Background
Community-acquired pneumonia (CAP) is a common disease that may become severe, leading to admission to intensive care units (ICU) [
1]. CAP etiology is usually bacterial; however, the causative role of respiratory viruses emerged recently [
2]. Multiplex polymerase chain reaction (mPCR) kits screen a large panel of respiratory viruses, and nowadays are available in clinical practice. They were used within several studies among adult ICU patients with CAP [
3‐
6]. High rates of positivity were reported, up to 49 % [
4], with strong variations in the distribution of viral species according to the population, the season, and the geographic area. However, the causative role of respiratory viruses identified in the respiratory tract during pneumonia is still debatable, since respiratory viruses might be present in asymptomatic adult subjects [
7,
8]. Some experimental data focusing on virus-bacteria interactions during respiratory tract infections supported a pathogenic role of respiratory viruses during pneumonia [
9]. In mice, the coinfection of influenza with
S. pneumoniae [
10],
L. pneumophila [
11] or
S. aureus [
12] impaired the anti-influenza immune response and increased the mortality. Similar synergistic results are obtained with
S. pneumoniae and respiratory syncytial virus [
13], or
S. pneumoniae and rhinovirus [
14].
In humans, the pathogenic role of respiratory viruses in virus-bacteria coinfected patients remains unclear. We conducted a comprehensive observational study among adult ICU patients with CAP, to compare clinical characteristics, biological presentation, and outcome according to the presence of virus in the respiratory tract.
Methods
Study design and patient selection
We conducted a retrospective monocenter observational study in the 26-bed ICU of the Bichat Claude Bernard University Hospital (Paris, France). During the study period, all consecutive patients having undergone a mPCR in the respiratory tract within the 72 hours following their ICU admission were screened. Medical records were independently reviewed by two physicians. All patients with a final diagnosis of pneumonia were included (see definitions for population selection in Additional file
1).
Data collection
AT ICU admission and during ICU stay, data regarding demographics, comorbidity, clinical examinations, laboratory and radiological findings, microbiologic investigations, and therapeutic management were collected (for details, see Additional file
1). Mortality was defined as death from any cause within 30 days of hospitalization.
Pneumonia severity was assessed through the Pneumonia Severity Index (PSI) [
15], and the Simplified Acute Physiologic Score (SAPS) II [
16].
Microbiological evaluation
Respiratory tract specimens underwent Gram staining and quantitative culture for bacterial pathogens. Urine antigen testing of S. pneumoniae and L. pneumophila used the BinaxNOW kits (Alere, Jouy en Josas, France). The immunoglobulin (Ig) antibodies testing for C. pneumoniae and M. pneumoniae was considered positive if IgM antibodies were identified or if a significant increase in IgG antibodies was observed between paired serum samples.
The respiratory mPCR were performed either in nasopharyngeal (NP) swabs or in lower respiratory tract (LRT) specimens, usually bronchoalveolar lavage fluid otherwise endotracheal aspirate. During the study period, different mPCR kits were used (for details, see Additional file
1). Respifinder® 19 (Pathofinder, Maastricht, The Netherlands) and Filmarray Respiratory Panel (BioFire Diagnostics, Salt Lake City, UT, USA) could not detect bocavirus nor differentiate rhinovirus and enterovirus; therefore, rhinovirus and enterovirus results were grouped as picornavirus (rhinovirus).
Either in blood samples or in bronchoalveolar lavage fluid, the cytomegalovirus PCR used the CMV R-gene® kit (Argene, Verniolle, France ) or the QS-RGQ® kit (Qiagen, Hilden, Germany), and the herpex simplex virus PCR used LightCycler® HSV (Roche, Basel, Switzerland).
Classification of patients according to pathogens
A bacterium was considered as a causative pathogen of the pneumonia if this bacterium fulfilled at least one criterion (for details, see Additional file
1). A virus identified with PCR was always considered as a causative pathogen of the pneumonia.
Pneumonia was defined as: (i) bacterial, if microbiological investigations revealed at least one bacterium and no virus; (ii), viral, if microbiological investigations revealed at least one virus and no bacterium; (iii) mixed (virus-bacteria), if microbiological investigations revealed at least one virus and one bacterium; and (iv) no etiology, if microbiological investigations revealed no virus and no bacterium.
Endpoints
The primary endpoint was to identify presentation and prognosis-specific features in virus-bacteria coinfected patients. Comparisons focused on microbiological data, biological findings and radiological patterns on admission, ICU course and hospital outcome. A composite criterion named “complicated course” included hospital death or mechanical ventilation for more than 7 days. The second endpoint was to describe the epidemiology of respiratory viruses in adult patients admitted to the ICU for a CAP. Patients were clustered into four groups according to the microbiological etiology of pneumonia: bacterial, viral, mixed, and no etiology.
Matching procedure
To better control for impact of the bacterial pathogen on our main findings, we also designed a subgroup analysis comparing patients with bacterial and mixed viral-bacterial CAP, matched on the bacterial pathogen. If more than one bacterium was identified in cases, we sought for a control with the same bacterial combination.
Data presentation and statistical analysis
Continuous data were expressed as median [first through third quartiles] and were compared using the Kruskall-Wallis test followed by pairwise Mann-Whitney test. Categorical data were expressed as number (percentages) and were evaluated using the chi-square test or Fisher’s exact test. p values less than 0.05 were considered significant. A univariate logistic regression with clinically relevant variables was used to identify variables associated with a complicated course. A multivariate conditional logistic regression, including variables with p value less than 0.10 in the previous step, was used to identify variables independently associated with complicated course. Similar statistical analyses were performed to identify variables independently associated with hospital death and mechanical ventilation for more than 7 days in survivors at day 28. Quantitative variables that did not validate the log-linearity assumption were transformed into categorical variable according to their median value. Missing data were imputed to the median or to the more frequent value. The accuracy of the final model was tested using area under the receiver operating characteristic curve analysis and the Hosmer-Lemeshow chi-square test. An additional multivariate conditional logistic regression, limited to bacterial and mixed groups, was performed to search specifically for an association between virus-bacteria coinfection and complicated course. Comparisons in the subgroup analysis of bacteria-matched patients involved univariate conditional logistic regression followed by multivariate conditional logistic regression to assess associations between microbiological diagnosis and complicated course, adjusting for clinically relevant variables. Analyses were performed using the SAS software package (SAS Institute, Cary, NC, USA).
Discussion
This retrospective study investigated the impact of the mixed viral-bacterial coinfection on the presentation and outcome of ICU patients with CAP. Real-time mPCR tests identified at least one virus in the respiratory tract of 56.3 % of patients. Specific biological and radiographic features, including high serum levels of creatine kinase, a low platelet count, and a high incidence of alveolar-interstitial infiltrates were observed in these patients, who presented also the higher status severity on hospital admission and the higher frequency of hemodynamic and respiratory failures during ICU stay. The viral-bacterial coinfection was independently associated with a complicated course. These findings were confirmed by a subgroup analysis comparing bacteria-infected and virus-bacteria coinfected patients.
In this study, more than one patient out of two (56.3 %) were infected with at least one virus, in line with a recent report on ICU ventilated patients with CAP [
4]. This finding illustrated the high yield of an aggressive diagnostic strategy with a broad panel mPCR on respiratory tract specimens. Elsewhere, the rate of viral documentation reported in adult ICU patients with CAP was slightly lesser, from 23 to 49 % [
3‐
5].
Respiratory tract specimens for bacterial test were recovered in a high proportion of patients (87.9 %), including a LRT specimen in 68.4 % of patients. Bacterial documentation was obtained in 52.3 % of patients, in the range of other studies on ICU patients with CAP (36–82 %) [
3,
4]. Interestingly, the rate of bacteremia, 6.3 %, was markedly lower than usually observed [
17,
18]; it might be attributable to a high pre-referral exposure to antibiotics (44.3 %). Overall, 82.8 % of patients had a microbiological diagnosis, a high rate in line with that of Karhu and colleagues [
4].
The predominant viruses were influenza viruses and picornavirus (rhinovirus) (21.8 % and 12.6 %, respectively). Previous studies in ICU patients with CAP reported a much lower rate of influenza infection, from 2 to 10 % [
3‐
6]. It might be explained by low influenza vaccine coverage in our population. Unfortunately, the influenza vaccination history was usually not available in medical records, preventing any conclusion on this point. The incidence of rhinovirus was consistent with that of previous reports [
3,
5]. Interestingly, a “viral phenotype” is emerging, mainly characterized by high serum levels of creatine kinase, and high frequency of alveolar-interstitial infiltrates on chest X-ray. It was not yet reported, since previous works having largely used mPCR in ICU patients with CAP did not specifically record biology and radiographic data [
3‐
6]. Chest X-ray patterns, notably ground glass opacities, were consistent with previous data [
19]. Elevated levels of serum creatine kinase could be attributable to the rhabdomyolysis associated with viral infections, especially with influenza and parainfluenza viruses [
20]. The trend toward a high serum level of cardiac troponin T in virus-infected patients might suggest acute myocardial injury. Furthermore, alveolar-interstitial infiltrates could indicate pulmonary edema. Unfortunately, brain natriuretic peptide dosage and echocardiogram data were not available, although left ventricular dysfunction has been previously described in influenza A infection [
21].
Virus-bacteria coinfection was observed in one patient out of four, consistently with previous reports (9–39 %) [
3,
4,
22]. It was identified as independently associated with a complicated course. This finding is original, since previous works that studied CAP patients requiring ICU admission did not point out any relationship between viral-bacterial coinfection and severity [
3,
4]. Karhu and colleagues studied a limited cohort (n = 49), whereas Choi and colleagues observed a low rate of virus-bacteria coinfected patients in their cohort (18/198, 9 %), thus preventing any analysis on outcome in both studies. In hospitalized patients with CAP, some data suggested that the viral-bacterial coinfection might be associated with high-risk classes of PSI [
23], higher length of hospital stay [
24], and higher mortality [
25]. More specifically,
S. pneumoniae-influenza and
S. pneumoniae-rhinovirus coinfections were correlated with a more severe illness in hospitalized patients [
22,
26,
27]. In our study, in order to avoid overinterpreting the data, we decided to consider respiratory viruses as a homogeneous group of pathogens. This might be criticized, since the pathogenicity probably differs from one viral species to another. However, our results remained similar when taking into account non-influenza viruses only. Further studies with larger populations may explore this point, with comparing the prognosis of CAP patients according to the type of virus as well as the different virus-bacteria combinations. Our findings might have a therapeutic impact. Currently, only very few medications targeting non-influenza respiratory viruses are available in clinical practice (cidofocir, ribavirine, immunoglobulins). But some novel antiviral drugs, targeting mainly respiratory syncytial virus and parainfluenza virus, have shown promising results in immunocompromised patients [
28,
29] and in human volunteers [
30]. Whether these upcoming drugs would be of interest in virus-infected CAP patients requiring ICU admission is questionable. We identified the virus-bacteria coinfected patients to be at risk of complicated ICU course, so further studies might explore potential benefits of the upcoming antiviral drugs in this high-risk population.
Our study has several limitations. First, this is a monocenter study, so the generalization of our results should be cautious. Second, this study included patients with pneumonia that required ICU admission. It means that we did not study mild to moderate pneumonia, preventing any conclusion on this population. Third, the study was retrospective so we did not control the microbiological investigations. By definition, a mPCR was performed in the respiratory tract of every included patient because it was the criteria for patient screening. But some other microbiological testings were only occasionally performed, i.e. cytomegalovirus and herpes simplex virus PCR, or
C. pneumoniae and
M. pneumoniae antibodies testing. Furthermore, the retrospective design prevented us obtaining a number of data, which were rarely reported in medical records by physicians, including vaccine history, symptoms before hospital referral, and duration of symptoms before ICU admission. Fourth, only patients having undergone a mPCR in the respiratory tract within the 72 hours following their ICU admission were screened; this might suggest a confounding of indication. Fifth, we chose a composite endpoint, decided a priori. Indeed, considering the predictable low hospital mortality, we did not choose hospital death as primary endpoint, because the low frequency of the event (death) would have favored the absence of significant difference between groups. Sixth, we made the assumption that a virus identified with PCR was a causative pathogen of the pneumonia. This might be criticized since respiratory viruses might be present in asymptomatic adult subjects [
7,
8]. Seventh, almost half the patients (n = 77, 44.3 %) received antibiotics prior to ICU admission. Thus some patients may have had false-negative findings regarding bacterial infection and may have been falsely included in the viral group instead of the mixed group or to the no etiology group instead of the bacterial group.
Acknowledgements
The authors are very grateful to the Assistance Publique - Hôpitaux de Paris (AP-HP) for their participation to the publication costs.