Details of this observational study have been presented elsewhere (10), so the following represents an abbreviated version. Between December 1, 1998 and December 31, 2000, 844 consecutive adults with pneumococcal bacteremia were enrolled in 21 hospitals in 10 countries on 6 continents. Patients were monitored for at least 14 days after the first positive blood culture or longer if the patient remained hospitalized.

All patients 15 years of age and older who had at least one blood culture positive for S. pneumoniae during the 25-month study period were included. Patients were classified as “elderly” if they were 65 years of age or older; “immunosuppressed” if they had human immunodeficiency virus infection, hematologic malignancy, or an autoimmune disorder, had received either organ or bone marrow transplant or cancer chemotherapy within 4 weeks of pneumococcal bacteremia, or had undergone prior splenectomy; and/or as having “underlying chronic disease” if they had heart, lung, liver, or renal disease or diabetes mellitus. Patients were defined as critically ill as calculated by a Pitt bacteremia score greater than 4, as described previously (10, 11). The APACHE (Acute Physiology and Chronic Health Evaluation System) II score was calculated for patients admitted directly to the intensive care unit (ICU).

Antibiotic therapy was eligible for analysis if the total daily dose of an agent was at least the minimum dose recommended for treatment of systemic infection. Patients who did not receive antibiotic(s) on the day of admission were excluded. Monotherapy was defined as receipt of the same single antibiotic within the first 2 days of obtaining a positive blood culture; likewise, combination therapy was defined as receipt of the same two antibiotics within the first 2 days of positive blood culture. Patients who received no antibiotic therapy or delayed treatment (more than 24 hours after admission), or had no consistent antibiotic regimen (e.g., 1 day of monotherapy plus 1 day of combination therapy) over the initial 2 days of admission were excluded.

Differences in categorical variables were calculated by χ² test with the Yate correction or Fisher exact test. The Kaplan–Meier product limit method was used to construct survival curves for patients receiving combination and monotherapy regimens. The patients were stratified by severity of illness and the survival curves were compared using the Mantel–Cox test (12). A logistic model was used to adjust for additional risk factors in evaluation of the impact of combination therapy on 14-day mortality. Factors associated with increased mortality by univariate analysis were entered into the model in addition to receipt of combination therapy.

Eight hundred forty-four consecutive cases of pneumococcal bacteremia were enrolled. Patients were excluded if they received no antibiotic therapy (43); received no antibiotic therapy on Day 2 (30); received delayed (more than 24 hours after admission) antibiotic therapy (23); underwent an inconsistent regimen (e.g., 1 day of monotherapy plus 1 day of combination therapy) (86); or underwent different monotherapy or combination therapy regimens on Day 1 and Day 2 (70). The remaining 592 patients were available for analyses of monotherapy versus combination therapy.

As we have reported previously, 16.5% of patients (139 of 844) died by Day 14 (10). The risk of death was almost eightfold greater for critically ill patients compared with patients who were less ill at hospital admission (54.6 versus 7.3%, p = 0.0001). The 14-day mortality was not significantly different for all patients receiving combination versus monotherapy (10.4 versus 11.5%, p = NS). Among critically ill patients (n = 94), however, combination antibiotic therapy was associated with lower mortality (14-day mortality, 23.4 versus 55.3%, p = 0.0015) (Figure 1)

. The most common monotherapy regimens prescribed for the critically ill patients were β-lactam agents (43, with 25 receiving a third-generation cephalosporin), azithromycin (2), ciprofloxacin (1), and clindamycin (1). The most frequent combination therapies prescribed were β-lactam/macrolide (14), vancomycin/β-lactam (12), β-lactam/aminoglycoside (7), vancomycin/other antibiotic (4), β-lactam/quinolone (4), double β-lactam therapy (2), β-lactam/chloramphenicol (2), β-lactam/trimethoprim-sulfamethoxazole (1), and clindamycin/quinolone (1).

This difference in mortality remained significant when analyzed according to in vitro activity, as defined by the National Committee for Clinical Laboratory Standards (NCCLS), of the monotherapy or combination drug therapy. Penicillin-susceptible isolates (minimal inhibitory concentration, less than 0.12 μg/ml) and penicillin-nonsusceptible isolates (minimal inhibitory concentration, 0.12 μg/ml or more) were distributed equally between the two groups (Table 1)

. One hundred percent (94 of 94) and 98.9% (93 of 94) were susceptible to ceftriaxone and cefotaxime, respectively, using recent NCCLS breakpoints (13). When all treatments were active in vitro, mortality for combination versus monotherapy was 19.4 versus 60% (p = 0.0006), respectively. When at least one of the drugs of the combination therapy was active in vitro, mortality for combination therapy versus monotherapy was 18.2 versus 60% (p = 0.0003), respectively. No patient received combination therapy in which both antibiotics were inactive in vitro. The demographics for the monotherapy versus combination therapy groups were comparable except for human immunodeficiency virus therapy and mechanical ventilation (Table 1). When adjusted for mechanical ventilation and human immunodeficiency virus positivity by multivariate analysis, combination therapy remained a significant factor in decreasing mortality (Table 2) . Severity of illness as measured by APACHE II or Pitt bacteremia score was essentially identical for the two groups.

One hundred twelve patients were admitted to the ICU on admission; note that not all were critically ill, and therefore the number is not 94. The severity of illness was nearly identical in patients receiving combination therapy and monotherapy (mean Pitt bacteremia score, 5.0 versus 5.5, respectively; APACHE II score, 19.0 versus 19.2, respectively). Nevertheless, the patients receiving combination therapy had a significantly lower 14-day mortality than those receiving monotherapy (8.2 versus 23.1%, p = 0.03). The superiority of combination therapy was also evident in various logistic regression models of ICU patients (data not shown).

The combination regimens were further examined to determine whether the difference seen in mortality rates between the two groups of critically ill patients was due to a specific antibiotic or combination of antibiotics. This was not the case; when compared with monotherapy, mortality rates were lower for antibiotic combinations that included β-lactams (26.8%, p = 0.007), vancomycin (6.3%, p = 0.0006), or macrolides (14.3%, p = 0.007). Numbers were too small to allow analysis of fluoroquinolones, aminoglycosides, and clindamycin. The impact on mortality for combination regimens was most evident when a second antibiotic was added to a β-lactam agent (the most commonly prescribed therapy for both groups). The mortality was 26.8% (11 of 41) for those who received a β-lactam–containing combination regimen as compared with 58.4% (24 of 41) (p = 0.004) for those who received a β-lactam as monotherapy. Because no patient received vancomycin as monotherapy and only two patients were given a macrolide as monotherapy in the critically ill group, no comparisons were made between combination versus monotherapy when either vancomycin or a macrolide was included in both treatment arms.

Varying durations of receipt of antibiotics have been used to define combination therapy versus monotherapy in the literature; our analyses assessed mortality with a 2-day duration (see Methods for definition). Thus, we also analyzed the 14-day mortality for patients receiving antibiotic therapy for 1 day (as defined by Waterer and coworkers [7]) (n = 775 patients) as well as 3 days (n = 510 patients). Mortality was significantly lower for combination therapy compared with monotherapy if the 1-day duration was used (43% [31 of 72] versus 60% [45 of 75], p < 0.026) or a 3-day duration was used (11.4% [4 of 35] versus 42.8% [12 of 28], p < 0.008).

When therapy was analyzed by country of origin, insufficient statistical power existed for analysis for some individual countries; when the number of cases exceeded 12 in a country, there was a clear-cut trend for superiority of combination therapy in those countries. When adjusted for country, patients receiving combination therapy remained a significant factor in decreasing mortality in logistic regression models (data not shown).

Findings from this large, prospective, international investigation suggest that the administration of combination antibiotic therapy results in increased survival among critically ill patients with pneumococcal bacteremia (Figures 1A and 1B). There was a more than twofold increase in survival among recipients of combination antibiotic therapy as compared with that for recipients of monotherapy regimens. Receipt of combination therapy and mechanical ventilation were significant factors in improving outcome by both univariate and multivariate analyses. The findings of significant increases in mortality for both critically ill and elderly patients (10), two well recognized subgroups at increased risk of dying, bolster the validity of the study's principal observation that outcome is improved with combination antibiotic therapy for severely ill patients with pneumococcal bacteremia.

The in vitro antibiotic resistance for monotherapy or one of the two antibiotics in the combination regimen did not significantly affect mortality. The clinical impact of in vitro resistance as defined by current NCCLS guidelines has been described in an earlier report derived from this study in which penicillin and third-generation cephalosporin resistance in vitro was not a risk factor for mortality (10). Specifically, outcomes in both critically ill and noncritically ill patients were not influenced by whether they received concordant or discordant therapy based on pneumococcal in vitro susceptibility results.

There are several strengths of the current investigation that are not shared by the three previous retrospectively conducted studies (6–8) that examined the impact of combination therapy on mortality. Our study enrolled 844 consecutive patients who were prospectively enrolled from 21 hospitals in 10 countries of 6 continents between December 1, 1998 and January 2001. In contrast, the three previously published studies (6–8) were retrospectively conducted and enrolled patients from three geographically confined areas, two (6, 7) of which were in the United States and one in Spain (8). The applicability of findings from these earlier investigations may be limited because case-fatality rates of community-acquired bacteremic pneumococcal illness differ greatly by geographic locale (14). The numbers of patients included in these three trials were fewer than that enrolled in our study. Our patient enrollment was performed more recently in the era of increasing antibiotic resistance, whereas the other studies were performed in 1978, 1991, and 1996, respectively. A retrospective analysis of a hospital claims-made database, which included more than 44,000 patients seen between 1997 and 1999 for community-acquired pneumonia, also found that combination antibiotic therapy reduced mortality (15).

It should be noted that the definition of combination antibiotic therapy was only 1 day in the Waterer and coworkers study (7) and was not specified in the other two retrospective studies (6, 8). Although our analyses given in the Results applied to the 2-day definition of duration of antibiotic therapy, we also assessed the impact of combination therapy 1 and 3 days after admission because our previous study showed that 64.5% of these patients died not more than 72 hours after blood culture had been obtained (10). Mortality for combination therapy remained significantly lower for all durations used (Results). We evaluated quality of support services other than antibiotics by analyzing a subgroup of patients admitted to the ICU. Again, combination therapy led to significantly lower mortality than monotherapy. Finally, mortality was significantly lower for combination therapy versus monotherapy when controlled for ICU admission and country of origin.

There were limitations to our study. First, as with other studies exploring this issue, these results were not from a randomized controlled study. However, the combination and monotherapy groups were generally comparable (Table 1), and severity of illness as measured by APACHE II and Pitt bacteremia scores was identical.

Second, we were unable to define optimal duration of combination therapy because antibiotic therapy was changed in 35% of patients by Day 5. Patients receiving monotherapy who were doing poorly often received additional antibiotics, whereas patients receiving combination therapy often had antibiotics tapered because of knowledge of in vitro susceptibility or improvement in clinical status.

Third, we were unable to clearly delineate the basis for the superiority of the combination regimen, and we were unable to precisely determine which specific components of combination therapy would be most effective. Earlier retrospective analyses (6, 8) of two smaller study populations suggested that the addition of a macrolide to a β-lactam was the best combination regimen. In our study not only were regimens that included macrolides as one of the combination agents associated with better outcomes, but nonmacrolide combination regimens were also successful in reducing mortality among the severely ill patients. Thus, the unique role of macrolides possessing antiinflammatory activity and/or coverage against atypical copathogens postulated by other investigators could not be supported.

In the report of a retrospective study of pneumococcal bacteremia, Austrian and Gold demonstrated that some patients, despite appropriate antibiotic therapy and critical supportive care (as was available more than four decades ago), “cannot be prevented from dying” (16). It is difficult to make direct comparisons between this retrospective study (14) and our prospective observational study. Nevertheless, the dramatic reduction in early mortality among severely ill patients in our current investigation has led us to speculate that combination antibiotic therapy might salvage some critically ill patients who might be otherwise “destined to die” (6).

We emphasize that the potential benefits of combination antibiotic therapy in our study and in the Waterer and coworkers study (7) were limited to more severely ill patients. For noncritically ill patients, there was no difference in the 14-day mortality for patients treated with monotherapy versus combination therapy (Figure 1A). Because the overall mortality of patients with mild illness due to bacteremic pneumococcal illness is so low, minimal benefit will be gained by adding a second antibiotic. As mentioned, the optimal duration of combination therapy is undefined, but because mortality is highest within the first 3 days (Figure 1) (10), it seems both intuitive and reasonable that 3–5 days might be sufficient. If our conclusions are confirmed by other prospective studies, we recommend that clinicians target combination antibiotic treatment only for those patients who are critically ill, and limit the duration of the combination to 3–5 days. Such an approach may be beneficial yet minimize antibiotic overuse.