Respiratory Dysfunction in Patients With Sepsis: Protective Effect of Diabetes Mellitus

Singapore General Hospital is the largest acute tertiary care hospital in Singapore. With 1600 beds, it serves approximately one-fourth of Singapore’s population of 4.98 million people.¹¹  Each year the hospital handles approximately 80 000 patient discharges. Data on all hospitalized patients 18 years or older, from January 1, 2004, to December 31, 2008, were collected from the hospital’s data warehouse, Information Technology Department, Singhealth Group. The data included demographic information such as ethnicity, sex, age, and marital status and clinical characteristics such as dates of admission and discharge, dates of ICU admission and discharge, up to 10 International Statistical Classification of Diseases and Related Health Problems, 9th Revision, Australian Modification (ICD-9-AM) diagnosis codes, up to 10 ICD-9-AM procedure codes, discharge status, and disposition at discharge. The research protocol was approved by the ethics committee of Singapore General Hospital, and no informed consent was required.

Sepsis was defined^1,3  as the presence of any of the ICD-9-AM codes listed in Table 1. In previous validation studies,^1,12  this method had positive predictive values of 88.7% to 99.8% for identifying true cases of sepsis. Diabetes mellitus was defined by ICD-9-AM code 250, which has a positive predictive value of 95%.¹³ 

Organ dysfunctions, type of organ dysfunction, source of infection (ie, the anatomical site of infection, which included respiratory; genitourinary; skin, soft tissue, or bone; gastrointestinal; central nervous system; cardiovascular; and others), and associated surgical and medical conditions were defined by using a combination of ICD-9-AM diagnosis, diagnostic codes and procedure codes^1,2,14  (Table 2). Chronic comorbid conditions were identified and were used to calculate the well-established adaptation of the Charlson Comorbidity Index by Deyo et al.¹⁵ 

Categorical variables were reported as percentages and continuous variables as means with standard deviations. Because of their skewed distribution, length of hospital stay and the number of comorbid conditions were summarized by using geometric means with 95% confidence intervals. Among patients with sepsis, variables were compared between those with and without diabetes mellitus by using χ² tests, 2-tailed t tests, and Mann-Whitney tests as appropriate. Logistic regression was used to assess the association of respiratory dysfunction with diabetes mellitus, and adjustments were made for age, sex, ethnicity, ICU admissions, number of comorbid conditions, and the source of infection. The Hosmer-Lemeshow χ² goodness-of-fit test was used to exclude nonsignificant variables and to identify clinically plausible interaction terms. The area under the receiver operating characteristic curve was also presented for the regression model. All P values were 2 sided. The level of significance was set at the conventional level of α=.0 5. STATA, version 10.0 (StataCorp LP, College Station, Texas) was used for all data analysis.

Of the 383 238 patients 18 years and older hospitalized during the period 2004 through 2008, a total of 9221 (2.4%) met the criteria used to define sepsis. Among the patients with sepsis, 2943 (31.9%) had diabetes mellitus. The patients with diabetes were generally older (mean 65.5 years; SD, 12.4) than those without diabetes (59.8 years; SD, 17.4; P < .001) and were predominately women (53% vs 48.5%; P < .001). The percentage of patients managed medically did not differ significantly (P = .87) between those with both sepsis and diabetes (66.5%) and those with sepsis only (66.7%). The percentage of patients who received ICU care also did not differ significantly (P = .28) between those with both sepsis and diabetes (16.5%) and those with sepsis only (17.4%). Patients with both sepsis and diabetes had significantly more comorbid conditions than did patients with sepsis only (geometric mean, 1.9 vs 1.3; P < .001). The demographic profile, clinical characteristics, and outcomes of patients with sepsis are given in Table 3.

Patients with both sepsis and diabetes had a significantly lower frequency than did patients with sepsis only for respiratory (27.6% vs 33.4%), gastrointestinal (25.5% vs 29.3%), cardiovascular (0.7% vs 1.9%), and neurological (0.8% vs 1.4%) sources of infection (Figure 1). However, patients with both sepsis and diabetes had an increased frequency of renal (32.6% vs 26.2%); skin, soft tissue, or bone (15.6% vs 13.1%); and other (19% vs 15.1%) sources of infection.

Among all patients with sepsis, 10.9% had respiratory dysfunction, 31.5% had renal dysfunction, 19.2% had cardiovascular dysfunction, 11.8% had hematological dysfunction, and 7.6% had other organ dysfunction (Figure 2). Patients with both sepsis and diabetes were more likely than those with sepsis only to have renal (46.5% vs 24.4%) and metabolic (4.7% vs 3.6%) dysfunction and less likely to have respiratory (9.4% vs 11.6%), hepatic (1.1% vs 1.7%), and hematologic (9.4% vs 12.9%) dysfunction. The occurrence of other organ dysfunctions differed little between the 2 groups.

Multivariate logistic regression analysis showed that diabetes mellitus was significantly negatively associated with respiratory dysfunction (adjusted odds ratio, 0.80; 95% confidence interval, 0.66–0.98; P = .03) after adjustments were made for age, sex, ethnicity, ICU admission, number of comorbid conditions, and infection sources, including respiratory; genitourinary; skin, soft tissue, or bone; and gastrointestinal sources (Table 4). The goodness of fit of the regression model was assessed by using the Hosmer-Lemeshow test (χ₈²  =8.3; P =.4 0). The area under the receiver operating characteristic curve was 0.92 (P < .001).

Evaluation of hospital mortality and length of stay (Table 3) indicated that sepsis patients with and without diabetes mellitus had comparable hospital mortality (19.2% vs 20.0%; P = .37). This finding was evident in the subgroup of patients with respiratory dysfunction as well (56.7% vs 52.9%; P = .29). Similarly, the hospital length of stay did not differ significantly between the 2 groups (geometric mean, 12.1 days vs 12.2 days; P = .53) or in the subgroup of patients with respiratory dysfunction (15.9 days vs 17.8 days: P = .21).

In this study of 5 years of discharge data from a major tertiary acute care hospital, we found that respiratory dysfunction was less likely to develop in patients with both sepsis diabetes mellitus than in patients with sepsis only. This difference persisted even after adjustments were made for age, sex, ethnicity, ICU admission, the number of comorbid conditions, and the source of infection.

Our study was the first on the effect of diabetes mellitus on respiratory dysfunction in an Asian population, and our results confirm and extend the previous published results for a Western population.⁹  In an observational cohort study⁶  of 375 patients treated with mechanical ventilation for more than 48 hours, higher glucose levels in the first 24 hours of respiratory failure were associated with a decreased risk for acute respiratory distress syndrome (ARDS; P = .025). In another large cohort of 688 critically ill patients at risk for ARDS from sepsis, trauma, massive transfusion, and aspiration, diabetes protected against the development of ARDS, even after adjustments for potential confounders such as age, clinical risk for ARDS, severity of illness, and transfusion (adjusted odds ratio, 0.58; 95% confidence interval, 0.36–0.92).¹⁶  In a third independent cohort study¹⁷  of 160 patients with sepsis, diabetes mellitus was detected in only 24% of patients with acute lung injury, the less severe form of ARDS, and in 43% of patients without acute lung injury (P = .01). Investigators in other clinical trials^8,18  have also reported reduced occurrence of ARDS or acute lung injury in sepsis patients with diabetes mellitus. Furthermore, in a large epidemiological study involving 12.5 million sepsis cases from the US National Hospital Discharge Survey from 1979 to 2003, Esper et al⁷  found that patients with diabetes mellitus were less likely to have acute respiratory failure than were patients without diabetes (9% vs 14%; P < .05) and were more likely to have acute renal failure (13% vs 7%; P<.05). The reduced likelihood of respiratory dysfunction in the patients with diabetes mellitus occurred irrespective of the source of infection.

Our results and these published findings support the view that respiratory dysfunction is less likely to develop in patients with sepsis and diabetes mellitus than in patients with sepsis only. Nevertheless, the results of 2 studies contradict this finding. In a large population-based case-control study¹⁹  and a cohort study,²⁰  the risk of pneumonia-related hospitalizations and deaths was increased in patients with type 2 diabetes. These contradicting study results emphasize that the underlying mechanism for the effect between diabetes mellitus and respiratory dysfunction is still not well understood.

Neutrophils play an essential role in the host inflammatory response against infection.²¹  Neutrophils from patients with diabetes can have impaired biological responses, including reduced bactericidal activity and impaired chemotaxis.^22,23  Compared with neutrophils from age and sex-matched controls, neutrophils from patients with diabetes adhere with less affinity to endothelial cell monolayers.^24,25  In addition, neutrophils from diabetic patients produce 40% less leukotriene B4 than do neutrophils from healthy volunteers.²⁶  Also, neutrophils from uninfected patients with poorly controlled diabetes did not have the same bactericidal activity as did neutrophils from healthy volunteers or patients with well-controlled diabetes.²³  Furthermore, defects of neutrophil chemotactic, phagocytic, and microbicidal activities in experimental studies in diabetic rats and mice have been reported.^21,27  These abnormalities in neutrophil function may protect the lungs by decreasing the ability of the neutrophils to migrate into the organ and to produce oxidant damage. Impaired neutrophil function or altered neutrophil-endothelial interactions have been proposed as a possible mechanisms for reduced incidence of respiratory dysfunction.^22,23 

Likewise, hyperglycemia can have immunomodulatory effects. Hyperglycemia in itself has an independent effect on neutrophil and endothelial properties, resulting in decreased injury in response to endotoxin.²²  In an in vitro study,²⁸  hyperglycemia increased production of anti-inflammatory cytokines such as interleukin 10 and promoted mitochondrial dysfunction. Impaired neutrophil function resulting in decreased intracellular bactericidal activity, opsonic activity, and innate immunity has also been reported.²⁹  Such effects could theoretically modulate the intense inflammation associated with respiratory organ dysfunction and may explain the decrease in respiratory dysfunction, but what the effects contribute to the increased incidence of other organ dysfunctions, such as renal failure, is not known. Further studies on these differential effects are needed.

From the clinical perspective, the development of organ failures determines the course and prognosis of most sepsis patients. Patients who usually die of sepsis in the course of different comorbid conditions have their deaths attributed to these conditions rather than to sepsis. Among specific organ dysfunctions, respiratory dysfunction is the most common, and then, in order, cardiovascular and renal failure.

The finding that patients with both sepsis and diabetes are less likely than patients with sepsis only to have respiratory dysfunction and more likely to have renal failure will have a significant impact on the management of sepsis patients. The effect of diabetes in patients with sepsis will be augmented by the increasing incidence of diabetes among Asian patients.³⁰  It would affect the need of support systems and equipment resource in ICUs and high-dependence units where sepsis patients receive care. The effect of diabetes could also be a prognostic indicator in patients with sepsis.

Our study had several limitations. The advantages of hospital discharge data are well known. The data are relatively inexpensive to obtain and use, are more reliable than other sources of data, cover a large population over multiple years, and are being used extensively in health services and health policy research.³¹  Nevertheless, hospital discharge data are administrative and are not intended for research purposes. The data are subject to coding errors and omissions of important diagnoses and complications, conditions that could have affected our results.³²  However, the coding errors might occur across all patient groups, a possibility that offsets the possible impact of the errors. Although detailed chart reviews could provide more reliable data, chart reviews are time-consuming and labor intensive.³³ 

The data we used were generated from a single, large acute care tertiary hospital, a situation that limits the generalizability of our results. Furthermore, lack of data on patients’ medication history and/or duration of existing comorbid conditions might affect the interpretation of the results. Another possible confounding variable is chronic alcohol abuse, which may increase the incidence of ARDS in critically ill patients.²³  Unfortunately, we do not have the alcohol intake history of all patients whose data were used in our study.

In conclusion, our results confirm and extend the results of other studies done in Western populations, suggesting that in this group of Asian patients, respiratory dysfunction is also less likely to develop in patients with both sepsis and diabetes mellitus than in patients with sepsis only.

Y. Yang designed and performed the study and contributed to the analysis of data, interpretation of data, and writing of the final draft of the manuscript. Z.H. Abdul-Salam contributed to the study design, the analysis of data, and contributed to the manuscript. B.C. Ong contributed to the study design, the analysis of data, and the writing of the manuscript. K.S. Yang contributed to the study design, the analysis of data, and the manuscript. We acknowledge Mr Jeffrey Fong, Mrs Bessie Ang, and Mrs LiSee Lou for their assistance in database preparation.