Background
Acute exacerbations of chronic obstructive pulmonary disease (AECOPD) negatively affect hospitalisation, readmission, disease progression and mortality rates in patients with COPD [
1]. Severe AECOPD are mainly triggered by bacterial infection, viral infection or environmental agents, with the most common causes of bacterial infection being
Streptococcus pneumoniae, Haemophilus influenzae and
Moraxella catarrhalis [
2‐
4]. Thus, current recommendations for antimicrobial treatment are aminopenicillin with or without clavulanic acid, a macrolide or a tetracycline [
5,
6]. AECOPD are infrequently caused by microorganisms—such as
Pseudomonas aeruginosa, methicillin-resistant
Staphylococcus aureus (MRSA),
Stenotrophomonas maltophilia or enterobacteria—that are resistant to these treatments. Guidelines and previous studies of severe AECOPD suggest that these patients have increased frequencies of exacerbations, previous antibiotic use, previous hospital admissions and more severe airflow limitations [
2,
5,
7,
8].
At least 30% of COPD patients are colonised by a potential pathogen when in a stable phase of their disease; however only 0.5% are colonised by
Enterobacteriaceae,
P. aeruginosa or
S. maltophilia [
9]. Also, AECOPD are associated with the overgrowth of potential pathogens and with the occurrence of
P. aeruginosa in the lower airway [
10]. Knowing the risk factors to microorganisms resistant to conventional antibiotic treatment (MRCT) in AECOPD could lead to improved prophylaxis and empirical antimicrobial treatment.
We hypothesised that specific factors predict the presence of MRCT. Our primary aim was to identify the risk factors associated with infection by MRCT. Our secondary aim was to compare the clinical characteristics and outcomes of patients with AECOPD resulting from MRCT against those with AECOPD from other causes.
Methods
Study design and patients selection
This observational cohort study was performed between January 2009 and December 2015, and included all patients admitted with a diagnosis of AECOPD to the Respiratory Department of the Hospital Clinic, Barcelona, Spain. COPD was defined according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines [
5], with spirometry performed in a stable disease phase and at least six months prior to hospital admission. Patients with a smoking history of 10 pack-years were considered positive smokers. A worsening of respiratory symptoms compared with the preceding days, and which required a change in home care medication, was used as a clinical definition for AECOPD [
5,
11]. Exacerbation severity was based on the respiratory symptoms/signs and the presence of potential indications for hospitalisation [
5]. The exclusion criteria were: 1) documented history of asthma or bronchiectasis as the predominant illness and 2) clinical pneumonia or acute heart failure identified at admission.
Ethics approval and consent to participate
The study protocol was approved by the Hospital Research and Ethics Committee (CEIC 2008/4106) and the study was conducted in accordance with good clinical practice guidelines and the declaration of Helsinki. Written informed consent was obtained from all enrolled patients.
Microbiological evaluation
Sputum samples were obtained at admission for bacterial culture, before starting antibiotic therapy. Routine antimicrobial susceptibility testing included the disc diffusion method or E-test for
P. aeruginosa. The results of susceptibility testing were interpreted according to the European Committee on Antimicrobial Susceptibility Testing guidelines [
12]. Multidrug-resistant (MDR), extensively drug resistant (XDR) and pan-drug resistant bacteria were categorised according to criteria set out by Magiorakos et al. [
13]. The quality of sputum samples was assessed using the Murray and Washington scoring system. Patients with poor quality sputum samples (> 10 epithelial cells or < 25 leucocytes) were excluded from analysis [
14]. Patients with mycobacterial, fungal isolation (e.g., Aspergillus or Candida) or
Nocardia spp. were also excluded.
Patients were classified into 3 groups: 1) patients with the isolation of microorganism sensitive to conventional treatment (MSCT) according to GOLD guidelines (i.e., aminopenicillin with clavulanic acid, a macrolide or a tetracycline); 2) patients with MRCT isolation, (i.e., P. aeruginosa, MRSA, S. maltophilia, Enterobacteriaceae producer of extended spectrum of beta lactamase and Acinetobacter baumannii); and 3) patients with negative microbiology results who did not receive antibiotics in the 7 days previous at admission.
Previous antibiotic treatment was not considered as inclusion/exclusion criteria in MRCT or MSCT groups. Nobody patient used macrolide as chronic treatment, thus it was not considered as variable.
Clinical measurements and outcomes
Demographic variables, body mass index (BMI), smoking history (former smoker was considered as those patients who quit smoke more than 12 months), presence of co-morbidities measured by Charlson index [
15], baseline dyspnoea grade based in modified medical research council (mMRC), COPD severity score measured by a questionnaire (COPDSS) [
16] and BODEx index (i.e., BMI, airflow obstruction, dyspnoea and exacerbations) [
17], use of long-term oxygen therapy (LTOT) and use of domiciliary medications (i.e., inhaled bronchodilators, such as short-acting β
2 agonist [SABA], long-acting β
2 agonist [LABA], anticholinergics or inhaled corticosteroids) were recorded at hospital admission. Characteristics of any exacerbations during the previous year, any previous antibiotic treatment (3 months before admission) and any microorganism isolated in the previous year were also recorded. Vital signs (body temperature, respiratory rate, heart rate and blood pressure) were assessed at admission. Arterial blood gases and laboratory parameters (i.e., leukocytes, haematocrit, haemoglobin, C-reactive protein, glucose and creatinine) were recorded at admission and at day 3.
Variables relating to clinical progression included length of hospital stay (LOS), use of non-invasive mechanical ventilation (NIMV), use of invasive mechanical ventilation (IMV) and intensive care unit (ICU) admission during the initial hospitalisation. Data on prognosis (cumulative number of deaths for all-causes and time to death) were recorded at 30 days, 1 year and 3 years.
Statistical analysis
We report the number and percentage of patients for categorical variables and the median and interquartile range (IQR) for continuous variables. Categorical variables were compared using the chi-square test, and continuous variables were compared by one-way analysis of variance or the nonparametric Kruskal–Wallis test. Post-hoc pairwise comparisons were carried out via the Bonferroni method to control for the experiment-wise error rate. Survival curves were obtained using the Kaplan–Meier method and compared using the Gehan-Breslow-Wilcoxon test. Patients lost to follow-up were censored in the survival analysis.
Multinomial logistic regression analyses were used to examine the associations between microbial aetiology groups (i.e., MRCT or MSCT relative to unknown aetiology) and risk factors (i.e., baseline characteristics and clinical presentation). Variables were included in the multivariate model when univariate comparisons yielded a level of significance of
p < 0.05 due the limited number of patients in the MRCT and MSCT groups and in order to exclude bias related to overestimation or underestimation of regression coefficient variance. The final multivariate model was calculated in a stepwise forward selection procedure (p
in = 0.05, p
out = 0.10). To identify the problem of collinearity, we calculated the r coefficient of 2 variables; that is, if 2 independent variables were highly correlated (r > | ± 0.30|), the variable with the largest variance was excluded from the multivariate analysis [
18]. The association between LOS and risk factors was also tested in simple and multiple analyses, and similar inclusion criteria were applied for the linear regression analysis (
p < 0.05). The odds ratios (ORs) or beta coefficients (βs) and their 95% confidence intervals (CIs) were estimated. The Cox and Snell R
2 and the Nagelkerke R
2 were calculated to assess the overall fit of the multinomial logistic regression model and the R
2 for the linear regression model. The area under the receiver operating characteristic (ROC) curve of the multivariate model to predict MRCT was calculated. Internal validation of the prediction models was conducted using ordinary nonparametric bootstrapping with 1000 bootstrap samples and bias-corrected, accelerated, 95% CIs [
19]. The same logistic regression analyses for microbial aetiology groups were also performed but using a multinomial logistic regression model for MRCT with only P. aeruginosa or MSCT relative to negative microbiology.
We investigated the missing data patterns for covariates, assumed missing at random [
20], and used multiple imputation [
21] to generate 5 datasets to evaluate the prediction performance for the microbial aetiology group. The model for multiple imputation included all covariates of the risk models as well as the microbial aetiology group. For simplicity, in the evaluation of the performance we filled in missing values with the first set of imputed values from the multiple imputation.
The level of significance was set at 0.05 (two-tailed). All analyses were performed with IBM SPSS Statistics 23.0 (Armonk, New York).
Discussion
In our study, we analysed 3 well-characterised groups, comparing patients with MRCTs against controls groups of patients with MSCTs and patients with negative microbiology and no previous antibiotic use. Our analyses revealed that not currently smoking, ≥ 2 AECOPD episodes or ≥ 1 admission for AECOPD in the last year, and a low systemic inflammatory response at admission were independent risk factors for AECOPD caused by an MRCT. However, although patients with MRCT had longer hospital stays, they did not have higher mortality or more severe AECOPD than the control groups. At baseline, patients with MRCT had more severe disease, as measured by the dyspnoea scale, COPDSS scale, BODEx index and history of previous AECOPD. There were no differences in symptoms or pulmonary gas exchange features at admission.
AECOPD are events that mark disease progression, and as taken into account by the GOLD guidelines [
5], are as important as airflow limitation. Indeed, it is evident that there are patients who are susceptible to frequent exacerbations, and in these, the most important predictor of future episodes is the history of AECOPD [
22]; however, the association with microbiologic data has been poorly analysed to date.
A low systemic inflammatory response was a risk factor for MRCT isolation in this study, which could be due to lower virulence or reduced ability to produce acute phase reactants in the presence of these microorganisms. Similar results were observed in patients with community acquired pneumonia or ventilator associated pneumonia in whom
P. aeruginosa was isolated [
23,
24].
There was also an association between smoking status and MRCT isolation, specifically in favour of non-current smoking status. It is known that smoke increases upper respiratory tract colonisation of
S. pneumoniae, H. influenzae, M. catarrhalis and
Streptococcus pyogenes [
25], and that smoking facilitates colonisation of the lung with these bacteria [
9]. This is probably related to the decreased phagocytic ability of alveolar macrophages and the decreased cytokine response associated with smoking [
26,
27]. The association between smoking status and MRCT isolation in this study should not, therefore, weaken the recommendation for smoking cessation for all patients. Other explanations to this point maybe those individuals who develop respiratory symptoms due to more severe disease being more likely to quit smoking.
Previous studies have produced controversial data about the presence of
P. aeruginosa in isolates, though they have tended to show that sensitive
P. aeruginosa had higher mortality [
28‐
31]. We found no differences in mortality between patients with
P. aeruginosa, including those with MDR strains. In other respiratory diseases, such as cystic fibrosis or non-cystic fibrosis bronchiectasis, microbiologic isolation of
P. aeruginosa and MRSA has been shown to have an important role in disease progression [
32‐
36]. Although there is evidence that eradication with antibiotic treatment would be beneficial in these diseases, there is no such evidence that similar benefits would exist for patients with COPD.
The role of antimicrobial treatment remains controversial in AECOPD. With the exception of patients who require mechanical ventilation and ICU admission, the benefits of antibiotic treatment are limited, and are mainly observed in patients with purulent sputum or in those with AECOPD graded as type I by the Anthonisen classification [
37‐
41]. The effect of inadequate antibiotic treatment is poorly understood in patients with AECOPD. In this study, we did not observe any differences in outcomes between patients with inadequate and adequate empiric antibiotic treatment.
The predictive factors identified in this study represent the first step in the development of a prediction model. To move forward, the potential model will need to undergo external validation with larger patient cohorts from multiple centres. We could also apply the results of internal validation techniques to understand how likely this model will be replicable to future studies and to studies at other centres. Bootstrapping techniques were applied to our data, and the results indicated that the coefficients obtained from the prediction model were quite robust. Notably, previous P. aeruginosa isolation was the one factor that the bootstrap analysis indicated might have limited repeatability in future work. Thus, we opted to remove previous P. aeruginosa isolation from the overall model and include it in a specific multivariate analysis for P. aeruginosa. In the real-world clinical setting where this prediction model could be used, previous P. aeruginosa isolation remains an important clinical characteristic that can play a substantial role in decision making.
We did not observe differences in the majority of outcomes when comparing MRCT vs non-MRCT exacerbations. However; length of stay was longer in the MRCT group. This is an important outcome to be taken into account to make efforts in predicting and treating these microorganisms in AECOPD.
Our study has some limitations that should be acknowledged. First, the study was carried out at only one centre in Spain. Second, the small sample limited the analysis of specific factors per bacterium. There is limited information about MRCT isolation in patients with AECOPD, and where there is, it is mainly for bacteria other than
P. aeruginosa. A confirmation of our results in a large and well balanced, international cohort of AECOPD is therefore desirable. Finally, other limitation of this study was the use of sputum cultures and the potential difficulty to distinguish between colonization and infection. However, we only accepted samples of good quality and we did not culture those of low quality. In addition this is the usual way to diagnose lower airway infection in AECOPD in the majority of studies, given that performing bronchoscopy in these patients is extremely difficult. Moreover we validated sputum cultures some years ago in comparison with bronchoscopic samples [
42].
Acknowledgements
We are indebted to all medical and nursing colleagues for their assistance and cooperation in this study. We thanks to Michael Maudsley by his assistance in language review.