al. Relating mean blood glucose and glucose variability to the risk of multiple episodes of hypoglycaemia in type 1 diabetes.Diabetologia2007502553–2561
4. Rodbard D. The challenges of measuring glycemic variability.J Diabetes Sci Technol20126712–715
5. Rodbard D. Glycemic variability: Measurement and utility in clinical medicine and research–one viewpoint.Diabetes Technol Ther2011131077–1080
6. Siegelaar SE, Barwari T, Kulik W, et al. No relevant relationship between glucose variability and oxidative stress in well-regulated type 2 diabetes patients.J Diabetes Sci Technol2011586–92
7. Eslami S, Taherzadeh Z, Schultz MJ, et al. Glucose variability measures and their effect on mortality: A systematic review.Intensive Care Med201137583–593
8. Paul E, Bailey M, Pilcher D. Risk prediction of hospital mortality for adult patients admitted to Australian and New Zealand intensive care units: Development and validation of the Australian and New Zealand risk of death model.J Crit Care201328935–941
9. Merlo J, Chaix B, Ohlsson H, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: Using measures of clustering in multilevel logistic regression to investigate contextual phenomena.J Epidemiol Community Health200660290–297
10. Paul E, Bailey M, Kasza J, et al. The ANZROD model: Better benchmarking of ICU outcomes and detection of outliers.Crit Care Resusc20161825–36
11. Bilotta F, Rosa G. Glycemia management in critical care patients.World J Diabetes20123130–134
12. Egi M, Finfer S, Bellomo R. Glycemic control in the ICU.Chest2011140212–220
13. Krinsley JS. Understanding glycemic control in the critically ill: 2011 update.Hosp Pract (1995)20113947–55
14. Krinsley JS. Glycemic control in the critically ill - 3 domains and diabetic status means one size does not fit all!Crit Care201317131
15. Braithwaite SS. Glycemic variability in hospitalized patients: Choosing metrics while awaiting the evidence.Curr Diab Rep201313138–154
16. Dungan K, Chapman J, Braithwaite SS, et al. Glucose measurement: Confounding issues in setting targets for inpatient management.Diabetes Care200730403–409
Keywords:

critically ill; glucose variability; hospital mortality; hypoglycemia

Supplemental Digital Content

Copyright © 2019 The Authors. Published by Wolters Kluwer Health, Inc. on behalf of the Society of Critical Care Medicine.