Community-acquired pneumonia (CAP) is one of our most important diseases, and recent studies have shown that the annual incidence of CAP is 4.7 cases per 1000 adults in Spain and 7–10 cases per 1000 population in the United States, with >500,000 associated hospitalizations and 45,000 associated deaths occurring each year in the United States [1, 2]. More than 1.3 million patients with CAP were hospitalized in 1998, and 3.9 million ambulatory care visits to physician offices, outpatient departments, and emergency departments were made in 1997 in the United States [3, 4]. The estimated cost of an episode of CAP in an inpatient is $6000–$7000, compared with <$200 for a case of CAP in a nonhospitalized patient . Macrolides, which include the closely related azalides, are frequently prescribed for the treatment of outpatients with CAP .
Lonks et al.  described 19 patients who developed breakthrough pneumococcal bacteremia due to macrolide-resistant strains while receiving empirical treatment with orally administered macrolides. These 19 patients included 13 adults and 6 children seen at 3 US centers and 1 Spanish center in 1991–2000. The patients were receiving empirical treatment for CAP or other febrile illnesses. The macrolides administered included azithromycin, clarithromycin, erythromycin, and josamycin. Seventeen of the patients developed breakthrough bacteremia after receiving oral therapy for 2–9 days (mean duration of therapy before bacteremia developed, 3.9 days), and 1 patient developed bacteremia after receiving macrolides intravenously for 3 days (2 doses of cefuroxime were also administered intravenously), followed by azithromycin given orally for 2 days. The remaining patient had been receiving clarithromycin daily for >30 days. Fifteen patients had lobar pulmonary infiltrates noted at the time of breakthrough bacteremia, whereas 4 had no focus of infection. In addition, empyema was present in 2 patients and meningitis was present in another patient. All patients subsequently were successfully treated with intravenously administered β-lactams (18 patients) or vancomycin (1 patient).
These 19 patients join the other anecdotal cases of macrolide treatment failure in patients with macrolide-resistant pneumococcal infections. The authors then went on to identify all cases of macrolide-resistant pneumococcal bacteremia that were seen in their institutions during the same period. They then ascertained whether any of these affected patients had received macrolides before presentation, to determine whether additional cases had occurred. If cases of macrolide-resistant disease occurred in these patients, this would implicate problems with compliance or bioavailability, rather than resistance, as possible reasons for treatment failure. Lonks et al.  identified 67 such cases and did not identify any additional patients who had received macrolides before presentation. Therefore, the likely explanation for treatment failures that occurred among patients who received macrolides before presentation was indeed resistance, not lack of compliance or poor bioavailability. In a comparison of macrolide-resistant cases that occurred among patients who were receiving a macrolide at the time of presentation with bacteremia (n = 19) and those that occurred among patients who did not receive a macrolide at the time of presentation with bacteremia (n = 67), only one characteristic, mean age, was found to be significantly different (mean age, 40 years for patients who had received macrolides vs. 55 years for those who had not [P = .039]). Only 1 (5%) of the 19 patients who had received macrolides presented with meningitis, compared with 9 (13%) of the 67 patients who had not received macrolides (P = .447). There were no deaths in the former group, compared with 12 deaths (18%) in the latter group (P = .06). These outcomes suggest that although there was considerable morbidity among patients with breakthrough bacteremia, the higher mortality rate noted for patients who had not received macrolides before presentation was likely related to the severity of disease at presentation and the older age of these patients.
Lonks et al.  also performed a case-control study of patients with pneumococcal bacteremia to determine whether any “controls” (i.e., patients with bacteremia due to a macrolide-susceptible pneumococcus) had experienced failure of macrolide therapy. This was achieved by matching each “case patient” (i.e., a patient with bacteremia due to a macrolide-nonsusceptible pneumococcus) with up to 2 controls on the basis of hospital location, sex, age, and year of presentation. If any controls had received macrolides, this would suggest that treatment failure was not directly related to macrolide susceptibility. Lonks and colleagues identified 141 case-matched controls, who were compared with 86 case patients. Whereas only 19 of the 86 case patients included in their report had received oral macrolides, as shown in the analysis discussed above, none of the controls had received macrolides before presentation with bacteremia (P < .001). This finding elegantly demonstrated that taking a macrolide was highly likely to be associated with treatment failure as a consequence of macrolide resistance. The only other statistically significant differences between the 2 groups were black race (none of the case patients vs. 11% of the controls; P = .015) and meningitis (12% of the case patients vs. 4% of the controls; P = .0059). The higher incidence of meningitis among case patients could be related to the virulence of the pneumococcal serogroups associated with macrolide resistance (macrolide resistance is predominantly found in serogroups 6, 9, 14, 19, and 23), whereas the pneumococcal serogroups associated with macrolide susceptibility (in the controls) were likely to be more widely distributed.
Evaluation of the significance of these findings needs to take the following factors into consideration: (1) the spectrum of pathogens associated with CAP and outcome on the basis of severity of disease; (2) the natural history of CAP, pneumococcal pneumonia, and bacteremia in various age groups; (3) the in vitro susceptibility of pneumococci to macrolides; and (4) the pharmacokinetic and pharmacodynamic properties of macrolides and the outcome of macrolide-susceptible and -resistant pneumococcal infections in animal models and in humans. Finally, these factors need to be integrated into recommendations for empirical treatment of CAP. These issues will be discussed in turn.
Spectrum of Pathogens Associated with CAP and Outcome According to Severity of Disease
A review of 16 studies of CAP, each of which involved >100 patients, showed that Streptococcus pneumoniae was the most common cause of CAP identified in 14 of these studies . Haemophilus influenzae was the second most common cause in 6 studies, followed by Legionella pneumophila and Mycoplasma pneumoniae. Allowing for the failure of older studies to recognize atypical or intracellular pathogens, a reasonable estimate of the distribution of pathogens in outpatients with CAP is 40%–60% for S. pneumoniae and 10%–20% each for H. influenzae, atypical or intracellular species, and other bacterial pathogens.
Disease severity also has a marked influence on the outcome of CAP. In the Pneumonia Outcome Research Team (PORT) study, 14,199 patients with CAP were stratified into 5 classes on the basis of disease severity . Patients <50 years of age who had no underlying diseases, normal mental status, a pulse rate of <125 beats/min, a respiratory rate of <30 breaths/min, a systolic blood pressure of >90 mm Hg, and a temperature of 35°C–40°C were assigned to class I, whereas those who did not meet these criteria were assigned to classes II–V on the basis of a scoring system. This stratification was correlated with the mortality rate, which was 9.8% overall and increased by class: 0.1% for class I, 0.6% for class II, 2.8% for class III, 8.2% for class IV, and 29.6% for class V. This classification was recently validated by a study of 533 patients with CAP in Spain, where mortality rates, both overall and as stratified by PORT class, were very similar to those found in the PORT study .
Natural History of CAP and of Pneumococcal Pneumonia and Bacteremia
Although these are important issues, there is a lack of adequate information about the untreated natural history of each major etiologic agent in CAP. The PORT study showed that the mortality rate correlates more closely with age, severity of disease, and presence of underlying diseases than with the pathogen involved. A report of the only CAP study to include a no-treatment arm was published in 1938: the mortality rate among untreated, hospitalized patients with acute-onset lobar pneumonia was 27%, compared with 8% among patients treated with sulfonamidopyridine, an early sulfonamide . It is interesting to note that, even in this 1938 publication, the mortality rate was associated with age; in the treatment group, the mortality rate was 2.9% among patients 8–39 years of age and 19.4% among patients ⩾40 years of age, compared with rates of 16.2% and 50%, respectively, among patients in the untreated group. The mortality rate among hospitalized patients with pneumococcal bacteremia is currently still 10%–25% and is thought to result from irreversible disease processes and older age [12–14]. Conversely, death among outpatients with CAP essentially does not occur, because patients who do not respond to therapy are hospitalized . However, the frequency of hospitalization of outpatients who do not respond to therapy is not well documented, although it has been estimated that 20% of patients for whom first-line therapy fails are hospitalized .
Data on variation in the outcome of CAP according to the pathogen involved are also difficult to find, particularly for outpatients, because disease resolves in virtually all outpatients and because placebo-controlled studies are not conducted for ethical reasons. Nevertheless, we can glean some information from CAP outpatient studies in which such agents as β-lactams were used to treat infections, such as M. pneumoniae infection, that are not affected by these agents. Several studies have shown no difference in outcome between treatment with antibiotics that do not provide coverage for these bacteria and those that do [16–18]. Factors that could discriminate among agents, such as speed of resolution, unfortunately are not included in these publications.
In Vitro Susceptibility of Pneumococci to Macrolides
Pneumococci were initially highly susceptible to macrolides, with the MICs of erythromycin, clarithromycin, and azithromycin ⩽0.12 µg/mL . However, resistant strains have emerged during the past 25 years, and 2 patterns of resistance are now common with the worldwide spread of resistant clones . First, some strains have acquired the erm gene, which codes for a ribosomal methylase, and are usually highly resistant to macrolides, lincosamides, and streptogramin B agents (MLSB phenotype; MICs of >64 µg/mL), although MICs can be lower in inducible variants. Second, other strains have acquired the mef gene, which codes for a macrolide efflux pump, which does not affect lincosamides or streptogramin B agents (M phenotype; macrolide MICs of 1–64 µg/mL). On occasion, strains are also found that have other mechanisms of resistance, such as mutations in ribosomal binding sites of macrolides or in ribosomal proteins associated with amino acid chain elongation [20–23]. Isolates with the M phenotype predominate among macrolide-resistant pneumococci in the United States, whereas isolates with the MLSB phenotype predominate in most other countries . Recent US studies have shown that 20%–30% of S. pneumoniae isolates are macrolide resistant, with two-thirds of resistant strains having the M phenotype and one-third having the MLSB phenotype [6, 25].
Current macrolide MIC breakpoints for pneumococci allow for good discrimination between susceptible and resistant strains. However, some investigators have suggested that M phenotype strains may be susceptible under selected circumstances (see the following section). Of note, macrolide resistance among pneumococci has increased in parallel with the increase in use of newer long-acting macrolides [6, 26]. In the United States, this increase has been associated with strains with the M phenotype, whereas in other countries, this increase has been associated with strains with the MLSB phenotype. A promising development for pneumococcal therapy has been the introduction of ketolides, such as telithromycin, which are related to macrolides; these agents appear to be active in vitro and in vivo against macrolide-resistant pneumococci, but they are not as active against H. influenzae .
Pharmacokinetic and Pharmacodynamic Properties of Macrolides
Determinations of in vitro susceptibility have to be interpreted to be clinically meaningful, and this is done by establishing interpretative criteria for MICs, which denote susceptible, intermediate, and resistant categories. The upper limit of susceptibility is generally determined on the basis of the pharmacokinetic and pharmacodynamic properties of each agent in animal models and in humans [28, 29]. Unbound serum levels have been found to be the best reflection of pharmacokinetic properties. The time when such unbound serum levels exceed MICs best reflects the pharmacodynamic properties of time-dependent agents (e.g., macrolides and β-lactams), and the ratios of the unbound serum area under the curve to MIC or the ratios of peak unbound serum levels to MIC best reflect the pharmacodynamic properties of concentration-dependent agents (e.g., quinolones and aminoglycosides). On the basis of studies that correlate pharmacokinetic and pharmacodynamic parameters with bacteriologic eradication in patients or in animal models, susceptibility breakpoints for macrolides have been determined as ⩽0.5 µg/mL for erythromycin, azithromycin, and clarithromycin [30, 31]. These breakpoints are based on the dosing regimen used and the site of infection, and they apply to approved oral dosing regimens and to infections other than meningitis and urinary tract infections; for the latter infections, the concentration profiles of the infection site differ substantially from the serum concentration profiles. They may not apply to intracellular pathogens, because drug distribution within intracellular compartments is much more variable.
Application of the aforementioned macrolide MIC breakpoints to the 2 most common extracellular pathogens associated with CAP, S. pneumoniae and H. influenzae results in 20%–30% of strains of the former organism and >95% of strains of the latter organism being classified as macrolide resistant in the United States [6, 25]. These interpretations are in agreement with numerous animal models of infections due to these pathogens as well as with bacteriologic outcome studies of acute otitis media and sinusitis in humans [30–34]. These studies show that macrolide-susceptible pneumococci respond to macrolide therapy, whereas macrolide-resistant pneumococci of both phenotypes, as well as all H. influenzae isolates, do not respond to therapy. A successful clinical outcome for community-acquired acute respiratory tract infection does not necessarily reflect antimicrobial efficacy, because of the high rate of spontaneous resolution of such disease in outpatients [9, 10] and because of the poor correlation between bacteriologic and clinical outcome in acute otitis media (referred to as the “Pollyanna phenomenon”) .
Some investigators have invented a new body compartment for drugs, such as macrolides and quinolones, that are concentrated intracellularly. This compartment is named alveolar or epithelial lining fluid (ELF), and drug concentrations in such fluids are postulated to correlate with the outcome of pneumonia. This theory is used to explain why macrolides, which have inadequate pharmacokinetic properties on the basis of unbound serum levels, are actually active in vivo against isolates with “low-level resistance,” such as S. pneumoniae with the M phenotype and all H. influenzae. ELF consists of alveolar lining fluid, which is obtained bronchoscopically by intrabronchial lavage [36–39]. After centrifugation of the fluid to remove macrophages, drug levels in ELF are determined and corrected for dilution on the basis of urea concentrations in blood and ELF. This theory has several associated problems that need to be explained. First, drug is predominantly concentrated within alveolar macrophages in ELF, rather than in the aqueous phase of the fluid, in which drug concentrations are low. Extracellular pathogens are exposed only to these aqueous-phase levels. Second, drug in ELF may prevent alveolar epithelial penetration of bacteria, but it is hard to envision how this fluid affects pathogens that have penetrated beyond the alveolar epithelium. Third, drug levels in ELF have mainly been determined in healthy volunteers, whereas ELF levels in actual patients have not been documented. Finally, animal models have been unable to show correlations between bacterial eradication and ELF levels, whereas correlation with unbound serum levels has repeatedly been demonstrated. With the use of dosing regimens equivalent to dosing regimens for humans, bacterial eradication of macrolide-resistant pneumococci with erm- or mef-mediated resistance, as well as eradication of H. influenzae, does not occur in rodent pneumonia models, whereas bacterial eradication does occur with macrolide-susceptible pneumococci [32–34].
Empirical Treatment of CAP in the ERA of Macrolide Resistance
The aforementioned factors need to be considered in relation to the spectrum of pathogens associated with CAP and to the development of guidelines for the empirical treatment of these infections. Current guidelines continue to recommend oral administration of macrolides for the treatment of outpatients with CAP [2, 40–43]. If macrolides are active against S. pneumoniae with M phenotype in vivo, then macrolides are reasonable choices in areas in which such strains are common. However, in light of the evidence discussed above, macrolides may not be as attractive an option.
Although the number of cases of breakthrough pneumococcal bacteremia specified in the article by Lonks et al.  is not large (19 cases) considering the large number of cases of pneumococcal pneumonia that occur, and although, fortunately, there were no deaths among these patients, the patients all had considerable morbidity, which was reflected by their need for hospitalization as well as by the development of empyema in 2 patients and meningitis in 1. In addition, Lonks and colleagues show that some patients could have been treated more aggressively, on the basis of such risk factors as age >50 years, immunosuppression, or underlying chronic pulmonary, renal, or hepatic disease. In areas in which the prevalence of macrolide-resistant pneumococci is high, oral administration of macrolides would be a reasonable option only for patients in PORT class I or II who have no high-risk factors for drug-resistant S. pneumoniae infections, such as recent antibiotic use, recent hospitalization, and day care exposure. However, macrolides have no advantage over β-lactams, unless patients are allergic to β-lactams. In the era of antibiotic-resistant pneumococci, it is somewhat ironic to note that amoxicillin, 875–1000 mg given orally twice daily, is active against ⩾90% of pneumococci worldwide, whereas macrolides are active against <80% of isolates in most countries, with the prevalence of resistance exceeding 80% in some countries in the Far East . On the basis of the high incidence of CAP, we need answers to questions about the real efficacy of antimicrobials in CAP, according to the pathogen involved, the patient age group, the severity of disease, whether outpatients with pneumonia due to “atypical” pathogens actually benefit from receiving antibiotics, and the most cost-effective approach to treating this disease.