Supplementary Material
Methods
This was a retrospective matched case control study conducted in an adult tertiary teaching hospital in Brisbane, Australia. The study was approved by the ethics committee of the Princess Alexandra Hospital (Approval number = HREC/11/QPAH/109).
All positive blood cultures from 1st January 2002 to 31st January 2012 were identified from microbiology laboratory records. Patients prescribed a glitazone during this period were identified from computerised pharmacy records. Patients who appeared on both lists had their medical record manually reviewed. Patients whose blood culture grew organisms likely to be contaminants (coagulase negative staphylococci, Bacillus spp., Proprionibacterium spp., Corynebacterium spp. and Micrococci) were excluded from further analysis. Potential cases were only included if they were actually on a glitazone at the time of the hospital admission that coincided with the blood culture being taken. This was noted from either a medication chart or list of current medications in the medical record.
For each subject three to four matched controls not taking a glitazone were identified from the pool of positive blood cultures. Controls were matched as close as possible to gender, age, causative organism and collection date. The blood culture collection date was matched to +/- 6 months, which was extended to +/- 12 months where there were limited numbers of potential controls. For cases with multiple potential controls those with an elevated glycosylated haemoglobin (HbA1c) (>5.9%) were preferentially included in an attempt to increase the number of diabetics in the control group. Organisms were placed into clinically relevant groups that included; anaerobic bacteria; Enterobacteriaceae; Pseudomonas aeruginosa; Streptococci and Enterococci; methicillin sensitive Staphylococcus aureus (MSSA); methicillin resistant Staphylococcus aureus (MRSA); fungi; and enteric gram negative organisms known to produce inducible beta-lactamases (Enterobacter spp., Serratia spp., Citrobacter freundii, Acinetobacter spp., Proteus spp, Providencia spp., and Morganella morganii).
Potential confounders were recorded. The Charlson comorbidity index was used to quantify the extent of comorbid disease [1]. Time from blood culture collection to the administration of antibiotics to which the organism was subsequently found to be susceptible was recorded. The likely source of infection was determined from review of the clinical record and the results of laboratory and radiological investigations. The site (community, hospital or ICU) where the infection was acquired was recorded. Infections were considered to have been acquired in either the hospital or ICU if the first manifestation occurred greater than 48 hour following arrival in the respective area. Usual medications as well as those administered during the hospital stay were recorded.
Continued glitazone use was defined as missing less than 2 doses in the 7 days following the time the positive culture was collected.
Outcome Variables
Survival to hospital discharge was determined directly from review of the clinical record. 90-day mortality was determined from the clinical record and hospital and regional databases. The development of severe sepsis and septic shock was determined using the 1991 American College of Chest Physicians / Society of Critical Care Medicine Consensus Conference definitions [2]. The need for intensive care unit admission was recorded.
Statistical methods
The analysis was performed using conditional logistic regression. For each outcome we calculated the crude odds ratios (OR) for glitazone users versus controls. We then created multivariate models to assess the effect of potential confounders. Each model included diabetes, source (nosocomial versus community), Charlson score and time to antibiotics (hours) and was used to calculate adjusted ORs. Lastly we used bivariate analysis to identify differences between the two groups in terms of medications (using each medication as the outcome). All analyses were completed using R (version 3.0.2).
Results
We identified 14 472 positive cultures during the study period. Comparing this with pharmacy records we found 1083 positive cultures in 82 patients who had been prescribed a glitazone during this same time frame. Sixty – seven patients were excluded (positive culture a probable contaminant, not taking a glitazone at time of hospital admission or medical record unable to be located), leaving 15 patients in the final cohort who were taking a glitazone at the time of the bacteraemia (figure 1). Six (40%) of these patients were taking rosiglitzone and 9 (60%) were taking pioglitazone. Eleven (73%) had their glitazone continued through their hospital stay. From each of these cases we identified a pool of 54 controls for comparison.
Baseline characteristics of cases and control cohorts are presented in table 2. Mean age of the study was 60 years with a range of 25 to 81. Just over half were men. All glitazone patients had a history or diabetes compared to only 39% of controls. The mean Charlson comorbidity index of 3 (2 SD) was not different between the groups. Just over half the infections were acquired in the community. Only one patient in each group acquired their infection whilst in the ICU.
The main potentially confounding medications are listed in table 3. There was a high proportion in each group taking immunosuppressant medication (27% and 39% in cases and controls respectively). There was also a high proportion on statin therapy (52% and 31%). Cases were significantly more likely to be taking sulphonylureas, metformin and either angiotensin converting enzyme inhibitors or aldosterone receptor blockers.
Severe sepsis developed in approximately 40% of patients (40% of glitazone users and 39% of controls). Overall a fifth of patients were admitted to the intensive care unit (27% of glitazone users and 20% of controls), with no difference between the groups (table 1). Twenty percent of glitazone users developed septic shock compared to only 7% of controls however this difference was not statistically significant. Hospital mortality was similar in both groups (7% in cases and 11% in controls, p = 0.70). Odds ratios, adjusted for confounders (diabetes, source (nosocomial versus community), Charlson score and time to antibiotics) were not significantly different between the groups for any of the clinical outcomes (Table 1).
Discussion
Key findings
This study showed no difference in clinical outcomes from bacteraemia for a group of diabetics taking a glitazone compared to matched controls.
Strengths
The main strength of our study was the ability to include many of the potential confounding variables such as time to appropriate antibiotics, age, co-morbid disease burden, site of infection and organism. In addition we used important patient centred endpoints such as mortality, development of severe sepsis and need for intensive care unit admission.
Limitations
Our study was limited by the low number of patients taking glitazones resulting in reduced power to detect a meaningful difference in clinical outcomes. This is in stark contrast to other agents, such as statins that that have been extensively investigated for a possible benefit in sepsis by retrospective observational studies [3]. In addition glitazone use is restricted to patients with diabetes which limits both the comparison and generalizability of any findings to the general population with infection. Glitazone prescriptions are in decline [4, 5]. The low uptake of glitazones may be related to their cost, concerns about side effects, or a lack of evidence that they improve patient centred outcomes. This study had small case numbers and if the use of these agents in diabetes continues to decrease case control studies such as this will be increasingly difficult, if not impossible to perform. Additional clinical exploration of the promising suggestions of a role for glitazones in sepsis from animal models will therefore require different methodology and study design.
Diabetes is an obvious confounder in this study. However the relationship between diabetes and sepsis is complex and incompletely understood. Diabetics have increased susceptibility to infections [6, 7]. However the influence of diabetes on outcomes in patients with sepsis is much less clear. There have been conflicting results from observational studies, with some studies showing higher mortality in diabetics [8], others similar mortality [9] and yet others reduced mortality [10]. In addition diabetics have been found to have reduced rates of acute respiratory distress syndrome and higher rates of acute kidney injury [10, 11]. A recent meta-analysis found that critically unwell diabetic patients admitted to an intensive care unit had similar mortality to non-diabetics [12]. It is possible that either the diabetic state or medications used to treat diabetes could have a protective effect in the event that sepsis develops.
Blood glucose levels were not consistently measured in the study population and we are unable to determine the influence of glycaemic control on outcomes. Sepsis is associated with insulin resistance and hyperglycaemia even in non-diabetics (so called stress hyperglycaemia). Hyperglycaemia has many potentially deleterious effects including oxidative stress, endothelial dysfunction and inhibition of leukocyte function [13]. It is associated with worse outcomes in observational studies with many demonstrating a dose response relationship [14, 15]. However meta-analyses of randomised controlled trials have failed to confirm better outcomes from improved glycaemic control [16, 17]. Whether or not hyperglycaemia is simply a marker of disease severity or directly contributes to poor outcomes is thus yet to be definitively established. In addition the efficacy of glitazones in improving inpatient glycaemic control has not been rigorously studied.
We did not collect data on cardiovascular outcomes. Given the concerns about rosiglitazone’s cardiovascular safety [18], any future studies in this area should include cardiovascular complications.
The two glitazones studied here have important pharmacodynamic differences, which may affect their efficacy and side effect profiles. Rosiglitazone is a more potent PPAR gamma agonist [19] while pioglitazone is also a partial PPAR alpha agonist [20]. In addition different PPAR gamma agonists have been shown to lead to different patterns of gene expression [21, 22]. This could relate to different conformational changes in PPAR gamma following glitazone binding, leading to differing affinity of the PPAR gamma-glitazone complex for co-activators [23]. It is unclear how important these differences are in relation to their potential use in sepsis. However given that glitazones have the ability to regulate the expression of over one hundred genes across many different biological systems these potential differences require further investigation. Unfortunately we did not have patient numbers to explore this issue.
Comparison with other studies
Glitazones have shown promise in small animal studies. Several studies using rodents who have undergone cecal ligation puncture have found improved survival, reduced organ dysfunction and reduced pro-inflammatory cytokine levels in rodents who receive a glitazone either before [24-26], at the time of [27], or up to 6 hours after the septic insult [24]. Similar effects on cytokine levels and organ dysfunction have been found in rodent models using endotoxin injection [28, 29]. Several of these studies also found reduced leukocyte infiltration of tissues including the liver and lung [24], and the kidneys [28]. No large animal models have been used. There are many differences between animals model and patients and many potential sepsis treatments have failed to show clinical benefit in human trials following successful animal data [30].
The sole study of glitazone use for sepsis in humans is a phase I/II randomised control trial of 140 patients with Plasmodium falciparum malaria [31]. In this study a four-day course of rosiglitazone improved parasite clearance and reduced levels of interleukin 6 and monocyte chemoattractant protein 1. Some or all of the effect on cytokine levels may be explained by glitazone-induced up-regulation of CD-36, which mediates macrophage clearance of parasitized erythrocytes. It is not clear whether the effects seen in malaria can be translated to sepsis caused by other organisms.
A phase I/II study of pioglitazone examine the effect on pro-inflammatory biomarkers in adolescents with severe sepsis and septic shock is expected to be completed in October 2015 (Clinical Trials.gov Identifier: NCT01352182).
Conclusion
We found no difference in clinical outcomes in those taking glitazones when compared to matched controls, however our study was severely limited by the low number of cases. The utility of observational studies this area is likely to be limited given the small numbers of glitazone users and concerns around potential toxicity noted in the diabetic population.
Table 2. Baseline characteristics.
(n = 15) / Controls
(n = 54)
Age (years) - mean (SD) / 60 (12) / 61 (11)
Male Gender - n (%) / 8 (53) / 29 (54)
Nursing home resident - n (%) / 0 (0) / 1 (18)
Comorbid Disease
Charlson score - mean (SD) / 3.87 (2.00) / 3.03 (1.98)
Diabetes - n (%) / 15 (100) / 21 (39)
Abnormal HbA1c (>5.9%) - n (%) / 11 (73) / 13 (24)
Infection Details
Source - n (%)
Respiratory tract
Intra-abdominal
Urinary tract
Central line associated
Skin and soft tissue
Other / 1 (7)
1 (7)
5 (33)
4 (27)
2 (13)
2 (13) / 7 (13)
10 (19)
12 (22)
8 (15)
12 (22)
5 (9)
Organisms - n (%)
Methicillin sensitive Staphylococcus aureus
Methicillin resistant Staphylococcus aureus
Other gram positive cocci
Enterobacteriaceae
Pseudomonas aeruginosa
ESCAPPMs[1]
Anaerobes / 2 (13)
1 (7)
1 (7)
6 (40)
1 (7)
3 (20)
1 (7) / 8 (15)
4 (7)
3 (6)
21 (39)
3 (6)
11 (20)
4 (7)
Setting infection acquired - n (%)
Community
Hospital
Intensive care / 8 (53)
6 (40)
1 (7) / 29 (54)
24 (44)
1 (2)
Time to appropriate antibiotics - n, (%)
<1hr
1-5.99
6-11.99
12-23.99
>24hrs
Mean
Median
SD / 4 (27)
2 (13)
1 (6)
4 (27)
4 (27)
25
15
34 / 15 (28)
15 (28)
4 (7)
3 (6)
17 (31)
20
3
31
Table 3. Medications
n (%) / Controls
n (%) / Odds Ratio
(95% CI) / P value
Metformin / 6 (40) / 3 (6) / 0.11
(0.02 – 0.53) / 0.006
Sulphonylurea / 7 (47) / 8 (15) / 0.13
(0.03 – 0.66) / 0.014
Insulin / 5 (33) / 7 (13) / 0.27
(0.07 – 0.99) / 0.05
Statin / 8 (53) / 17 (31) / 0.40
(0.12 – 1.34) / 0.137
Antiplatelet / 7 (47) / 16 (30) / 0.47
(0.15 – 1.49) / 0.202
Angiotensin converting enzyme inhibitors / angiotensin receptor blockers / 12 (80) / 12 (22) / 0.12
(0.03 – 0.58) / 0.008
Beta-blockers / 4 (27) / 16 (30) / 1.00
(0.26 – 3.82) / 1.000
Immunosuppressant / 4 (27) / 21 (39) / 1.50
(0.38 – 5.86) / 0.569
Figure 1. Flow diagram identifying study case cohort.