Description of the condition
Community acquired pneumonia (CAP), which includes pneumonia acquired in the community at large, but excludes cases in nursing and long-term care facilities, is a common condition that carries a high burden of mortality and morbidity, particularly in the elderly. Prospective studies carried out in the UK, Finland and the USA have estimated the annual incidence of CAP in community-dwelling adults at 5 to 11 cases per 1000 adult population; the incidence is known to vary markedly with age, being higher in the very young and the elderly (Foy 1979; Jokinen 1993; Woodhead 1987).
It is the most important cause of death from infectious causes in high-income countries and the sixth most important cause of death overall (Mandell 2007). CAP can be caused by a broad range of pathogens including bacteria, atypical agents (Chlamydophila pneumoniae (C. pneumoniae), Mycoplasma pneumoniae (M. pneumoniae), Legionella pneumophilus (L. pneumophilus)) and viruses (Mandell 2007). In fact, more than 100 different microorganisms have been associated with CAP (Loeb 2002). Furthermore, a patient with CAP can be infected with more than one microbe, as in the case of a bacterial superinfection of an underlying influenza infection. The most common pathogens include, depending on the patient population tested, Streptococcus pneumoniae (S. pneumoniae) (usually by far the most common), C. pneumoniae, Haemophilus influenzae (H. influenzae), M. pneumoniae and influenza viruses (Loeb 2002; Mandell 2007).
Significant costs are associated with the diagnosis and management of CAP. In the UK, 22% to 42% of adults with CAP are admitted to hospital (Guest 1997; Woodhead 1987), and of those, 5% to 10% need to be admitted to an intensive care unit (BTS 1992; Torres 1991).
Description of the intervention
Antibiotics are the mainstay in the treatment of CAP, since the causative organisms usually respond to them. Consequently, CAP contributes significantly to antibiotic use, which is associated with the development of bacterial resistance. In treating patients with CAP, the choice of antibiotic is a difficult one. Factors to be considered are the possible etiologic pathogen, the efficacy of the substance, potential side-effects, the treatment schedule and its effect on adherence to treatment, the particular regional resistance profile of the causative organism, and co-morbidities that might influence the range of potential pathogens (such as in cystic fibrosis) or the dosage (as in the case of renal insufficiency).
Why it is important to do this review
Many clinical trials have been performed to evaluate and compare the efficacy of antibiotics for CAP. However, the vast majority of them were conducted in hospitalized patients. These patients usually suffer from more severe manifestations of the disease and often have other co-morbid conditions that affect their response to treatment and their time to recovery. Consequently, it is unclear as to what extent results comparing the efficacy of different antibiotics in hospitalized patients can be extrapolated to outpatients. Numerous guidelines exist to aid clinicians with the treatment of CAP: in recent years, guidelines have been published, among others, by the American Thoracic Society (ATS 2001), the Infectious Diseases Society of America (IDSA 2000, updated December 2003; IDSA 2003 and March 2007; Mandell 2007), the British Thoracic Society (BTS 2001, update: BTS 2004), the Canadian Community-Acquired Pneumonia Working Group (CCAPWG 2000), the European Respiratory Society (Woodhead 2005), a German Guidelines Group (Hoeffken 2005), the Gulf Cooperation Council (Memish 2007), the Japanese Respiratory Society (JRS 2005), the Latin American Thoracic Association (ALAT 2001, update: ALAT 2004), the South African Thoracic Society (SATS 2007), and the Swedish Society of Infectious Diseases (Hedlund 2005; update: Strålin 2007). All these guidelines include recommendations for the choice of antibiotic treatment for CAP in ambulatory patients. However, the evidence on which these recommendations are based are derived mainly from studies carried out almost exclusively in hospitalized patients. Although many studies have been published concerning CAP and its treatment, there is no concise summary of the available evidence concerning its treatment in unselected ambulatory outpatients.
This review is an update of our first review (Bjerre 2004) and, like it, addresses the comparative efficacy of antibiotic treatments for community acquired pneumonia (CAP) in outpatients above 12 years of age.
- To assess and compare the efficacy of individual antimicrobial therapies with respect to clinical, radiological and bacteriological outcomes in unselected adult outpatients with CAP.
- To assess and compare the efficacy of antibiotic drugs across drug groups.
Criteria for considering studies for this review
Types of studies
RCTs of antibiotics in adolescent and adult outpatients with CAP reporting on clinical parameters, cure rates and/or mortality were considered for inclusion.
Types of participants
Outpatients of either gender over 12 years of age with the following:
- Symptoms and signs consistent with an acute lower respiratory tract infection associated with new radiographic shadowing for which there is no other explanation (for example, not pulmonary edema or infarction); and
- The illness is the primary clinical problem and is managed as pneumonia.
(modified from the criteria for CAP as defined by the British Thoracic Society (BTS 2001)).
Types of interventions
All double-blind randomized controlled comparisons of one antibiotic and a placebo or at least two antibiotics used to treat CAP were considered. Trials comparing two doses, two treatment durations or two different application forms (intravenous versus oral for example) of the same drug were not included. However, trials comparing two different pharmacologic formulations of a same substance (for example, microspheres versus pure substance) were included, as they are likely to differ in their pharmacodynamic and pharmacokinetic properties, and thus may differ in their efficacy.
Comparisons involving intravenous drugs are usually carried out in a hospital setting. However, as this might occasionally be performed in an ambulatory setting, we did not a priori exclude studies dealing with intravenous drug applications.
Trials allowing concurrent use of other medications such as antitussives, antipyretics, bronchodilators, or mucolytics were included if they allowed equal access to such medications for participants in both arms of the trial.
Types of outcome measures
- Clinical response: improvement of signs and symptoms, usually at a pre-defined test-of-cure (TOC) visit. Where possible, duration of clinical signs and symptoms were used as outcome measures. We used a clinical definition of cure as the primary outcome since radiographic resolution lags behind clinical improvement (Macfarlane 1984).
- Radiologic response: resolution or improvement of a new finding on chest X-ray after antibiotic therapy.
- Bacteriologic response: negative sputum culture in patients previously found to have had pathogens in their sputum.
Search methods for identification of studies
We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library, 2009, issue 1) which contains the Cochrane Acute Respiratory Infections Group's Specialised Register; MEDLINE (January 1966 to February week 2, 2009), and EMBASE (January 1974 to February 2009).
MEDLINE and CENTRAL were searched using the search strategy shown below. We combined the MEDLINE search string with the Cochrane Highly Sensitive Search Strategy for identifying randomized trials in Medline: sensitivity maximizing version (2008 revision) (Lefebvre 2008). The search string was adapted for EMBASE, as shown in Appendix 1.
1 exp Anti-Bacterial Agents/
4 exp Pneumonia/
5 exp Community-Acquired Infections/
6 and/4-5 (3356)
7 community acquired pneumonia.mp.
9 3 and 8
Searching other resources
Studies were also identified by checking the bibliographies of studies and review articles retrieved, and if necessary by contacting the first or corresponding authors of relevant studies. In our first review of this topic, published in 2004 (Bjerre 2004) we had contacted the following antibiotics manufacturers to identify any additional published or unpublished studies: Abbott, AstraZeneca, Aventis, Boehringer-Ingelheim, Bristol-Myers-Squibb, GlaxoSmithKline, Hoffmann-LaRoche, Lilly, Merck, Merck Sharp & Dohme, Novartis, Pfizer, Pharmacia, Sanofi, and Yamanouchi. This search yielded no new studies.
We decided not to contact pharmaceutical companies for future review updates. This decision was made for two reasons: first, because of the very low yield of this search strategy, compared to the significant amount of time it requires; and second, because this search strategy provides an unfair advantage to unpublished studies carried out by industry, as opposed to government or academia, where an equivalent search strategy is not readily available.
For the same reason, we decided to only include studies that have been fully published in peer-reviewed journals. No language restrictions were applied to the search and selection process.
Data collection and analysis
Selection of studies
Two review authors (LMB, TJMV) used the titles and abstracts of the identified citations to exclude trials that clearly did not meet our inclusion criteria in the first publication of this review (Bjerre 2004). In this update, this was done by LMB. If the review author felt that a study might possibly fulfil the inclusion criteria, the full paper was obtained for further study.
Two review authors (LMB, TJMV in the first review, LMB in this updated review) independently reviewed articles having passed this initial screen to determine whether they met the inclusion criteria of the review.
Studies could be excluded for any one of the following reasons: if they were not truly randomized; if they were conducted exclusively in hospitalized patients; if they only compared two doses or two application forms of the same substance; if the indication for treatment consisted of a mix of diagnoses (most commonly: acute bronchitis, exacerbation of chronic bronchitis, and pneumonia) and the results were not reported separately for each diagnostic group.
Another reason for exclusion was that some studies included a mix of in- and outpatients without reporting the data separately for these two sub-groups. Whenever this was the case, we contacted the trial authors to obtain separate data for outpatients only.
Studies including only bacteriologically evaluable patients were also excluded, because these studies typically included only patients with positive cultures of pathogens susceptible to study antibiotics or excluded patients with serologic confirmation of infection with atypical agents (such as M. pneumoniae or C. pneumoniae). A priori exclusion of patients with resistant strains as well as of patients with non-bacterial or atypical causes of CAP would falsely increase the treatment success rate to levels that would be unrealistic in real practice. We chose to exclude these "narrow-focus" studies because we are interested in the efficacy of treatment in patients as they present to their general practitioner (GP), that is, unselected and unfiltered. We consider this essential to the generalizability of our results.
Studies were also excluded if the diagnosis of pneumonia was not confirmed by chest X-ray. This exclusion criteria was necessary to ensure that only participants with a very high likelihood of having pneumonia be included in the review, since this was the patient population in which the efficacy of various treatment alternatives was to be assessed.
Furthermore, studies were excluded if the total number of patients was less than 30, because below this limit, the estimate of a binomial parameter (in this case, the proportion of patients cured or improved) becomes too unstable (Armitage 1994).
Finally, in this update, we added the following exclusion criterion: we excluded studies of antibiotics that have been withdrawn from the market or are no longer licensed for the treatment of outpatients with CAP, due to severe adverse effects. For example, studies assessing the following fluoroquinolones were excluded: gatifloxacin, grepafloxacin, sparfloxacin, temafloxacin and trovafloxacin. Consequently, a study (Ramirez 1999) that had been included in our first review (Bjerre 2004) was excluded from the present review.
Data extraction and management
The following data were extracted from each study, whenever possible:
- description of participants, in particular: age range and gender of participants, smoking status, co-morbidities;
- description of potential pathogens identified and their antimicrobial resistance profiles;
- description of intervention;
- description of control therapy;
- total number of participants in each arm of the trial;
- study setting;
- mean duration of symptoms in each arm of the trial;
- clinical, radiographic and bacteriologic cure rates in each arm of the trial;
- the number of patients lost to follow-up;
- types of adverse effects experienced and number of patients experiencing adverse effects
- the number of drop-outs due to adverse effects;
- proportion of patients admitted to hospital in each arm of the trial;
- mortality rates in each arm of the trial;
- study sponsor.
There were no unreconcilable disagreements. Review authors were not blinded to the identity and affiliation of the study authors.
Assessment of risk of bias in included studies
The risk of bias tables in RevMan (version 5.0.18) were used to systematically assess the risk of bias in included studies. (See 'Characteristic of included studies' table).
Measures of treatment effect
For dichotomous outcome data, an estimate of the common odds ratios with approximate 95% confidence intervals (CIs) was estimated using the Mantel-Haenzel approach. This was done using RevMan software.
Unit of analysis issues
The unit of analysis was the individual patient. All included studies were RCTs without any design particularities, such as cross-over design or multiple interventions, that would warrant special attention to the units of analysis.
Dealing with missing data
Missing data arising, for example, from failure to report on outcomes such as radiological cure rates in some studies, were to be dealt with by excluding the specific outcome from the Data and analysis section for the study in question. As for data missing from individual studies (more specifically, patients lost to follow-up), whenever possible, data from the clinical per-protocol population were used, because this excluded patients who had not been sufficiently exposed to the study drug to be able to potentially benefit from the drug.
Assessment of heterogeneity
The assessment of heterogeneity was carried out by means of the Chi-square test for heterogeneity, available in RevMan software.
Assessment of reporting biases
If applicable, reporting / publication bias was assessed using funnel plots.
Whenever possible, data was synthesized using a fixed-effect meta-analytic model (Mantel-Haenszel odds ratio, available in RevMan software).
Sensitivity analyses including and then excluding small studies (less than 30 patients) as well as, if relevant, excluding any low quality studies were conducted, to assess the impact of such review decisions on the outcome effect.
Description of studies
Results of the search
This search yielded a total of 1828 references in our first review (1966 to 2003) (Bjerre 2004), and the current updated search (2003 to 2009) yielded an additional 1298 records, for a grand total of 3126 records. Some records were double entries, due to the overlapping content of databases.
Six RCTs involving a total of 1857 patients aged 12 years and older diagnosed with community acquired pneumonia were included in this review: Anderson 1991 and Chien 1993, already included in our 2004 review (Bjerre 2004), and the new trials by D'Ignazio 2005, Drehobl 2005, Kohno 2003, and Mathers Dunbar 2004. The trials included varying numbers of patients, the largest having 499 patients (Drehobl 2005), the smallest 107 (Anderson 1991). The mean size of studies included in the analysis was 310 patients, the median size 315.
All trials enrolled outpatients with CAP. In all trials, the diagnosis of CAP was based on clinical signs and symptoms as well as radiographic findings in all patients. The signs and symptoms used as diagnostic criteria included combinations of the following: fever, chills, recent onset of productive cough, pleuritic chest pain, dyspnoea, pyrexia, tachypnoea, dullness to percussion, egophony, rales, localized reduced breath sounds and bronchial breath sounds. In all six trials, patients were treated exclusively as outpatients.
Patient inclusion and exclusion criteria
Three trials included only adult patients, one trial (Chien 1993) also included adolescents 12 years of age and older, and two included adolescents 16 years of age and older (Drehobl 2005, Kohno 2003). One trial reported including patients as old as 91 and only one used older age as an exclusion criterion (in the study by Kohno 2003, patients were up to 80 years of age). Overall, the trials excluded patients with conditions that could have affected the treatment or interfered with follow-up. Exclusion criteria were reported in sufficient detail in all study reports. The most common criteria reported were: pregnancy and lactation, women not using adequate contraception (usually oral contraceptives or a barrier method), history of allergic reaction to the study drugs, recent treatment with or concomitant use of an antimicrobial agent, concurrent medication with ergotamine, cyclosporin, antacids (except H2-antagonists) or digitalis, conditions affecting GI absorption, severe renal or hepatic impairment, terminal illness or conditions precluding study completion, infectious mononucleosis, HIV/AIDS, and prior participation in the study.
The trials varied with respect to the antibiotics studied (Figure 1). Two trials (Anderson 1991; Chien 1993) studied the same antibiotic pair (clarithromycin and erythromycin). All other trials studied different antibiotic pairs, namely clarithromycin versus azithromycin microspheres (Drehobl 2005), clarithromycin versus telithromycin (Mathers Dunbar 2004), azithromycin microspheres versus levofloxacin (D'Ignazio 2005), and telithromycin versus levofloxacin (Kohno 2003).
|Figure 1. Overview of included studies and antibiotic pairs studied. * indicates studies new to this review; shaded ovals indicate quinolones (gyrase inhibitors), white ovals indicate macrolides|
A large number of studies were excluded because they were conducted exclusively in hospitalized patients. Furthermore, a number of studies reported including a mix of in- and outpatients without reporting data separately for these two subgroups. For these studies, we contacted the trial authors to try to obtain separate data on outpatients. Out of seven trial authors, only two responded, and both were unable to provide us with the necessary data.
Risk of bias in included studies
The extent of reporting was variable between studies, but was generally good to very good. Compliance with treatment was explicitly assessed by pill count in three studies (Anderson 1991; Chien 1993; Mathers Dunbar 2004). None reported any difference in the number of pills remaining between the two groups. However, in the Chien 1993 study, 40 participants were excluded because they received "less than the minimum therapy" (seven days) and these patients were distributed unevenly across the two groups (10 in the clarithromycin group and 30 in the erythromycin group). In the two studies using azithromycin microspheres (D'Ignazio 2005; Drehobl 2005), the compliance in the azithromycin group was 100% in both studies, because the drug was administered in a single dose under directly observed therapy (DOT) at the initial treatment visit.
Regarding co-interventions with other medications, most studies excluded patients whose co-medication included certain drugs such as other antibiotics, chemotherapeutics or anti retrovirals. Only one study (Chien 1993) reported how many patients were excluded because of forbidden co-medication.
|Figure 2. Risk of bias graph: review authors' judgements about each methodological quality item presented as percentages across all included studies.|
|Figure 3. Risk of bias summary: review authors' judgements about each methodological quality item for each included study.|
All trials were randomized, double-blind evaluations comparing two antibiotics. None of the articles reported any test of effectiveness of the blinding procedures used.
Incomplete outcome data
Withdrawals were generally reported in sufficient detail as to the reasons for withdrawal. The number of patients lost to follow-up was reported in all studies. Losses to follow-up appeared to be minor, amounting to a maximum of 10% of the initially randomized patients. One study (Chien 1993) did not present intention-to-treat analysis results.
Since only two studies (Anderson 1991; Chien 1993) addressed the efficacy of the same antibiotic pair, and both studies provided the same information about outcomes, there were no missing data issues in the combined analysis of the data arising from these two studies.
There were no concerns about the selective availability of data, or selective reporting of outcomes. However, because all included studies were supported by the pharmaceutical industry, there are concerns that studies with results unfavourable to the drug under investigation may not have been published, thus potentially leading to publication bias.
Other potential sources of bias
The main concern about other potential sources of bias was that all included studies were sponsored by pharmaceutical companies manufacturing the antibiotics under study.
Effects of interventions
The success rates for each of the treatment arms of the six trials are shown in the Data and analyses section of this review, Analyses 1 to 5 (including sub-analyses). 'Success' was defined as cure or improvement, be it clinical, bacteriological or radiological, as assessed at a predefined follow-up visit ('Test-of-cure' (TOC) visit). Overall, success rates, be they clinical, bacteriological or radiological , were very high, usually ranging from 87% to 96%. Except for bacteriological success in the study by Kohno 2003, none of the clinical, bacterial or radiological success rates differed significantly among treatment arms within each of the studies, nor did they achieve clinical - or statistical - significance when the results of the two studies of clarithromycin versus erythromycin (Anderson 1991; Chien 1993) were pooled together. As for the Kohno 2003 study, the bacteriological success rate (i.e. eradication of previously identified pathogen) significantly favoured Levofloxacin ( Analysis 5.2). Most failures were due to failure to eradicate H. influenzae, so the authors looked at the clinical success rate of patients with H. influenzae at baseline, which turned out to not be significantly different between levofloxacin and telithromycin ( Analysis 5.3).
Comparisons across antibiotic groups
The only comparisons across antibiotic groups are provided by the studies by D'Ignazio 2005 and Kohno 2003 whereby, in each of these studies, a macrolide and a quinolone are compared. Again, there was no significant difference in clinical or bacteriological success, except in the Kohno 2003 study, as detailed above in the 'Efficacy' section; radiological outcomes were not reported separately for the two treatment arms.
In all studies, the most common side effects attributable to the study drugs were gastrointestinal side effects. There were significant differences in the occurrence of side effects attributed to the study drug in four of the six studies (Anderson 1991; Chien 1993; D'Ignazio 2005; Mathers Dunbar 2004), affecting anywhere between 12% and 38% of patients. In the two studies comparing clarithromycin with erythromycin (Anderson 1991; Chien 1993), there were significantly more side effects in the erythromycin group, the majority being gastrointestinal side effects. However, this was not reflected in the rate of side effects leading to withdrawal from the study, which was not significantly different across treatment arms. However, as noted above, in the Chien 1993 study, 40 patients were excluded because they received "less than the minimum therapy" (seven days) and these patients were distributed unevenly across the two groups (10 in the clarithromycin group and 30 in the erythromycin group). Although not listed as drop-outs due to side effects, it is quite plausible that these differences in pre-study drop-out rates were due to the unfavourable gastrointestinal side effects of erythromycin.
Various pathogens were identified with varying frequency across studies. The proportion of samples yielding an identifiable pathogen ranged from 19% (Anderson 1991) to 53% (D'Ignazio 2005). H. influenzae was the most common pathogen identified by Anderson 1991 (62% of positive cultures) and Kohno 2003 (43% of positive cultures), whereas S. pneumoniae was the main causative organism in the Chien 1993 (56% of positive cultures) and Mathers Dunbar 2004 (52% of positive cultures) studies. On the contrary, D'Ignazio 2005 and Drehobl 2005 reported C. pneumoniae as being the most common pathogen (20% and 23% of positive cultures, respectively).
Serologically identified pathogens
The most frequently identified pathogen in the studies by Anderson 1991, Chien 1993 and Kohno 2003 was M. pneumoniae which represented 69% (Anderson 1991), 74% (Chien 1993) and 52% (Kohno 2003) of positive serology results, with C. pneumoniae accounting for the remainder. In the study by D'Ignazio 2005, C. pneumoniae was predominant, representing 61% of atypical pathogens, with M. pneumoniae accounting for the remaining 39%. Likewise in the studies by Drehobl 2005 and Mathers Dunbar 2004, C. pneumoniae represented just over half of atypical pathogens (53% and 52%, respectively). Only one patient tested positive for Legionella pneumoniae (L. pneumoniae) in the Mathers Dunbar 2004 study, and no samples were positive for Chlamydia psittaci (C. psittaci ) in any of studies.
Summary of main results
The overwhelming feature of this review update remains the paucity of relevant evidence that could be identified and included in the review. Nonetheless, in this update, four new studies were included (D'Ignazio 2005; Drehobl 2005; Kohno 2003; Mathers Dunbar 2004). Inclusion of these studies did not alter the conclusions of our previous review.
Unfortunately, only two studies (Anderson 1991; Chien 1993), which had already been included in the previous version of this review, focused on the same antibiotic pair; all other studies dealt with different antibiotic pairs, so that, once again, no formal meta-analysis of the data could be carried out. At most, it can be stated that individual study results did not reveal significant differences in efficacy between various antibiotics and antibiotic groups, but that there were some significant differences with respect to the frequency of side effects. Given this current state of affairs, it is not possible to make strong evidence-based recommendations regarding the choice of antibiotic to be used for the treatment of community acquired pneumonia in ambulatory outpatients. Under such circumstances, other factors such as tolerability, duration and frequency of treatment, and cost will take on more importance in determining the choice of treatment.
Overall completeness and applicability of evidence
One important reason for this lack of evidence is that a large number of the trials originally identified were conducted in hospitalized patients and therefore are not necessarily relevant to the treatment of ambulatory patients. It could be argued that the inclusion/exclusion criteria for this review were too strict and that this is the reason why so few studies were retained. However, we do believe that the criteria we applied are necessary in order to validly address the question of the efficacy of treatment of CAP in ambulatory patients. In particular, it could be argued that the decision to exclude studies based on size is not desirable, since one aim of the review is to pool results and that each study therefore would contribute some information. However, we felt that this criterion was necessary to exclude studies where the number of patients with pneumonia was so small that randomization could no longer be expected to achieve a balanced distribution of confounders, both known and unknown, across study groups.
As for the requirement that the diagnosis of CAP be confirmed by a chest radiograph, we felt that this was necessary to avoid diagnostic misclassification, which could, for example, have led to the inclusion of patients with bronchitis into the review. This could have biased the estimation of the efficacy of various antibiotic treatments, either differentially or non-differentially, depending on the distribution of non-CAP cases across treatment groups. Indeed, most recent clinical guidelines recommend the routine use of chest X-rays to confirm a suspected pneumonia (ATS 2001; CCAPWG 2000; Mandell 2007). However, we are aware that this diagnostic test is often not used in practice and that patients are therefore treated empirically according to the clinical findings and the severity of the clinical picture. In such a situation, GPs should realize that patients with an empirical diagnosis of pneumonia (i.e. diagnosis without chest X-ray) are probably on average, less severely ill than the subjects in the trials we reviewed.
The diversity of pathogens identified as most common causative organisms in our present review underscores the need for conducting studies of CAP treatment in a variety of different geographical locations, and also points to a possible limitation of such studies, namely their questionable generalizability to other clinical and, particularly, geographical situations than the ones under study.
Finally, a lot of potentially useful information is lost because investigators often included a mix of in- and outpatients in their studies without reporting results separately for each of these subgroups. Investigators and journal editors should be strongly encouraged to report such data separately, as these patients have different co-morbidity profiles, and, potentially, respond to treatment differently. Putting heterogenous patient groups into the same analytic basket may reduce the generalizability and thus the usefulness of such study results. Contacting trial authors after publication to get additional data, sometimes years after a study was published, is a time-consuming process with a very low yield, as was our experience, and that of other Cochrane reviewers (for example, Robenshtok 2008).
Quality of the evidence
Overall, the quality of the included studies was relatively good, although there were some differences in the completeness of reporting. The fact that we chose to include only double-blind, controlled, prospective RCTs led to the a priori exclusion of studies of lesser quality.
Potential biases in the review process
By choosing to include only studies published in peer-reviewed journals and by choosing to no longer contact pharmaceutical companies for information on unpublished studies, we believe that we have contributed to increasing study quality and reducing bias in our review to a minimum. Furthermore, by excluding data from studies focusing on selected sub-groups of ambulatory participants (such as participants with suspected bacterial pneumonia), we believe we have maximized the generalizability of our review results to unselected patients presenting to their GP.
It is noteworthy that, once again, all studies meeting inclusion criteria for our review were sponsored by pharmaceutical companies. This could potentially introduce a publication bias, as it would be in the interest of manufacturers not to publish studies yielding unfavourable results about their products. We are of the opinion that there is an urgent need for industry-independent research into the treatment of CAP in ambulatory patients.
Agreements and disagreements with other studies or reviews
As mentioned above, we feel that our decision to exclude studies focusing only on a subset of outpatients (for example, "only bacteriologically evaluable patients" or "excluding patients with atypical pneumonia") is necessary to maintain the generalizability of our results to patients presenting to first-line physicians. In contrast, a recent meta-analysis we encountered in the process of searching the literature (Maimon 2008) was more inclusive. The object of this meta-analysis was different from that of our review. However, the target population (outpatient with CAP, treated as outpatients) was the same. Despite the authors' stated focus on "the inclusion of only randomised prospective double-blind studies using only oral therapy exclusively in outpatients" (p.1974), this work included a number of open-label (non-blinded) studies, studies including a mix of in- and outpatients (for which sub-group data was reportedly obtained from study authors) as well as narrow-focus studies (i.e. studies focusing exclusively on bacterial pneumonia, for example). Of the 13 studies included in this meta-analysis, none met the inclusion criteria for our review, which leads us to question the generalizability of this meta-analysis' results to unselected patients presenting in general practice.
Finally, a recent RCT confirmed that patients with CAP of moderate severity (Fine Score II or III) can be treated as safely and effectively as outpatients than as inpatients (Carratala 2005), at only a fraction of the cost. This further underscores the need for solid, evidence-based data on the treatment of patients with CAP in the ambulatory care setting.
Implications for practice
Currently available evidence from RCTs is insufficient to make evidence-based recommendations for the choice of antibiotic to be used in the treatment of CAP in ambulatory patients. At most, it can be stated that individual study results do not reveal significant differences in efficacy between various antibiotics and antibiotic groups.
Implications for research
Multi-drug, multi-drug-group, double-blind comparisons conducted in various geographical settings are needed to provide the evidence necessary for practice recommendations if these are to be applicable in the ambulatory setting. Study conditions should ensure that diagnosis and management of patients with CAP is as similar as possible to real practice, while still ensuring that the study question is addressed in a valid way. Finally, in studies recruiting a mix of in- and outpatients, it is imperative that data be reported separately for these two sub-groups.
We thank the authors we contacted who kindly replied to our requests for additional information: Dr. Lorenzo Aguilar, Dr. Claude Carbon, Dr. Lars Hagberg, Dr. Karen Higgins, Dr. Shigeru Kohno, Dr. Hartmut Lode, Dr. Lala Mathers Dunbar and Dr. Antoni Torres Martí. Many thanks to Dr. Frederike Behn for precious help with parts of the data extraction process in the first version of this review (Bjerre 2004). We also wish to thank the following people for commenting on this updated draft: Clare Jeffrey, Anne Lyddiatt, Tina Tan, Mark Jones, and Roger Damoiseaux. Last but not least, we thank the Acute Respiratory Infections Group editorial team for enduring support and guidance, in particular Liz Dooley (Managing Editor), as well as Ruth Foxlee and Sarah Thorning (respectively, past and present Trials Search Co-ordinators).
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Differences between protocol and review
- Index terms
Appendix 1. Embase.com search strategy
1. 'antibiotic agent'/exp AND [embase]/lim
#2. antibiotic*:ti,ab AND [embase]/lim
#3. #1 OR #2
#4. 'pneumonia'/exp AND [embase]/lim
#5. 'communicable disease'/exp AND [embase]/lim
#7. 'community acquired pneumonia'/exp AND [embase]/lim
#8. #4 AND #5
#9. 'community acquired pneumonia':ti,ab AND [embase]/lim
#10. #7 OR #8 OR #9
#11. #3 AND #10
#12. 'randomized controlled trial'/exp OR 'double blind procedure'/exp OR 'single blind procedure'/exp OR
'crossover procedure'/exp AND [embase]/lim
#13. random*:ti,ab OR factorial*:ti,ab OR crossover*:ti,ab OR 'cross over':ti,ab OR assign*:ti,ab OR allocat*:ti,ab OR volunteer*:ti,ab OR 'single blind':ti,ab OR 'single blinding':ti,ab OR 'single blinded':ti,ab OR 'double blind':ti,ab OR 'double blinded':ti,ab OR 'double blinding':ti,ab AND [embase]/lim
#14. #12 OR #13
#15. #11 AND #14
Last assessed as up-to-date: 6 July 2009.
Protocol first published: Issue 2, 2000
Review first published: Issue 2, 2004
Contributions of authors
Lise M. Bjerre (LMB) screened abstracts and full articles for inclusion into the review, decided, in agreement with Theo J. M. Verheij (TJMV), which articles to include, extracted the data from these articles, performed the quantitative analyses, wrote the text, tables and figures of the review, and made modifications to the review based on the comments of peer and consumer referees.
Michael M. Kochen (MMK) co-wrote the protocol, defined the search strategy and did a preliminary screening of abstracts for inclusion into the review. He also critically reviewed and edited the text of the review at various stages in its development.
Theo J. M. Verheij (TJMV) co-wrote the protocol, screened abstracts and full articles for inclusion into the study and decided, in agreement with LMB, which articles to include. He also critically reviewed and edited the text of the review at various stages in its development.
Declarations of interest
None of the authors have any potential conflicts of interest to declare.
Sources of support
- None, Not specified.
- None, Not specified.
Differences between protocol and review
In our first review of this topic published in 2004 (Bjerre 2004), as per protocol we contacted the following antibiotics manufacturers to identify any additional published or unpublished studies: Abbott, AstraZeneca, Aventis, Boehringer-Ingelheim, Bristol-Myers-Squibb, GlaxoSmithKline, Hoffmann-LaRoche, Lilly, Merck, Merck Sharp & Dohme, Novartis, Pfizer, Pharmacia, Sanofi, and Yamanouchi. This search yielded no new studies. Starting with this update, we decided to no longer contact pharmaceutical companies to ask about unpublished studies. This decision was made for two reasons: first, because of the very low yield of this search strategy, compared to the significant amount of time it requires; and second, because this search strategy provides an unfair advantage to unpublished studies carried out by industry, as opposed to government or academia, where an equivalent search strategy is not readily available.
In this update, we excluded studies of antibiotics that have been withdrawn from the market or are no longer licensed for the treatment of outpatients with CAP, due to severe adverse effects. For example, studies assessing the following fluoroquinolones were excluded: gatifloxacin, grepafloxacin, sparfloxacin, temafloxacin and trovafloxacin.
Finally, in this update, we applied the new risk of bias tools (tables, summary figures and graphs) newly made available in RevMan Version 5. These tools were not available at the time our protocol and our first review (Bjerre 2004) were written and published.
Medical Subject Headings (MeSH)
MeSH check words
* Indicates the major publication for the study