Etiology of community-acquired pneumonia and diagnostic yields of microbiological methods: a 3-year prospective study in Norway

Background Despite recent advances in microbiological techniques, the etiology of community-acquired pneumonia (CAP) is still not well described. We applied polymerase chain reaction (PCR) and conventional methods to describe etiology of CAP in hospitalized adults and evaluated their respective diagnostic yields. Methods 267 CAP patients were enrolled consecutively over our 3-year prospective study. Conventional methods (i.e., bacterial cultures, urinary antigen assays, serology) were combined with nasopharyngeal (NP) and oropharyngeal (OP) swab samples analyzed by real-time quantitative PCR (qPCR) for Streptococcus pneumoniae, and by real-time PCR for Mycoplasma pneumoniae, Chlamydophila pneumoniae, Bordetella pertussis and 12 types of respiratory viruses. Results Etiology was established in 167 (63%) patients with 69 (26%) patients having ≥1 copathogen. There were 75 (28%) pure bacterial and 41 (15%) pure viral infections, and 51 (19%) viral–bacterial coinfections, resulting in 126 (47%) patients with bacterial and 92 (34%) patients with viral etiology. S. pneumoniae (30%), influenza (15%) and rhinovirus (12%) were most commonly identified, typically with ≥1 copathogen. During winter and spring, viruses were detected more frequently (45%, P=.01) and usually in combination with bacteria (39%). PCR improved diagnostic yield by 8% in 64 cases with complete sampling (and by 15% in all patients); 5% for detection of bacteria; 19% for viruses (P=.04); and 16% for detection of ≥1 copathogen. Etiology was established in 79% of 43 antibiotic-naive patients with complete sampling. S. pneumoniae qPCR positive rate was significantly higher for OP swab compared to NP swab (P<.001). Positive rates for serology were significantly higher than for real-time PCR in detecting B. pertussis (P=.001) and influenza viruses (P<.001). Conclusions Etiology could be established in 4 out of 5 CAP patients with the aid of PCR, particularly in diagnosing viral infections. S. pneumoniae and viruses were most frequently identified, usually with copathogens. Viral–bacterial coinfections were more common than pure infections during winter and spring; a finding we consider important in the proper management of CAP. When swabbing for qPCR detection of S. pneumoniae in adult CAP, OP appeared superior to NP, but this finding needs further confirmation. Trial registration ClinicalTrials.gov Identifier: NCT01563315. Electronic supplementary material The online version of this article (doi:10.1186/s12879-015-0803-5) contains supplementary material, which is available to authorized users.


Background
Community-acquired pneumonia (CAP) is a common disease and a significant cause of morbidity and mortality worldwide [1][2][3]. Differences in epidemiology of pathogens make the knowledge of local etiology crucial for the appropriate choice of empirical antimicrobial treatment, which has a major impact on the prognosis of the patient [4][5][6][7]. An important rationale for microbial testing is to enable pathogen-directed therapy, and thus avoid unnecessary antibiotic use [8]. In clinical practice causative pathogens often remain unknown, and the majority of patients are treated empirically [8]. Moreover, published data on etiology of CAP in Norway [9], as well as in several other countries, are hampered by old data, limited samples and detection methods confined to bacterial culture.
The use of diagnostic methods based on polymerase chain reaction (PCR) have resulted in increased detection of many bacterial and viral pathogens associated with CAP [8]. One of the advantages of PCR is its ability to identify respiratory pathogens also after initiation of antibiotic therapy [10]. Some recent studies have shown that a high microbial yield can be achieved when realtime PCR assays are combined with conventional diagnostic methods like bacterial cultures, urinary antigen assays and serology [11][12][13][14][15][16]. Such extensive approaches have led to recognition of viruses as important causes of CAP and that coinfections are more common than previously thought [17,18].
The aim of this study was to describe etiology of CAP in adults admitted to a general hospital in South-Eastern Norway using conventional methods and real-time PCR assays for the detection of respiratory pathogens. We also investigated the diagnostic yields of the methods applied, especially the potential benefit of PCR.

Patients and study design
Adults (≥18 years old) with suspected pneumonia admitted to Medical Department, Drammen Hospital, Vestre Viken Health Trust-an acute care 270-bed general hospital in the province of South-Eastern Norway serving a mixed rural and urban population of about 160 000 inhabitants-were consecutively evaluated for inclusion in a prospective study from January 2008 to January 2011. CAP was defined as (i) presence of a new pulmonary infiltrate on chest radiograph, (ii) rectal temperature >38.0°C, and (iii) at least 1 of the following symptoms or signs: cough (productive or nonproductive), dyspnea, respiratory chest pain, crackles or reduced respiratory sounds. These criteria had to be present within 48 h of admission. Patients were excluded from the study when the chest radiographic examination showed noninfectious causes such as pulmonary infarction, tumor or bronchiectasis, or if the patient was discharged from the hospital within the 2 weeks prior to admission. All study participants were invited to an outpatient follow-up 6 weeks after hospital discharge, including chest radiography and a convalescent blood sample. Consent was obtained at the time of recruitment. The study was approved by the Regional Committee for Medical and Health Research Ethics in South-Eastern Norway (reference number: S-06266a).

Clinical data collection
Clinical data were recorded on paper case record forms with missing data abstracted from medical records. Chest radiographic patterns were thoroughly examined by an independent experienced radiologist.

Microbial sample collection
Blood cultures (BD BACTEC™ Blood Culture System, Sparks, Maryland, USA) were obtained prior to commencing antibiotic therapy. Clinicians at the emergency room or on the wards collected nasopharyngeal (NP) swab samples-and, when possible, sputum specimens for Loeffler stain, Gram stain (optional) and culture-also before initiation of antibiotic therapy. In some cases, these samples were collected at the time of recruitment within 48 h after hospital admission. An additional NP swab and oropharyngeal (OP) swab from posterior oropharynx and both tonsillar pillars were collected for PCR analysis on separate transport media (Copan Flocked Swabs and UTM-RT Transport Medium System, Brescia, Italy). Urine sample was collected for specific antigen detection. Bronchoalveolar lavage (BAL) and diagnostic thoracentesis were performed on medical indication as judged by the treating clinician, usually when the pneumonia did not resolve, or-in the case of thoracentesis-the cause of build up pleural fluid was not known. Acute-phase serum samples were collected at the time of recruitment. Convalescent-phase serum samples were obtained at the time of outpatient follow-up, approximately 6 weeks after hospital discharge.

Conventional microbiological methods
Bacteriological specimens were cultured on standard media. BAL was cultured quantitatively in accordance with accepted methods [19]. Growth of specific pathogens detected in sputum and NP samples were judged semiquantitatively as low, intermediate or abundant. The most purulent part of sputum samples was processed for microscopy. Only samples displaying >25 polymorphonuclear leukocytes and <10 squamous epithelial cells per 100× power field were considered acceptable for culture [20]. For serology, commercially available assays were used according to the manufacturer's instructions for detection of Mycoplasma pneumoniae, Chlamydophila pneumoniae, Bordetella pertussis, and influenza A and B viruses (see Additional file 1). Urinary antigen detection tests for Streptococcus pneumoniae and Legionella pneumophila serogroup 1 were performed with the BinaxNOW pneumococcal urinary antigen test and the BinaxNOW Legionella urinary antigen test (Binax, ME, USA).

PCR
All NP and OP samples were analyzed by real-time quantitative PCR (qPCR) for detection of S. pneumoniae using primers specific for the pneumolysin (ply) gene as described by Greiner

Complete sample collection definition
A complete sample collection constituted the collection of blood, sputum and NP samples for culture; NP and OP samples analyzed for S. pneumoniae, M. pneumoniae, C. pneumoniae, B. pertussis and 12 types of respiratory viruses by use of PCR; serological testing for M. pneumoniae, C. pneumoniae, B. pertussis, and influenza A and B viruses; and urine antigen assays for detection of pneumococcal and L. pneumophila antigens.

Classification of etiology
Etiology was considered to be definite if any of the following criteria were met: a microorganism was cultured from blood or pleural fluid; BAL yielded growth of >10 5 colony-forming units per milliliter (cfu/mL) of one microbial species; urinary antigen test for S. pneumoniae or L. pneumophila was positive; or real-time PCR detection of L. pneumophila or P. jirovecii in sputum samples or BAL was positive.
Etiology was considered to be probable if any of the following criteria were met: bacteria cultured from sputum or NP swabs (at least intermediate growth was demanded for bacteria other than S. pneumoniae where any quantity of growth was accepted); real-time PCR detection of M. pneumoniae, C. pneumoniae, B. pertussis or a respiratory virus was positive; serological diagnosis of M. pneumoniae, C. pneumoniae or influenza viruses by CFT seroconversion (i.e., <10 to ≥20 or conversely), or a 4-fold titer rise or fall, or a high CFT titer (i.e., M. pneumoniae ≥160, C. pneumoniae ≥80, influenza A virus ≥160, influenza B virus ≥80) of either one or both of acute-and convalescentphase samples, or presence of IgM antibodies ≥840 U/mL for M. pneumoniae or ≥1.1 cut-off-index for C. pneumoniae; or serological diagnosis of B. pertussis by a 2-fold rise of PT-IgG (to >30 IU/mL in convalescent-phase serum) or fall (from >30 IU/mL in acute-phase serum), or presence of PT-IgG ≥80 IU/mL in acute-and/or convalescent-phase samples [22,23]. Pneumococcal DNA from NP samples [24] corresponding to ≥10 5 cfu/mL [25] by use of qPCR was also considered to be of probable significance. The same conclusions were drawn for OP samples based on the results from Principi et al. [26]. A case was considered coinfected if >1 pathogen, classified according to the above-mentioned criteria, was found.

Statistical analysis
Data were analyzed using PASW statistics software, version 20.0 (IBM SPSS, Chicago, IL). Categorical data were summarized using frequency counts and percentages. Continuous data were presented as mean and standard deviation (SD). The McNemar's test with Yates' continuity correction for paired proportions and the kappa statistics were used to compare the different techniques with respect to the percentage with positive results, and to evaluate the degree of agreement, respectively. Methods were selected for comparison based on level of diagnostic significance according to the classification; i.e., only techniques of which a positive result would give the same level of significance (definite or probable) were compared. In case of zero cells, the value of 0.5 was applied to each cell of the table in order to be able to perform the calculations. Kappa value <0.20 is poor, 0.21-0.40 fair, 0.41-0.60 moderate, 0.61-0.80 good, and kappa value >0.81 is very good agreement. For other category variables, groups were compared by use of Pearson's chi-square test or Fisher's exact test. In all instances a P of < .05 (2-tailed) was considered statistically significant.

Patients
Of the 320 screened patients, 267 patients were included in the study ( Figure 1). Median age was 66 years (range 19-100 years) ( Table 1). The average stay in the hospital was 6.8 days (range 1-75 days). Forty-eight (18%) patients were admitted to intensive care unit. The overall 30-day mortality rate was 3.7% (i.e., 10 of 267 patients died).

Microbiological etiology of CAP
A definite or probable etiology was established, by all methods, in 167 (63%) of 267 patients, 58 (22%) of whom diagnosis was definite. Bacterial etiology was established in 126 (47%) of patients; S. pneumoniae was most commonly detected ( Table 2). Twenty-five patients (9%) had bacteremia; 21 (84%) of these cases had S. pneumoniae isolated of which 2 cases had additional findings of either Haemophilus influenzae or Moraxella catarrhalis. One case had both H. influenzae and Haemophilus parainfluenzae isolated from blood, and Escherichia coli, Pseudomonas aeruginosa and group A streptococci was isolated from one case each. There were 7 cases of legionellosis (3 females and 4 males; age range 41-72 years), all caused by L. pneumophila serotype 1. Six of these patients had a documented travel history within past 4 weeks from a Mediterranean country (5 cases) or Asia (1 case), and 1 patient was probably infected while working on a supplyboat in the North Sea. Viral etiology was established in 92 (34%) of 267 patients; influenza viruses were most frequently detected followed by rhinovirus (Table 3). One patient had Pneumocystis jirovecii detected in BAL by use of real-time PCR.

Microbial patterns in CAP patients with an established etiology
A total of 251 pathogens were detected in the samples from 167 patients, the distribution patterns of the microbial agents identified in these mono-and coinfected patients are presented in Table 4. Multiple pathogens were detected in the samples from 69 (41%) cases; 1 additional pathogen (copathogen) in 54 (32%), and 2 copathogens in 15 (9%) cases. S. pneumoniae was identified in 44 (64%) of the 69 coinfected cases. Also, ≥1 copathogen was detected in 44 (54%) of 81 episodes with S. pneumoniae. This copathogen was due to 1 additional bacterial agent in 15 (19%) of cases; and/or ≥1 viral agent in 36 (44%). Accordingly, 36 (82%) of the 44 coinfected S. pneumoniae episodes were attributed to ≥1 viral agent. A pure bacterial etiology was established in 75 (45%) of the 167 patients; and a pure viral etiology in 41 (25%). Viral-bacterial coinfections were established in 51 (31%) of patients; the most common combinations were S. pneumoniae with influenza viruses (16 of 51 patients, 31%) or rhinovirus (14 of 51, 27%) (Additional file 1: Table S1).

Patients screened (n = 320)
Patients screened eliglible for participation in the study (n = 287)
The diagnostic yields obtained by the diverse methods are shown (Figure 2A and B). The corresponding contribution of each method for establishing bacterial or viral etiology in the study population is illustrated in Table 3 and Table 4, respectively. A pairwise comparison of the diagnostic yield obtained by each technique revealed interesting findings (Additional file 1: Table S2). First, OP swabs revealed significantly more positive results (34/235 positive, 14.5%) than NP swabs (14/235, 6.0%) for detection of S. pneumoniae by qPCR, P < .001. Secondly, for detection of B. pertussis serology exhibited significantly more positive results (13/256, 5.1%) than real-time PCR, in both NP and OP swabs (both 0/256, 0.0%) P = .001. Similar results were recorded for detection of influenza viruses by serology (34/236, 14.4%) compared to real-time PCR in NP swabs (15/236, 6.4%) P < .001; and serology (34/235, 14.5%) compared to real-time PCR in OP swabs (15/235, 6.4%) P < .001. Thirdly, there was no significant difference between sputum culture and NP swab culture for detection of bacteria (Figure 2A). Of note, sputum samples were collected from 165 (62%) patients; of these were only 73 (44%) deemed valid for culture by microscopy.

Microbial yield in patients with complete sample collection
Microbial etiology was established in 47 (73%) of 64 patients with complete sample collection ( Figure 3). The total diagnostic yield improved by 8% (47 vs. 42 of 64 cases, P = .44) when PCR was added to conventional methods; 5% (37 vs. 34) in cases with bacterial etiology; and 19% (26 vs. 14, P = .04) in cases with viral etiology. Cases with detection of ≥1 copathogen improved by 16% (22 vs. 12). Of 20 (31%) patients in whom S. pneumoniae was identified, 12 (60%) had ≥1 copathogen detected. This copathogen was due to ≥1 viral agent in 9 (45%) of patients. Etiology was established in 34 (79%) of 43 patients with complete sampling and who had not been prescribed antibiotics prior to hospital admission (not shown).

Seasonality
Adult CAP patients with viral findings showed marked seasonal variations during the study period; the proportion of cases with positive findings was significantly higher in the winter and spring compared to summer and fall (45% vs. 28%, P = .01) ( Figure 4A). Annually outbreaks of influenza viruses and S. pneumoniae contributed to the increase of these findings during the cold half of the year (Additional file 1: Figure S1). Only 2 cases of influenza A (H1N1) virus infection were detected, although the H1N1 pandemic reached our region in the middle of the inclusion period, May 2009. Detection rates of viral-bacterial coinfections in patients with CAP also varied considerably   from month to month (range 0%-60%), with peaks in January and February (not shown); the proportion of viral-bacterial coinfections was significantly higher in the winter and spring compared to summer and fall (39% vs. 10%, P = .01) ( Figure 4B). During winter and spring, viralbacterial coinfections were more frequently detected than either pure infection alone.

Discussion
This study is the first to describe etiology of CAP in hospitalized adults in Norway using conventional methods and PCR. The main results were: Firstly, etiology was established in the majority of patients, especially when samples were complete and no antibiotics were prescribed prior to hospital admission. Furthermore, PCR improved diagnostic yield, particularly in diagnosing viral infections. Secondly, S. pneumoniae and respiratory viruses (mainly influenza viruses and rhinovirus) were most commonly detected, usually with copathogens. Viruses were detected more frequently during winter and spring, which was also the time period they occurred most commonly in combination with bacteria. Thirdly, OP swabs gave more positive results than NP swabs for qPCR detection of S. pneumoniae; and serology gave more positive results than real-time PCR for detection of B. pertussis and influenza viruses. Establishing a microbiological diagnosis for patients with CAP is challenging. As is the case in any study of samples not obtained through invasive techniques and with quantitation, the link between presence and causal is still quite unclear. However, in this study we applied a vast array of available tests to determine the presence of known pathogens, including those only identified by real-time PCR, following recommendations for interpretation applicable to our region. The diagnostic yield of 63% is consistent with results of other studies, although reports vary considerably from 39% to 76% [11][12][13][14][15][16][27][28][29][30][31]. These variations may be attributable to differences in the epidemiology of pathogens, study population, diagnostic methods and available patient specimens. The high diagnostic yield that was achieved in the complete sampling group underscores that etiology, nowadays, can be established in the majority of CAP patients if multiple techniques are applied and sample collection is optimized.
The prevalence of S. pneumoniae, other "typical" bacteria and Legionella pneumophila, which almost exclusively occurred in "imported" cases, are in line with previous studies from Northern-Europe [7,11] and USA [32]. C. pneumoniae was expectedly uncommon [7,11]. The relatively high frequency of B. pertussis is probably explained  by an ongoing epidemic [33], and the low frequency of M. pneumoniae due to timing; the study was conducted between two 5-7 years epidemic cycles [34].
The potential role of viruses in causing pneumonia, both as the sole pathogen as well as in association with bacteria, is debated. Furthermore, the validity of upperairway samples to establish an etiologic diagnosis is controversial. However, the impacts of viruses per se on adult CAP are increasingly being recognized [35][36][37][38]. In line with other studies [11,28,29], we identified viruses by all methods in 34% of all cases and in 41% of cases with complete sampling. In contrast, conventional methods identified only 14% of all cases, which confirms the increased sensitivity of PCR reported by others [11,16,39]. As expected, the occurrence of adult CAP with viral findings varied considerably during the year, peaking in March at~70%, and was significantly higher during winter and spring compared to summer and fall. Influenza viruses were most frequently detected; 14% by serology compared to 6% by PCR, the difference between the two methods being statistically significant-and no more than 2 cases of influenza A (H1N1) virus infection were detected by PCR-illustrating the substantial value of serology in epidemiological studies. However, serology is generally not recommended for clinical decision making due to low specificity, which in our study may have led to an overestimation of numbers since a probable diagnosis was not only based on seroconversion but for some patients on high titer in single sample. Interestingly there was no increase in CAP admissions due to influenza viruses during the 2009 H1N1 pandemic, indicating that influenza viruses are major pathogens regularly occurring in patients hospitalized with CAP irrespective of the H1N1 pandemic. The low diagnostic yield of influenza viruses by PCR technique also underscores the clinical problem with false negative PCR influenza tests, and may suggest that sampling from both the upper and lower respiratory tract should be performed, at least in severely affected individuals [40]. Another explanation may be that the PCR influenza test is more likely to be negative later in the course of a secondary pneumonia to influenza virus infection leading to hospitalization.
Coinfections have previously been reported to account for 4% to 30% of adult CAP [28,30]. In our study, more than one pathogen was found in 26% of all patients and in 34% of patients with complete sampling. In agreement with recent studies, S. pneumoniae was the most common bacterial agent associated with coinfections, and a copathogen was found in more than one-half of the cases with the majority of pathogens being viral agents [11,12,14,41], usually influenza viruses or rhinovirus. One of our important observations, applying comprehensive and complete diagnostics, was an exceptionally high detection rate of viralbacterial coinfections in the course of the winter season, peaking at 60% in January and February. This number is considerably higher than the 39% recently reported [40], indicating that seasonal activity patterns of respiratory pathogens also impacts on the occurrence of adult CAP with viral-bacterial coinfections, which appear to be the dominant cause of pneumonia during those times.
Good-quality sputum culture is often difficult to obtain and reported yields are varying [1]. In our study, sputum samples were obtained from 62% of patients; of those obtained were 56% judged as inadequate for culture, and of those cultured revealed 21% a positive result. NP swab cultures were obtained from 99% of patients; only 13% of these revealed a positive result. Our concordance analyses indicate, however, that the two methods are complementary. PCR improved the yield by 8-15% when combined with conventional methods, proving that PCR is a valuable tool in addition to conventional methods. However, since serological tests have little clinical impact in the acute situation, our findings underscore the need for further improvements. Our results, using flocked swabs, indicate quite equal effectiveness of NP and OP sampling for real-time PCR detection of viruses and the concordance rate between the two specimens was good (~90%). These results are in contrast to those from two previous reports [29,42] that observed superiority of NP sampling, one of the studies using flocked swabs [42]. To our knowledge, no study comparing NP and OP sampling for qPCR detection of S. pneumoniae in adult CAP has yet been published: In line with the findings of Principi et al. [26], we found that OP swabs were significantly more effective than NP swabs for qPCR detection of S. pneumoniae. However, it is not known whether these might actually represent false-positives as, according to our analyses, OP swabs showed poorer specificity (90.2%) compared to NP swabs (96.8%) (Additional file 1). Furthermore, NP swabs were not assessed by direct fluorescent antigen for specimen adequacy prior to testing. However, excellent specimen adequacy of NP flocked swabs has recently been reported [43]. Nevertheless, due to the lack of diagnostic gold standards, accuracy of such tests is difficult to assess and more research is needed to confirm this finding.
Certain limitations concern data collection and analyses. Firstly, the requirement of rectal temperature >38.0°C may exclude predominantly elderly patients with CAP, who remain afebrile, and hence contribute to explain the low number of nursing home residents in the study. Secondly, some patients who met the inclusion criteria were not enrolled in the study and the complete microbial sampling could not be applied in every patient. We have, however, no reason to suspect these factors to induce systematic selection bias. Thirdly, not all sputum and NP culture samples were taken before administration of antibiotics, which influences the validity of the microbial etiology patterns found and the comparison of diagnostic tests.
Fourthly, specificity issues have been raised regarding the ply gene used, which has been identified in Streptococcus mitis. However, we used quantitative technique, and any DNA amount of S. mitis would probably have been low, at least in NP samples. Furthermore, we did not include a control group to determine the prevalence of adult pneumococcal nasopharyngeal carriage. The frequency of carriage in Norway hardly differs appreciably from the 4% reported from Sweden [44]. Also, we used a C q cut-off value of 10 5 cfu/mL to distinguish colonization from true infection [25]. Because we did not include control subjects we were unable to investigate the significance of the proportion of certain viruses, in particular rhinovirus, among patients with CAP. Finally, we included 48 (18%) patients with immunosuppression (i.e., primary or acquired immunodeficiency, active malignancy, immunosuppressive drugs) that may have an impact on the etiological results. However, this proportion of immunocompromized patients in our study does not exceed 25% that probably would have hampered comparison of etiological results with other studies from which such patients have been excluded [7]. Moreover, inclusion of these patients may also be a strength of the study as they better reflect the population being referred to this local hospital. These limitations have been considered in the analyses and presentation of results.