Introduction

The proportion of patients with diabetes admitted to the ICU is growing as a result of the worldwide increase in type 2 diabetes. In cardiac surgery patients, the growth is even more pronounced, since patients with diabetes-associated, often complex, multivessel macrovascular complications are preferably treated surgically rather than by percutaneous intervention [1, 2]. It is known that diabetic patients admitted to the ICU are more prone to develop complications [35], at least in part because of hampered immune cell function associated with the disease [6, 7]. One would expect an increased mortality rate for diabetic ICU patients, but the literature at this point is conflicting, with reports of increased, equal or even decreased mortality rates among this population compared to patients without diabetes. Also, there might be a difference in outcomes between various ICU populations, such as cardiac surgery and medical patients.

To better understand the role of diabetes in critical illness, we conducted a systematic review and meta-analysis of studies regarding mortality, including observational as well as intervention studies that reported ICU, in-hospital and/or 30-day mortality rates of ICU-admitted patients with diabetes.

Materials and methods

This study was conducted according to the recommendations of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group [8]. The study did not need ethical approval, since it was a retrospective analysis of anonymous data.

Data sources and search strategy

Together with the clinical librarian at our institution, an electronic search of MEDLINE and Embase from 1 May 2005 to 1 May 2010 was performed for studies that reported mortality of adult ICU patients, including both retrospective and prospective studies and both observational and intervention studies. The five-year limit was chosen because we expected that insulin treatment regimens and other therapies would be comparable among studies in this rapidly evolving field. The text terms and medical subheading, or MeSH, terms 'intensive care unit', 'critical care' and 'mortality' were combined. Since a preliminary search showed that diabetes was not always the primary interest of included studies, and to prevent publication bias, we narrowed the search by not using 'diabetes mellitus' as a search term. To avoid possible treatment-induced bias, randomized controlled trials comparing intensive insulin therapy regimes were excluded [9]. We limited our search to research performed in adult human patient populations and published in the English-language literature. Unpublished studies were not identified.

Study selection

Two reviewers (SES and MH) independently screened the records. Agreement on final inclusion was reached by consensus. Inclusion and exclusion criteria were defined a priori. A study was included if it reported crude ICU, in-hospital and/or 30-day mortality of ICU-admitted patients with or without diabetes and/or included univariate or multivariate analysis of mortality risk for diabetic patients represented as odds ratios (ORs), hazard ratios or relative risks. Studies reporting 28- or 30-day mortality rates were combined. We excluded studies in which diabetes was the reason for ICU admittance, that is, ketoacidosis, as well as studies concerning patients with gestational diabetes and ICU readmissions. When it was not possible to obtain the full paper from our institutional library or the Internet, the corresponding author was contacted if the population included in the study was larger than 500 patients.

Data extraction

From the included publications the following data were extracted: first author, year of publication, country where the work was performed, study design, type of ICU, population specification, reported mortality type, definition of diabetes and the numbers of patients with and without diabetes. Subsequently, the number of deaths among patients with or without diabetes and/or the mortality risk associated with the presence of diabetes were extrapolated. We contacted the corresponding author for additional information when the publication reported only a P value for mortality risk or when the exact number or proportion of patients with diabetes was not reported.

Study quality

No individual assessment of study quality was performed. Funnel plot analysis did not suggest any publication bias. We did not expect bias in outcome reporting, because death is a robust end point, and it is unlikely that patients were lost to follow-up, as the primary outcome was recorded in the hospital.

Data synthesis and statistical methods

The meta-analysis was performed using RevMan version 5 software (The Cochrane Collaboration, Oxford, UK). Analyses were performed separately for ICU, in-hospital and 30-day mortality as outcome variables. Data were synthesized using inverse variance with ORs as the effect measure. In the primary analysis, unadjusted results were used where possible. If only crude mortality data were provided, then the ORs and standard errors were calculated. An OR > 1 suggested that diabetes was associated with an increased risk of death. We stratified the analyses by ICU type: trauma, surgical, medical and mixed. Data were pooled using a random effects model because heterogeneity between studies was anticipated. Heterogeneity was assessed using the I2 statistic.

We performed sensitivity analyses to explore the robustness of the data and the influence of the following factors on effect size by repeating the analysis: (1) using a fixed instead of a random model, (2) leaving out the two largest studies, (3) making distinctions between cardiac and other types of surgery, (4) selecting and pooling only the five studies reporting unadjusted as well as adjusted outcomes and comparing these two effect sizes to assess the possible influence of confounders and (5) using adjusted instead of unadjusted effect sizes of those five studies in the in-hospital mortality analysis.

Results

Figure 1 summarizes the study identification and selection process. After removal of duplicates derived from the Embase and MEDLINE searches, 3,220 potentially interesting records were identified. Of these, we excluded 556 publications by reviewing the abstracts. After review of 2,664 full articles, 2,520 were excluded: 1,814 because no data at all were available regarding diabetes, 579 because mortality data were not available, 35 because they reported only long-term mortality and 2 duplicate publications were identified. We were not able to obtain the full texts of 90 potentially relevant records, of which 14 reported a sample size of more than 500 participants. Three records were excluded by data extraction because only a P value for mortality was reported and, after contacting the authors, the raw data could not be retrieved [1012]. Finally, 141 studies could be included in the analysis of ICU mortality (n = 50 [1362]), in-hospital mortality (n = 74 [26, 35]. These complications are associated with mortality, at least in the nondiabetic patient population. Apparently, diabetes is a risk factor only for the development of complications, but once acquired, the mortality risk is equal to or perhaps even lower than that of nondiabetic patients.

A possible explanation for the relative protection of critically ill patients with diabetes could lie in the different effects of stress-induced hyperglycaemia between the two groups. Hyperglycaemia is common in critically ill patients, and not only in those with diabetes. It is associated with mortality [156], but it has been shown that patients with diabetes are less affected than nondiabetic patients by high glucose levels [26, 77, 157, 158]. It might be that the relative protection from stress-induced hyperglycaemia counteracts the increased mortality risk due to an increased amount of complications. Other mechanisms may, of course, contribute [26]. Further studies are needed to unravel the exact effects of hyperglycaemia in patients with or without diabetes.

Because of the nature of the included studies, we mainly show unadjusted results. It might be possible that the baseline characteristics between patients with or without diabetes were different, which would influence the outcome. If there were a difference, it would seem likely that patients with diabetes would be more severely ill at the time of admission. Adjusting for severity of illness would have decreased the ORs, as shown in the sensitivity analysis, thus indicating a greater advantage only for patients with diabetes. This effect is seen in the in-hospital mortality outcomes among the trauma patient population. It might be possible that adjustment for severity of disease decreases the negative effect of diabetes seen in cardiac surgery patients. However, our results represent the mortality risk of the average patient with or without diabetes, irrespective of the differences between the populations associated with diabetes.

There are limitations associated with our meta-analysis. First, the published data do not allow differentiating type 1 from type 2 diabetes or insulin-treated from non-insulin-treated diabetes, and it might be possible that there are differences in outcomes between these groups. Furthermore, incomplete capture of diet-treated or undiagnosed diabetes in the included studies is likely, as most studies relied on the use of glucose-lowering drugs. Second, we could not retrieve mortality data from three studies. It is unlikely that including these three studies would have shifted the results to a disadvantage for patients with diabetes, since the P values for mortality related to the presence of diabetes in these studies were not significant. Third, for 90 potentially relevant studies, the full articles could not be retrieved. These studies comprised a total of 48,263 patients, which would have contributed only 0.39% to the total population, so we expect that this had little influence on the outcomes. Fourth, the heterogeneity among studies was quite large. Fifth, the crude mortality rates in the three surgical populations were driven by the studies included: the high mortality rate in the surgical ICUs is heavily weighted by general surgery studies, whilst for in-hospital and 30-day mortality figures, the cardiac surgery studies with much lower mortality rates dominated.

Conclusions

We show in this meta-analysis that patients with diabetes who are admitted to medical, mixed and trauma ICUs have chances of survival similar to those of patients without diabetes. Diabetes significantly increases mortality risk only in patients admitted after surgery, more specifically after cardiac surgery, a population with distinct characteristics of the underlying disease. Further studies are needed to unravel the pathophysiological mechanisms by which patients with diabetes seem to be relatively protected in nonsurgical settings, despite higher complication rates.

Key messages

  • Patients with diabetes admitted to medical, mixed and trauma ICUs have chances of survival similar to those of patients without diabetes.

  • Patients with diabetes should not be denied access to ICU facilities on the basis of their diabetes.

  • The presence of diabetes significantly increases mortality risk for patients admitted after cardiac surgery.

  • The pathophysiological mechanisms by which patients with diabetes have equal mortality rates despite higher complication risks in nonsurgical settings should be a subject of further research.