Shorter Versus Longer Courses of Antibiotics for Infection in Hospitalized Patients: A Systematic Review and Meta-Analysis

Secondary Outcomes

Three studies^30,31,42 of 286 patients (with VAP or intra-abdominal infection) evaluated the emergence of MDR organisms. The overall risk difference was d = −9.0% (95% CI, −19.1%-1.1%; P = .081). There was no statistically significant heterogeneity between studies (I² = 7.6%, P = .34).

Seven studies examined LOS—3 in the intensive care unit (ICU)^30,31,43 and 4 on the wards^32,36,40,41—none of which found significant differences between treatment arms. Across 3 studies of 672 patients, the weighted average for ICU LOS was 23.6 days in the short arm versus 22.2 days in the long arm. Across 4 studies of 235 patients, the weighted average for hospital LOS was 23.3 days in the short arm versus 29.7 days in the long arm. This difference was driven by a 1991 study⁴¹ of spontaneous bacterial peritonitis (SBP), in which the average LOS was 37 days and 50 days in the short- and long-course arms, respectively.

Three studies^32,41,43 of 186 total patients (with SBP or hospital-acquired infection of unknown origin) examined the cost of antibiotics. The weighted average cost savings for shorter courses in 2016 US dollars⁴⁸ was $265.19.

Three studies^30,33,43 of 618 patients evaluated cases of CDI, during 10-, 30-, and 180-day total follow-up. The overall risk difference was d = 0.7% (95% CI, −1.3%-2.8%), with no statistically significant heterogeneity between studies (I² = 0%, P = .97).

Study Quality

Included studies scored 2-5 on the Cochrane Collaboration Risk of Bias Tool (supplementary Figure 4). Four studies had an overall low risk of bias,^36,37,43,46 while 15 had a moderate to high risk of bias (supplementary Table 3).^{29-35,38-42,44,45,47} Common sources of bias included inadequate details to confirm adequate randomization and/or concealment (n = 13) and lack of adequate blinding (n = 18). Two studies were stopped early,^37,42 and 3 others were possibly stopped early because it was unclear how the number of participants was determined.^29,33,47 Covariate imbalance (failure of randomization) was present in 2 studies.^37,47 There was no evidence of selective outcome reporting or publication bias based on the funnel plots (supplementary Figure 5).

In this study, we performed a systematic review and meta-analysis of RCTs of shorter versus longer antibiotic courses for adults and adolescents hospitalized for infection. The rate of clinical cure was indistinguishable between patients randomized to shorter versus longer durations of antibiotic therapy, and the meta-analysis was well powered to confirm noninferiority. The lower 95% CI indicates that any potential benefit of longer antibiotics is not more than 1%, far below the typical margin of noninferiority. Subgroup analysis of patients hospitalized with CAP also showed noninferiority of a prespecified shorter treatment duration.

The rate of microbiologic cure was likewise indistinguishable, and the meta-analysis was again well powered to confirm noninferiority. Any potential benefit of longer antibiotics for microbial cure is quite small (not more than 4%).

Our study also demonstrates noninferiority of prespecified shorter antibiotic courses for mortality. Shorter- and longer-term mortality were both indistinguishable in patients randomized to shorter antibiotic courses. The meta-analyses for mortality were well powered, with any potential benefit of longer antibiotic durations being less than 2% for short-term and less than 6% for long-term mortality.

We also examined for complications related to antibiotic therapy. Infection recurrence was indistinguishable, with any potential benefit of longer antibiotics being less than 6%. Select infections (eg, VAP due to NF-GNB, catheter-associated UTI) may be more susceptible to relapse after shorter treatment courses, while most patients hospitalized with infection do not have an increased risk for relapse with shorter treatment courses. Consistent with other studies,⁸ the emergence of MDR organisms was 9% less common in patients randomized to shorter antibiotic courses. This difference failed to meet statistical significance, likely due to poor power. The emergence of MDR pathogens was included in just 3 of 19 studies, underscoring the need for additional studies on this outcome.

Although our meta-analyses indicate noninferiority of shorter antibiotic courses in hospitalized patients, the included studies are not without shortcomings. Only 4 of the included studies had low risk of bias, while 15 had at least moderate risk. The nearly universal source of bias was a lack of blinding. Only 1 study³⁷ was completely blinded, and only 3 others had partial blinding. Adequate randomization and concealment were also lacking in several studies. However, there was no evidence of selective outcome reporting or publication bias.

Our findings are consistent with prior studies indicating noninferiority of shorter antibiotic courses in other settings and patient populations. Pediatric studies have demonstrated the success of shorter antibiotic courses in both outpatient⁴⁹ and inpatient populations.⁵⁰ Prior meta-analyses have shown noninferiority of shorter antibiotic courses in adults with VAP^15,16; in neonatal, pediatric, and adult patients with bacteremia¹⁷; and in pediatric and adult patients with pneumonia and UTI.^3-6,18,19 Our meta-analysis extends the evidence for the safety of shorter treatment courses to adults hospitalized with common infections, including pneumonia, UTI, and intra-abdominal infections. Because neonatal, pediatric, and nonhospitalized adult patients may have a lower risk for treatment failure and lower risk for mortality in the event of treatment failure, we focused exclusively on hospitalized adults and adolescents.

In contrast to prior meta-analyses, we included studies of multiple different sites of infection. This allowed us to assess a large number of hospitalized patients and achieve a narrow margin of noninferiority. It is possible that the benefit of optimal treatment duration varies by type of infection. (And indeed, absolute duration of treatment differed across studies.) We used a random-effects framework, which recognizes that the true benefit of shorter versus longer duration may vary across study populations. The heterogeneity between studies in our meta-analysis was quite low, suggesting that the results are not explained by a single infection type.

There are limited data on late effects of longer antibiotic courses. Antibiotic therapy is associated with an increased risk for CDI for 3 months afterwards.¹¹ However, the duration of follow-up in the included studies rarely exceeded 1 month, which could underestimate incidence. The effect of antibiotics on gut microbiota may persist for months, predisposing patients to secondary infections. It is plausible that disruption in gut microbiota and risk for CDI may persist longer in patients treated with longer antibiotic courses. However, the existing studies do not include sufficient follow-up to confirm or refute this hypothesis.

Our review has several limitations. First, we included studies that compared an a priori-defined short course of antibiotics to a longer course and excluded studies that defined a short course of antibiotics based on clinical response. Because we did not specify an exact length for short or long courses, we cannot make explicit recommendations about the absolute duration of antibiotic therapy. Second, we included multiple infection types. It is possible that the duration of antibiotics required may differ by infection type. However, there were not sufficient data for subgroup analyses for each infection type. This highlights the need for additional data to guide the treatment of severe infections. Third, not all studies considered antibiotic duration in isolation. One study included a catheter change in the short arm only, which could have favored the short course.³³ Three studies used different doses of antibiotics in addition to different durations.^35,45,47 Fourth, the quality of included studies was variable, with lack of blinding and inadequate randomization present in most studies.