Selection process for randomized controlled trials (RCTs). CAP indicates community-acquired pneumonia.
The number of patients who died within the follow-up period (overall mortality). The left side depicts arms with atypical coverage; the right side, arms without atypical coverage. A value of less than 1 favors atypical coverage. Studies are subdivided by the antibiotic used as the atypical regimen, including quinolone therapy (01), macrolide therapy (02), and combined macrolide and quinolone therapy (03). The total indicates the total number of patients (sum of groups 01-03). CAP indicates community-acquired pneumonia; CI, confidence interval; I2, measure of inconsistency (see “Methods” section); n/N, number of patients/total number of patients in the study; RR, relative risk; and solid oblong diamond, total events.
The number of patients considered to have a clinical treatment failure. The left side depicts arms with atypical coverage; the right side, arms without atypical coverage. Studies are subdivided by the antibiotic used as the atypical regimen, including quinolone therapy (01), macrolide therapy (02), and combined macrolide and quinolone therapy (03). Total indicates the total number of patients (sum of groups 01-03). CAP indicate community-acquired pneumonia; CI, confidence interval; I2, measure of inconsistency (see “Methods” section); n/N, number of patients/total number of patients in the study; RR, relative risk; and solid oblong diamond, total events.
Shefet D, Robenshtok E, Paul M, Leibovici L. Empirical Atypical Coverage for Inpatients With Community-Acquired PneumoniaSystematic Review of Randomized Controlled Trials. Arch Intern Med. 2005;165(17):1992-2000. doi:10.1001/archinte.165.17.1992
Copyright 2005 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2005
Current guidelines of empirical antibiotic treatment for inpatients with community-acquired pneumonia recommend antibiotics whose spectrum covers intracellular (atypical) pathogens. No sufficient evidence exists to support the necessity of such coverage, whereas limiting it may reduce toxic effects, resistance, and expense. Our goal was to assess the efficacy of empirical coverage of atypical pathogens in terms of mortality and clinical and bacteriological success.
Systematic review and meta-analysis of randomized, controlled trials comparing treatment regimens with and without coverage of atypical pathogens. We searched MEDLINE, EMBASE, the Cochrane Library, and references. Relative risks (RRs) with 95% confidence intervals (CIs) were pooled using the fixed-effects model. The primary outcome assessed was all-cause mortality.
We included 24 trials encompassing 5015 patients. We found no studies of a drug without atypical coverage that compared it with the same drug supplemented with a drug with atypical coverage; nearly all compared a β-lactam with a single quinolone or macrolide. There was no difference in mortality between the 2 arms (RR, 1.13 [95% CI, 0.82-1.54]). Regimens with coverage of atypical pathogens showed a trend toward clinical success and a significant advantage to bacteriological eradication. Both disappeared when evaluating methodologically high-quality studies alone. These regimens further showed a significant advantage in clinical success for Legionella pneumophila, whereas no advantage for pneumococcal pneumonia was seen. There was no difference between study arms in the frequency of total adverse events.
Empirical antibiotic coverage of atypical pathogens in hospitalized patients with community-acquired pneumonia showed no benefit of survival or clinical efficacy in this synthesis of randomized trials.
Major guidelines for the treatment of community-acquired pneumonia (CAP) generally differentiate between outpatients, inpatients, and patients hospitalized in intensive care units.1- 4 Suggested antibiotic regimens for inpatients include a β-lactam combined with macrolides, or monotherapy with a respiratory fluoroquinolone. Although Streptococcus pneumoniae remains the leading pathogen in CAP, the rationale for a macrolide supplement or fluoroquinolone monotherapy lies in its ability to cover intracellular (atypical) pathogens such as Chlamydia pneumoniae, Mycoplasma pneumoniae, and Legionella pneumophila.
Coverage of the latter pathogen is recommended explicitly for patients in intensive care units. To our knowledge, no randomized, controlled trial has compared, as an objective, the superiority of antibiotic regimens containing coverage of atypical pathogens with regimens lacking such coverage. A systematic review of nonrandomized studies found a significant reduction in mortality when the antibiotic spectrum covered atypical pathogens in 6 of its 8 selected studies.5 However, all studies were cohort studies, and 2 were restricted to bacteremic pneumococcal pneumonia.6,7 In the largest study,8 the choice of such regimens was associated with an initial lower severity score, thus underlining the potential bias of nonrandomized studies. Whereas the advantage of combination therapy is unproved, dual therapy may increase toxic effects, resistance, and cost. Moreover, an antagonism between penicillin and erythromycin has been shown in vitro and in vivo against S pneumoniae isolates,9 the most prevalent pathogen causing CAP.10,11
The present review evaluates the need for empirical antibiotic coverage of atypical pathogens in adults hospitalized owing to CAP. It includes all randomized, controlled trials that compared an antibiotic regimen containing coverage of atypical pathogens with one not containing such coverage. The main outcome was mortality. Secondary outcomes included clinical efficacy, bacteriological failure, and adverse events.
We included randomized, controlled trials that assessed treatment of CAP in hospitalized adults and in which an antibiotic regimen containing coverage of atypical pathogens was compared with a regimen not containing such coverage. Regimens including a macrolide, fluoroquinolone, tetracycline, doxycycline, or chloramphenicol were considered to afford atypical coverage. Regimens lacking these drugs were considered regimens without atypical coverage. We included oral and intravenous therapies.
Trials that included mainly patients with major immunosuppressive states were not considered for this review. Trials with a dropout rate of more than 30% were excluded.
The search string combined community-acquired infections/pneumonia, inpatients, and antibiotic names and classes of atypical drugs identified in the previous section (string specified at The Cochrane Database of Systematic Reviews 2005, Issue 2).
Databases searched included CENTRAL (Cochrane Library Issue 4, 2004), MEDLINE (to August 2004), and EMBASE (to July 2003). We inspected references of identified studies for more trials and contacted corresponding authors for complementary information.
The primary outcome was overall mortality up to 30 days after the end of treatment. Secondary outcomes included clinical treatment failure, bacteriological eradication, and development of superinfections and adverse events, specifically gastrointestinal events or events resulting in treatment discontinuation.
Outcomes were extracted by intention to treat (ITT), including all individuals randomized in the outcome assessment. When data for ITT analysis were unavailable, available cases were assessed. Two reviewers (D.S. and E.R.) independently extracted data from included trials. Methodological assessment was performed using a component approach, including allocation generation and concealment, blinding, and analysis by ITT. Allocation generation and concealment were classified as adequate, unclear, or inadequate, using criteria from the Cochrane Reviewers’ Handbook 188.8.131.52 We did not assess a composite quality scale, because different scales may lead to discordant results.13 Sensitivity analyses were performed to assess the robustness of the findings per the following trial methods: allocation concealment, allocation generation, and blinding.
Relative risks (RRs) with 95% confidence intervals (CIs) are reported. We used a fixed-effects model and compared it with a random-effects model when significant heterogeneity between trials was observed. Heterogeneity of trial results was assessed by calculating a χ2 test of heterogeneity and the I2 measure of inconsistency. Significant heterogeneity was predefined as a χ2 test P value smaller than .1 or an I2 measure larger than 50%. We had anticipated between-trial variation in the estimation of morbidity and mortality for different geographic areas, age groups, sample size, and the drug affording atypical coverage (macrolides or fluoroquinolones). Subgroup analyses to assess the impact of these factors on the main results were performed. A funnel plot estimating the trial precision (logarithm of the RR for efficacy against sample size) was examined to estimate potential asymmetry.
Our search resulted in 994 references. Fifty-six publications were retrieved for full-text inspection,14- 69 of which 26 fulfilled the inclusion criteria. Two were withdrawn from analysis owing to unavailable data,44,45 and thus 24 trials are included in the review46- 69 (Figure 1 and Table 1).
Included studies were performed between 1982 and 2004 and encompassed 5015 patients. Inclusion criteria in all studies consisted of adults hospitalized with CAP. The number of participants was 100 or fewer in 8 trials and more than 100 in 16 trials (range, 40-808 participants). All trials were restricted to adults, with a mean age less than 65 years in 12 studies and 65 years or greater in 9. Among the latter, 2 studies were performed in nursing homes,54,62 and 1 exclusively included patients older than 70 years.65 Three studies did not report mean age. Ten studies provided the percentage of patients with chronic obstructive pulmonary disease, ranging from 25% to 52.5%.64 None analyzed separately results for patients with chronic obstructive pulmonary disease and/or smokers.
Pneumonia was defined by a combination of clinical signs, radiological confirmation (the sole criteria in 2 studies), laboratory values, and/or bacteriological evidence. Thirteen trials further included outpatients, patients with nosocomial pneumonia, and/or patients with bronchitis. In all cases, most of the patients had CAP or could undergo separate analysis.
The antibiotic regimens, dosages, and routes of administration are detailed in Table 1. In nearly all studies, the comparison was between monotherapy in the arm covering atypical pathogens and a β-lactam. We found no comparison of a β-lactam–macrolide combination with β-lactam monotherapy. Treatment duration was conveyed in 14 studies and was almost uniformly 10 days, with no difference between the arms. The main outcome in all studies was clinical treatment failure. Six studies mandated radiological resolution for success definition, and 1 required bacteriological eradication. None chose mortality as the primary outcome.
Eighteen trials assessed bacteriological failure (per patient or per pathogen). Only 8 performed serologic tests for atypical pathogens, of which 1 study found negative results for all tests,68 and 4 others did not fully report eradication rates. Superinfection and colonization rates were reported in only 5 studies each, precluding further evaluation.
Adverse events were addressed in all studies, although 2 did not specify the number of events per treatment arm.
Of the 24 included studies, adequate allocation concealment was reported in 6 and adequate allocation generation in 9. No information was available for the remaining studies. All studies of adequate allocation concealment were also of adequate allocation generation.
Seven studies reported results by ITT. Another 13 reported the number of dropouts per study arm, permitting reanalysis by ITT by assuming failure for all dropouts. Four studies did not refer to dropouts and were analyzed only by patients undergoing evaluation.
Follow-up duration was specified in 21 studies, of which 16 defined a specific time for outcome measurement. Follow-up ranged from the end of treatment to 3 months after. Overall mortality was assessed at the end of treatment or at follow-up in all studies. Data at the furthest point in time, up to 30 days, was chosen for analysis. At least 18 of the 24 studies were sponsored by pharmaceutical companies, all of which manufactured the drug with atypical coverage.
Twenty-three of the 24 studies could be evaluated for mortality, encompassing 4846 of 5015 randomized patients (96.6%) (Figure 2). Six studies reported no deaths, whereas 10 reported mortality rates of 0.4% to 5%; 6, 5% to 8%, and 1, 25%.51 There was no significant difference between the arms in the overall mortality rate (RR, 1.13 [95% CI, 0.82-1.54]) (Figure 2). The difference was nonsignificant when evaluating quinolones (RR, 0.98; 95% CI, 0.69-1.41) and macrolides (RR, 1.25 [95% CI, 0.52-3.01]). No heterogeneity was seen for the overall comparison.
Mortality was further analyzed by age, geographic area, and sample size, and the results disclosed no significant difference. Overall mortality in both arms was similar when analyzing studies per allocation generation, allocation concealment, blinding, and the ITT analysis (Table 2). In the funnel plot for overall mortality, results are symmetrically centered around the combined RR.
Clinical failure was the primary outcome in all studies, encompassing 4682 patients. No significant difference between study arms was observed (RR, 0.92 [95% CI, 0.82-1.03]) (Figure 3).
When we evaluated the different drug regimens, opposing trends were noticeable, with an advantage for quinolone monotherapy (RR, 0.89 [95% CI, 0.77-1.02]) and a disadvantage for macrolide monotherapy (RR, 1.17 [95% CI, 0.77-1.77]). Clinical failure with macrolide treatment was the only comparison in which heterogeneity was detected (χ23 = 6.68; P = .08; I2 = 55.1%). Reanalysis by the random-effects model did not alter the results. Relative risks were similar regardless of age or sample size. An advantage for coverage of atypical pathogens was statistically significant in the 13 European studies (RR, 0.82 [95% CI, 0.70-0.95]), but not in studies performed elsewhere.
When we analyzed studies by methodological quality, an advantage toward coverage of atypical pathogens was accentuated in studies of unclear or inadequate allocation concealment and allocation generation. In the analysis of studies of high methodological quality, the effect was nearly identical in the 2 arms (for adequate allocation generation, RR, 0.99 [95% CI, 0.82-1.19]; for adequate allocation concealment, RR, 0.98 [95% CI, 0.81-1.19]) (Table 2). In an ITT vs per-protocol design sensitivity analysis, no significant difference was found.
Clinical treatment failure rates were evaluated among patients with microbiologically documented infections. No significant difference between the study arms in the treatment of documented pneumococcal infections was detected (RR, 1.15 [95% CI, 0.81-1.63] among 16 studies and 906 patients). Data were insufficient to analyze cases of pneumococcal bacteremia. For atypical pathogens, a trend in favor of atypical coverage did not reach statistical significance (RR, 0.52 [95% CI, 0.24-1.10] among 4 studies and 158 patients). A significant advantage to coverage of atypical pathogens was found for eradication of Legionella species, with an RR of 0.17 and narrow 95% CIs (0.05-0.63), based on relatively few cases (n = 43). Sixty-one of 78 atypical cases and 9 of 20 cases of L pneumophila were successfully resolved in the arm without coverage of atypical pathogens.
Eighteen studies reported bacteriological eradication rates, encompassing 1968 patients and/or isolates. There was a statistically significant advantage to bacteriological eradication for the arm covering atypical pathogens (RR, 0.73 [95% CI, 0.59-0.91]), with no heterogeneity seen. However, in an analysis restricted to studies of adequate allocation generation and concealment, this advantage disappeared (RR, 0.96 [95% CI, 0.61-1.52]) (Table 2).
Adverse events per treatment arm were reported for 4261 patients. Total adverse events (RR, 1.02 [95% CI, 0.91-1.13]) and events requiring treatment discontinuation (RR, 0.98 [95% CI, 0.67-1.42]) were similar in both treatment arms, with no heterogeneity seen. Gastrointestinal events were reported in 15 studies and were significantly more common in the arm without atypical coverage (which consisted mainly of β-lactams) (RR, 0.73 [95% CI, 0.54-0.99]). However, the definitions of gastrointestinal events differed, some including abdominal pain and some diarrhea alone, thereby precluding an accurate comparison of antibiotic-associated diarrhea.
The objective of our review was to assess empirical antibiotic coverage of atypical pathogens in hospitalized patients with CAP, in terms of mortality and successful treatment. We found no difference in mortality between regimens with coverage of atypical pathogens and regimens without such coverage, persisting in all subgroup analyses. There was a nonsignificant trend toward clinical success to coverage of atypical pathogens, accentuated with quinolone monotherapy. The advantage disappeared when we evaluated high-quality methodological studies alone. A significant advantage in bacteriological eradication was detected in the coverage of atypical pathogens, especially in reference to Legionella species. This advantage was not demonstrated in an analysis restricted to studies of adequate allocation generation and concealment. There was no difference in the frequency of total adverse events between the 2 groups, although more gastrointestinal events (but not explicitly diarrhea) were noted in the arm without atypical coverage.
Mortality data were obtained for 96.6% of randomized patients. The overall mortality rate (adjusted mean mortality rate, 3.7%) was lower than that reported in the literature (eg, MedisGroups, 10.6%70; validation cohort inpatient mortality for the Pneumonia Patient Outcome Research Team, 8.0%71). This is surprising because nearly half of the studies target relatively severe pneumonia cases. Thus, patients recruited to randomized trials may not adequately represent all patients hospitalized with CAP.
Although mortality is the most significant outcome in a potentially lethal infection, all studies chose clinical failure as their primary outcome. This end point is subjective and should be studied with care. Our review clearly demonstrates its potential for bias. A trend in favor of clinical success for the arm covering atypical pathogens originated in studies with unclear allocation generation. Similarly, the clear statistical advantage of that arm, found in the overall analysis of bacteriological eradication rates, did not exist in an analysis restricted to studies of adequate allocation generation. Thus, we should be wary about relying solely on subjective outcomes when comparing treatment regimens for pneumonia, especially because pharmaceutical companies sponsored most studies and many studies were nonblinded.
The similar response of the young and old is somewhat surprising, as an advantage to atypical coverage would be expected in younger people with a higher prevalence of atypical pneumonia. Perhaps this prevalence diminishes in the hospitalized population. The clear advantage of the arm with atypical pathogen coverage in the successful treatment of L pneumophila infections is not surprising, although cases of atypical pneumonia (including L pneumophila) often resolved without such coverage. Coinfections with typical pathogens may explain some of these cases.
We had set out to investigate the contribution of coverage of atypical pathogens to empirical treatment of CAP in hospitalized patients. The most suitable study for our purpose would have been one comparing a drug without atypical coverage (eg, β-lactam) with a combination of that drug and a drug with atypical coverage (eg, β-lactam and a macrolide). None was found, although the need to add a macrolide to β-lactam therapy is a common dilemma manifested within the guidelines themselves. Furthermore, many studies included treatment arms that do not adhere to current guidelines. Therefore, our meta-analysis is chiefly based on comparison of various regimens without coverage of atypical pathogens to monotherapy, mainly quinolone monotherapy. Regarding this comparison, we found no advantage to coverage of atypical pathogens in terms of mortality or clinical success.
Our conclusion of no benefit might be due to lack of power when using available randomized trials. Large observational studies showed benefit for atypical coverage. However, correction for the baseline differences between patients given or not given atypical coverage in these studies may be impossible.
Studies designed specifically to evaluate the necessity of atypical coverage are needed. The optimal design would be a randomized controlled trial comparing the same β-lactam in both study arms with and without the addition of antibiotics against atypical pathogens. Studies must be of adequate generation concealment and allocation, and patients included should resemble more closely the general population of inpatients with CAP.
Correspondence: Daphna Shefet, MD, Department of Medicine E, Beilinson Campus, Rabin Medical Center, Petah-Tiqva, Israel (email@example.com).
Accepted for Publication: May 2, 2005.
Financial Disclosure: None.
Funding/Support: This study was supported by a European Union Fifth Framework Information Society Technologies program (TREAT), Brussels, Belgium.
Additional Information: A detailed protocol of the methodology used for this study was published in The Cochrane Library, where the full review has been accepted for publication.
Acknowledgment: We thank Claude Carbon, MD, Christian Chuard, MD, Francois Fourrier, MD, Daniel Genne, MD, Hartmut Lode, MD, Phillip Peterson, MD, Patrick Petitpretz, MD, and Jose Sifuentes Osornio, MD, for supplying complementary information for their trials; Karla Soares-Weiser, MD, PhD, for her guidance; Gabriel Izbicki, MD, for translation from German; and Rika Fujia for obtaining and translating the Japanese articles.