The Relation Between Hyperglycemia and Outcomes in 2,471 Patients Admitted to the Hospital With Community-Acquired Pneumonia

RESEARCH DESIGN AND METHODS

Capital Health (Edmonton, Alberta, Canada) is one of the largest integrated health systems in Canada and serves ∼1 million people. From 15 November 2000 to 14 November 2002, 2,785 adults were admitted with a clinical diagnosis of CAP by 318 different physicians in all six hospitals affiliated with Capital Health. Patients were excluded from this study if they did not have glucose measured at presentation; if they had physician-diagnosed or -suspected aspiration pneumonia, tuberculosis, or cystic fibrosis; if they were immunosuppressed (e.g., had HIV infection and CD4 count <250/mm³, were using >10 mg/day of prednisone or other immunosuppressive agents, were undergoing active treatment for cancer, or had a history of organ transplantation); if they were pregnant or nursing; or if they required direct admission to the intensive care unit. (Note that patients with CAP and diabetic ketoacidosis or hyperosmolar coma would have been directly admitted to intensive care units and thus would not be included in our study cohort.) After initial triage and assessment in the emergency department, an inpatient physician was consulted for hospital admission in all patients with Pneumonia Severity of Illness scores >90 points or those with lower scores in whom the emergency physician felt inpatient care was required (13). All admitted patients were treated in a standardized manner according to a regional critical pathway for CAP inpatients, which has been previously described (18). While choice of antibiotics, conversion from intravenous to oral therapies, and physiotherapy resources were standardized for all patients, management of comorbidities, such as diabetes, was left to the discretion of each patient’s attending physician. Six trained research nurses assisted with pathway implementation and prospectively collected standardized data on all study patients.

We decided a priori to examine different degrees of glycemia at admission (>11 mmol/l and ≥14 mmol/l), with the referent group for the bivariate and multivariate analyses being those patients with admission glucose <6.1 mmol/l (the level found to be optimal in a trial of aggressive versus conservative glucose control in intensive care unit patients) (19).

Adverse outcomes before hospital discharge were defined as death, any nonmetabolic complications (i.e., all in-hospital complications except for abnormalities of blood glucose), cardiac complications (acute coronary syndromes and/or heart failure), and nosocomial infections (i.e., in sites other than the lungs). Details were extracted directly from the medical charts. All outcomes were ascertained in an independent and blinded manner by the pneumonia pathway nurses and without knowledge of the hyperglycemia hypotheses outlined herein. The research nurses also assessed which patients had or had not met a priori–defined recovery milestones every 24 h (up to 5 days) after admission, which included becoming afebrile, respiratory rate <24/min, ability to take fluids by mouth, and oxygen saturation improving to ≥89%.

Statistical analyses

In addition to reporting summary statistics for the overall sample, we compared the frequency (for categorical data) and means (for continuous variables) between four patient subgroups defined by admission glucose levels. We examined for differences between the groups in covariates or outcomes using χ² tests for trend for categorical variables and Kruskal-Wallis test for continuous variables. Multiple logistic regression models were used to examine the relationship between a variety of covariates, admission glucose, and outcomes. The final multivariate model was fit using the forward stepwise technique with P to include set at 0.05 and P to exclude at 0.10. For all other comparisons, statistical significance was set at P < 0.05. All analyses were conducted using SAS version 8.2 (SAS Institute, Cary, NC).

RESULTS

Of the 2,785 patients admitted to non–intensive care unit settings with CAP during this study, 314 did not meet our eligibility criteria. The clinical features of the 2,471 patients forming our study sample are outlined in Table 1. Median age was 75 years, 52% were male, 15% were nursing home residents before admission, and 49% were taking at least four prescribed medications before admission. At the time of presentation, 178 of the 401 patients (44%) with a known diagnosis of diabetes and 101 of the 2,070 patients (5%) without a prior diagnosis had glucose exceeding 11 mmol/l. Overall, 9% of patients hospitalized with CAP died, 23% suffered an in-hospital complication, and the median length of stay was 6 days (Table 2).

As can be seen from the bivariate analyses in Table 3, a prior history of diabetes was not associated with mortality (odds ratio [OR] 1.00 [95% CI 0.69–1.45]) or in-hospital complication rates (1.14 [0.89–1.47]).

However, admission glucose was a significant predictor of adverse outcomes and prolonged length of stay (Table 2). Compared with those with values ≤11 mmol/l, patients with an admission glucose >11 mmol/l had an increased risk of death (13 vs. 9%, P = 0.03) and in-hospital complications (29 vs. 22%, P = 0.01). These differences were particularly pronounced in those without a prior diagnosis of diabetes (15 vs. 9%, P = 0.03 for death and 30 vs. 22%, P = 0.07 for in-hospital complications). Patients with an admission glucose >11 mmol/l were less likely to attain a priori–defined milestones for pneumonia recovery by the 5th day of admission (21 vs. 14% in those with admission glucose ≤11 mmol/l, P = 0.03) (Table 2). For example, by the 5th hospital day, 90% of patients with an admission glucose ≤11 mmol/l had their oxygen saturations return to normal or baseline compared with 83% of those with glucose >11 mmol/l (P = 0.02).

Compared with those patients with admission glucose ≤6.1 mmol/l, patients with admission glucose >11 mmol/l had a 73% higher (95% CI 12–168%) mortality risk and a 52% higher (12–108%) risk of in-hospital complications. Of note, these increased risks were not driven solely by those patients with admission glucose ≥14 mmol/l (as has been previously reported in the Pneumonia Severity Index). A breakdown of the risks in patients with admission glucose 11.01–13.99 and ≥14 mmol/l, along with covariates from the Pneumonia Severity Index, are presented in Table 3. Even after adjustment for other non–glucose-related covariates in the Pneumonia Severity Index (13), hyperglycemia on admission remained significantly associated with subsequent adverse outcomes. For each 1-mmol/l increase in admission glucose, risk of in-hospital complications increased 3% (0.2–6%) in the entire cohort and 5% (1–9%) in those with a history of diabetes; the risk of in-hospital death increased 8% (1–15%) per mmol/l in those with a history of diabetes.

CONCLUSIONS

Using a large dataset from all six hospitals in a large Canadian urban community linked by a common pathway for the treatment of CAP, this analysis indicates that individuals who are hyperglycemic at the time of admission to the hospital are at increased risk for adverse outcomes, including increased mortality, morbidity, and prolonged length of stay. The risk is graded and related to the degree of hyperglycemia; however, contrary to earlier reports (including the Pneumonia Severity Index) (13) and current clinical doctrine, the risk of adverse outcomes is increased at lower glucose levels than are currently felt to warrant intervention in non–intensive care unit patients. Over one-third of our patients with elevated admission glucose levels did not have a prior history of diabetes.

Our results confirm earlier studies suggesting a relationship between hyperglycemia at the time of admission and adverse outcomes (particularly nosocomial infections, cardiac complications, and death) in individuals admitted to the hospital with a wide range of clinical disorders (2–12). Akin with other investigators, we have shown that glucose levels are a more important prognostic factor in hospitalized patients than a prior diagnosis of diabetes (6). Furthermore, our data suggest that the occurrence of these complications is correlated with blood glucose on initial presentation even after adjusting for other confounding factors known to carry prognostic significance in CAP. Patients with hyperglycemia are prone to dehydration secondary to osmotic diuresis and exhibit a variety of perturbations of platelet function and endothelial function, as well as delayed chemotaxis, diminished granulocyte adherence, impaired phagocytosis, and reduced microbiocidal capacity (20–26). These abnormalities tend to develop when glucose exceeds 11 mmol/l and do improve in vitro with better glycemic control (20–26). An observational study, however, cannot be considered definitive, and alternative explanations could account for our findings. For example, hyperglycemia may not be a causal risk factor but merely a marker for an unmeasured confounder.

While there have not been any trials examining whether strict glycemic control improves outcomes in patients admitted to general medical wards, two trials provide some evidence that can inform this debate. In the first trial, 1,548 patients were admitted to a Belgian surgical intensive care unit (87% did not have diabetes, 73% had undergone coronary artery bypass graft, and the median APACHE II score was 9 [interquartile range 6–13]). Patients who were randomized to intensive blood glucose control (with a glucose-insulin infusion to keep their plasma glucose between 4.4 and 6.1 mmol/l) had 46% fewer septic complications and 34% lower mortality compared with those patients managed in conventional fashion (in whom mean fasting glucose averaged 8.5 mmol/l) (19). These results, taken in concert with another trial that demonstrated a 29% relative reduction in 1-year mortality after myocardial infarction in diabetic patients randomized to intensive glucose control (27), suggest that close attention should be paid to glycemic control in hospitalized patients irrespective of whether they have a prior diagnosis of diabetes.

Although this is a prospective study with rigorous collection of data in a standardized fashion and all patients had their pneumonia managed in a similar fashion as per a validated critical pathway, there are four limitations to our study. First, since glucose monitoring was not a component of the CAP care map, we were not able to examine changes in glucose levels during hospitalization in a standardized fashion. Nonetheless, because our results are based only on admission glucose, it is likely that we have underestimated the risks associated with hyperglycemia (given that the patients with the highest levels of glucose would have been more likely to be treated in the hospital setting). Second, even though we analyzed one of the largest community-based cohorts of hospitalized patients with CAP studied to date, we had insufficient numbers to examine the role of various antiglycemic treatments (e.g., insulin, metformin) on outcomes. Third, we did not routinely collect measures of long-term glucose control, such as HbA_1c. This would have allowed us to distinguish long-term poorly controlled diabetic patients (irrespective of a formal diagnosis) from those with hyperglycemia as a measure of significant physiological stress. However, given that a history of diabetes was not associated with outcomes on univariate or multivariate analyses in our study, we believe that acute hyperglycemia is more important than long-term glycemic control in predicting prognoses for patients admitted with CAP. Finally, although our cohort was population based and drawn from six different hospitals, all of our patients are from one Canadian health region, and whether our findings are generalizable may be questioned by some. However, our results are similar to those reported by Umpierrez et al. (6) from a community teaching hospital in Atlanta (where one-third of all hyperglycemic patients also did not have a prior history of diabetes and hyperglycemia was independently associated with mortality even after covariate adjustment in 2,030 consecutive adult admissions).

In conclusion, we found that a substantial proportion of our hospitalized patients with pneumonia had hyperglycemia and that this was associated with poor outcomes, even after adjusting for known prognostic factors in CAP. While observational studies (28,29) in non–intensive care unit settings echo the randomized trial evidence (19,27) from critical care units suggesting that stringent glucose control benefits ill patients, we believe that a large randomized trial is required to compare stringent glycemic control on regular inpatient wards with usual care in patients with CAP. The past few years have provided clear illustrations that the injudicious application of observational evidence or the overly zealous application of evidence from randomized trials conducted in highly specialized settings is fraught with potential harm (30,31). Based on our data, we estimate that a trial enrolling ∼900 patients with CAP (using the composite outcome of “death or in-hospital complication”) would be required to provide definitive answers about both the safety and the efficacy of stringent glycemic control versus usual care on general medical inpatient units.