Diabetes mellitus is associated with lower mortality in intensive care unit: a multicenter historical cohort study

Background: The association of pre-existing diabetes mellitus and outcomes among critically ill patients remains unknown. Methods: This retrospective study enrolled patients who were covered by the eICU Collaborative Research Database from 2014 to 2015. DM was the exposure of interest, and diabetic individuals were adjudicated by the medical history, and blood glucose level (BGL). We abstracted basic characteristics, laboratory variables, and primary exposures. ICU mortality was the primary outcome. Results: In a cohort of 134,429 critically ill patients (male 54.4%, median age 66 [54-77] years, BMI 28[24-33] kg/m 2 ), the prevalence of DM was 29%. In comparison with nondiabetics, DM patients were older, more obese, had higher Acute Physiology and Chronic Health Evaluation (APACHE)-IV score, and ICU admission BGL. In comparison with nondiabetics, pre-existing DM was associated with lower ICU mortality (OR: 0.846, 95%CI: 0.791-0.905). In multivariable logistic regression and Cox proportional hazard analyses, pre-existing DM was associated with decreased odds of ICU mortality in hyperglycemic patients (>163 mg/dL), higher APACHE IV score (>67), middle to old age (45-75 years), sepsis and morbid obesity (BMI>35 kg/m 2 ). Also, in comparison with nondiabetics, pre-existing DM was associated with lower mortality among those with higher mean BGL (>128 mg/dL), and higher mortality in lower mean BGL (<107 mg/dL). Conclusions: In comparison with nondiabetics, pre-existing DM is associated with a lower adjusted ICU mortality. This association is stronger in DM patients with hyperglycemia, obesity, sepsis, middle to old age, and higher APACHE IV score. without glucose level),


Introduction
Diabetes mellitus (DM) is a metabolic disorder with complex pathophysiology related to hyperglycemia, insulin resistance, oxidative stress, and abnormal lipid metabolism. DM is a well-known cause of multisystem chronic disorders, such as immune system, renal, neurological, cardiovascular, and vascular systems [1]. Although many studies have focused on glucose control in the intensive care unit (ICU), the effect of DM on the outcomes of critically ill patients remains unclear [2]. While in some studies, pre-existing DM is reported as an independent risk factor for mortality and other complications in speci c groups of critically ill patients (i.e., cardiac surgery [3], trauma [4,5], and sepsis [6]), other investigations showed no association with worse outcomes [7] or in some cases it was associated with improved survival [8] (e.g., sepsis [9][10][11]). These contradicting ndings could be due to inclusion bias, study populations, and methodological differences among the studies [12]. On the other hand, the primary focus has been on the association of DM, blood glucose level (BGL) and mortality, only a few studies studied the associations of the severity of baseline comorbidities, variety of exposures and mortality in ICU patients with DM.
The impact of pre-existing DM on the outcomes of ICU patients follows complex pathophysiologic mechanisms. Long term DM would lead to microvascular alteration and impairment, which, in turn, result in decreased oxygen delivery and organ dysfunction [13]. On the other hand, the DM-associated microvascular disease induces chronic hypoxia resulting in tissue preconditioning, which can potentially lead to more resistance to acute ischemic events [14]. Besides the microvascular insu ciencies, factors such as age, control of BGL [15], and body mass index (BMI) [16] may affect outcomes of patients with diabetes. For example, the "obesity paradox" has been observed, i.e., critically ill obese patients have lower mortality [17]; however, the potential bene ts and possible mechanisms in DM are still undiscovered [18].
We hypothesized that the pre-existing DM would be associated with increasesed the ICU mortality in patients admitted to ICU. Hence, the primary aim of this study was to assess the association between pre-existing DM and ICU mortality. We also investigated the impact of age, glucose level, BMI, exposures, and comorbidities on the mortality rate of diabetic critically ill patients.
De nition of DM Diabetes mellitus was de ned based on one of these two criteria as per eICU-CRD database guidelines. These criteria included 1) the medical history of insulin-dependent, or oral hypoglycemic agent dependent, or non-medication dependent diabetes, or 2) ICD diagnosis of diabetes mellitus during hospitalization (ICD-9 code 250.00 or ICD-10 code E11.9) and at least a glucose level of ≥ 200 mg/dL (11.1 mmol/L) at any time during ICU admission [19,21]. The patients with isolated or stress hyperglycemia (ICD-9 code 790.6) were not considered as diabetics.
To assure the reliability of diabetes mellitus adjudication, all the data were abstracted and recorded by CH and JW; then it was validated by BH and ZP. In case of disagreement, KK and JL dissolved the issue.

Data extraction
We extracted data from the eICU-CRD based on structure query language (SQL) with PostgreSQL 10.7 and Navicat Premium version 11.0.17. The abstracted baseline characteristics included age, gender, race, body mass index (BMI), admission type, and source of admission.
Laboratory variables included BGL (glucose level at ICU admission, mean level of the second ICU day, and mean level throughout ICU stay), lactate, white blood cell, hemoglobin, platelet, serum creatinine, blood urea nitrogen. The highest Acute Physiology and Chronic Health Evaluation (APACHE) -IV score [22], OASIS [23] scores, and Sequential Organ Failure Assessment (SOFA) within 24 hours of ICU admission were calculated. Charlson comorbidity index (CCI) was determined at hospital admission. The differences in APACHE IV and CCI elements in between DM and non-DM patients were assessed. Admission APACHE IV diagnosis category during ICU admission (i.e., sepsis, acute myocardial infarction [AMI], cerebrovascular accident/stroke [CVA], pneumonia, acute kidney injury [AKI] ) were recorded. The de nition of AKI was based on an increase in plasma creatinine level elevate ≥ 0.3 mg/dL within 48 hours after ICU admission and the ICD codes for acute renal failure (ICD-9 code 584.9; ICD-10 code N17.9). ICU and hospital mortality were recorded, and expected mortality based on the APACHE IV score was calculated. The proportion of patients who required mechanical ventilation and/or vasopressor administration and the length of ICU and hospital stays were recorded.

Statistical analysis
The normality of the distribution for continuous variables was assessed by the Kolmogorov-Smirnov test, and normally distributed variables were summarized as the means ± standard deviation (SD). The skewed variables were summarized as the median and interquartile range (IQR). We used Student t-test and Mann-Whitney U test as appropriate. We used Pearson's chi-squared test or Fisher's exact test to compare categorical variables, as necessary. We used binary logistic regression analysis to examine the association between the group allocations and ICU mortality, and the results were presented as the odds ratio (OR) and associated 95% con dence interval (CI). Regression models were used to determine the successive in uence of potential confounders on the relationship between speci c population and ICU mortality in patients with or without DM. Cox hazard regression models were used for survival analysis during ICU stay. We adjusted for ICU admission glucose level, APACHE IV, age, BMI, and different admission APACHE IV diagnosis category. All potential confounders were strati ed (BMI to the American Heart Association overweight criteria [24], age to the World Health Organization criteria, APACHE IV, and BGL to the quartiles).
Multivariable analysis was used to reduce the effect of residual confounding factors by adjusting for the baseline characteristics, including age, gender, race, BMI, CCI, APACHE IV, was along with the severity of the organ dysfunction (SOFA). All the statistical analyses were performed using the IBM SPSS version 24.0 (IBM Corp., Armonk, NY, USA). GraphPad Prism (version 8.0) was used to design gures. Twosided P-values < 0.05 represented statistical signi cance. Figure [22.97-31.45] kg/m 2 , p < 0.001) compared to non-DM. Patients with DM in comparison with nondiabetic patients also had signi cantly higher white blood cell, platelet, serum creatinine, blood urea nitrogen, glucose, and potassium levels, and signi cantly lower hemoglobin, serum albumin and sodium concentrations (all p < 0.001). DM patients in the ICU were admitted more frequently for sepsis (16.9% vs. 12.9%, p < 0.001) and AKI (2.3% vs. 1.1%, p < 0.001), whereas no statistically signi cant was found in AMI, CVA or pneumonia than non-DM. In the treatment, DM patients had higher need for mechanical ventilation (33.1% vs. 29.0%, p < 0.001) and vasopressor (19.1% vs. 16.5%, p < 0.001) when compared with non-DM.  P values indicate differences between DM and non-DM. P < 0.05 was considered as signi cant.
We divided patients based on age according to the World Health Organization guidelines into four groups (< 45 years; 45-59 years; 60-75 years; and > 75 years). Patients with DM had a similar mortality rate when compared with nondiabetic patients in each age group (Figure.   (Table 4).

Discussion
In this multicenter historical cohort study, pre-existing DM, when compared to non-DM patients, was associated with lower ICU and hospital mortality after adjustments. Indeed, we found that among patients with a higher prevalence of comorbid conditions and severity of illness, hyperglycemia, middle to old age ranges, being overweight, and sepsis, pre-existing DM was associated with lower mortality.
DM is considered as the leading cause of kidney dysfunction and non-traumatic lower limb amputation [25]. Diabetes is also associated with doubling risk of death due to cardiovascular disease when compared with patients without diabetes [26]. In a retrospective cohort of 10,320 critically ill patients, authors also demonstrated the presence of DM could modify the association between glucose control and ICU mortality, as diabetic patients tolerate a more extensive glucose variability when compared with nondiabetic ICU patients [8]. Graham [7]. In our study, even though patients with pre-existing DM were older, more obese, with had higher APACHE IV, their adjusted ICU, and hospital mortality were also lower than nondiabetics.
While the glycemic control is vital to ICU patients, the optimal target glucose range is still unclear, especially with different underlying disease. We noted the lowest mortality when blood glucose levels were near-normal (i.e., 103-127 mg/dl) in both DM and non-DM groups, regardless of glucose level that was used for the models or presence of pre-existing DM.
Admission ICU BGL was a predictor for mortality in critically ill patients. In ST-segment elevation myocardial infarction complicated by cardiogenic shock, admission BGL is an independent predictor of increased risk of death only among patients without DM [27], Kim and colleges also found admission BGL was associated with increased cardiac ICU mortality in non-DM critically ill individuals but not in critically ill patients who had DM [28]. In our research, we demonstrated patients with a history of DM in comparison with nondiabetics had signi cantly higher crude ICU mortality rate when the rst ICU admission glucose levels were low (i.e., < 103 mg/dL). Following adjustment for age, gender, race, BMI, APACHE IV (calculated without blood glucose level), CCI, and SOFA, the association mentioned above disappeared. Besides, the pre-existing DM was signi cantly associated with a lower adjusted ICU mortality when glucose levels were at the higher range (i.e., > 128 mg/dL) when compared with nondiabetics. Cox proportional hazard model indicated pre-existing DM was associated with decreased odds of ICU mortality in hyperglycemia (i.e., > 163 mg/dL).
Regarding BGL target in critical ill, Sechterberger and colleagues found that elevated BGL with its high variability were associated with ICU mortality among patients who did not have pre-existing DM [8], James SK and colleagues suggested that patients with diabetes may bene t from higher glucose target ranges than those without DM [29], these ndings have been reported in several other previous studies [30][31][32]. In our study, the mean glucose level on the second ICU day in the presence of DM history was associated with lower ICU mortality when it was > 128 mg/dL, and there was no association with ICU mortality when the mean BGL was < 107 mg/dL or near-normal (i.e., 107-127 mg/dL) range compared with non-DM patients. Regarding the mean blood glucose level throughout ICU stay, the pre-existing DM was associated with higher ICU mortality when the mean BGL was < 107 mg/dL, and with lower ICU mortality when the mean BGL was at higher range (i.e., > 124 mg/dL) compared with to non-DM patients.
The incidence of diabetes is directly correlated with age [33]. Old age (> 70 years) is also the independent risk factor for the mortality of critically ill patients, with the maximum rate of mortality being about 0.75% per year between the ages of 71 and 77 years [34]. In our study, the association of age and ICU mortality was similar to the previous literature. In multivariable logistic regression analysis and Cox proportional hazard model for the survival analysis, the potential association of pre-existing DM and ICU mortality showed an age-related inverse correlation (OR: 0.63, 0.77, 0.83, 0.93; HR: 0.74, 0,81. 0,86, 0.96 for an increase in each age group), although based survival analysis the signi cant difference was primarily among those in the middle to old age ranges, i.e., 45-59 and 60-75 years old. The possible reason for higher mortality in old diabetic critically ill patients included 1) restrictive diets in diabetic patients with old age may lead to accelerated deterioration of muscle and bone structure [35], and 2) older critically ill patients are admitted to ICU more often, which may lead to a higher rate of ICU-related complications, e.g., iatrogenic bloodstream infections [36].
Pre-existing DM also in uenced the mortality rate in different BMI striae. Dhalwani and colleagues showed that for the same level of obesity, mortality risk was higher in individuals without diabetes compared to those with diabetes [37]. Carnethon et al. also reported that adults within normal weight range at the time of DM diagnosis had higher mortality than adults who were overweight or obese [38]. Tobias and colleagues performed a similar study which did not replicate these results [39]. In our research, compare to patients without diabetes, patients with very high (> 35 kg/m 2 ) BMI range had lower mortality when were diabetic. Although pre-existing DM was associated with lower mortality in normal range of BMI (i.e., 18.5-24.9) compare to non-DM, we were not able to replicate this in the survival analysis or Cox proportional hazard model, which is similar to other studies. Therefore, we believe this could be due to chance, only.
Diabetes can lead to complications and organ dysfunction with the pathophysiological alterations in the chronic disease processes. However, the effect of pre-existing DM on acute conditions is still unclear. The most controversial area is sepsis. Among septic patients, a history of DM has been observed to be associated with higher [6,13], lower [9,11,40], and no change in mortality [12,41]. Recent studies also indicated lower mortality among DM and sepsis when compared to non-DM [9,11]. This veri es our ndings. The potential mechanism for this phenomenon include improved acute hyperglycemia tolerance [11] and reduced degree of the in ammatory cytokine response [42]. Regarding the association between DM and mortality in pneumonia patients, McAlister and colleagues found pre-existing DM had no in uence on mortality among patients with pneumonia [43], but more recent investigations reported in community-acquired pneumonia, DM was associated with worse prognosis, including increased rate of pleural effusion and mortality [44,45]. In our study, pre-existing DM was not associate with the mortality rate in pneumonia patients.
DM have been identi ed as a risk factor for developing AKI [46], but in recent years, some studies showed that diabetes might provide a protective effect against AKI, especially sepsis-associated AKI [47]. In our study, DM nearly had no in uence on the mortality of AKI patients, and a similar result noted between DMs and ICU mortality among patients with AMI and CVA. Although some studies showed a prevalent positive association between DM and AMI [48], and CVA related mortality [49], they mostly focused on the longer outcome, while increased ICU mortality due AMI or CVA was not elucidated.
We also noted that DM was associated with lower mortality in higher APACHE IV score when compared with non-DM. With further evaluation, among all elements in APACHE IV, the most in uential factor related to mortality differences among DM and non-DM was BGL. The contribution of BGL to the calculation of APACHE IV score among patients with DM was signi cantly higher than non-DM patients. This difference could explain why DM was associated with lower mortality in higher APACHE IV score compared to non-DM. Notably, DM was an independent variable in lower ICU mortality in hyperglycemia.
The potential mechanisms of lower observed ICU mortality among patients with pre-existing diabetes mellitus is not well de ned. However, this could be due to 1) improved tolerance to acute hyperglycemia episodes during critical illnesses among diabetics when compare those without diabetes [50], 2) long-term hypoxia-induced preconditioning and blunted in ammatory response in pre-existing DM [51], 3) reduced the degree of the in ammatory cytokine levels increasing (interleukin-6, tumor necrosis factor alpha [52], plasma protease factor VIIactivating protease [42]) when compared with patients without DM; and nally, 4) higher prevalence of therapies such as treatment with insulin [41], statins [53], metformin [54] and fenoldopam [55] may have protective effects due to their anti-in ammatory characteristics.
Our study has several limitations. Similar to any retrospective observational study, we are not able to imply any causal relationship. Prospective observational studies may be required to validate our ndings in associations between diabetes and mortality in ICU. The heterogeneity in diagnostic tools and treatment strategies between 208 hospitals may reduce the reliability of our ndings. Some missing laboratory variables in the eICU-CRD database, e.g., HbA1c, may increase the chances of biased results. The lack information of the treatment on DM may also lead to the variability of mortality analysis. Finally, we were not able to access the long-term outcomes information.

Conclusions
Compared with nondiabetics, pre-existing DM may be associated with lower adjusted ICU mortality. This association is particularly stronger when patients had hyperglycemia, higher APACHE IV score, or sepsis, were more obese or older.