Summary of main results

We identified six studies, which had enrolled a total of 1088 participants with pneumonia (92% of whom had ventilator‐associated pneumonia (VAP)) defined according to diverse radiological, clinical and microbiological criteria. There were notable differences in patient characteristics and interventions between studies, though for the most part this was not reflected in statistical heterogeneity.

Considering the five studies that compared fixed durations of antibiotic therapy for VAP (seven to eight days versus 10 to 15 days), a shorter course of therapy was associated with significantly reduced antibiotic exposure in terms of 28‐day antibiotic‐free days in two studies (mean difference (MD) 4.02 days; 95% confidence interval (CI) 2.26 to 5.78), without any significant increase in mortality, duration of mechanical ventilation or duration of hospital stay, and irrespective of pathogen (Chastre 2003a; Fekih Hassen 2009). One study reported that a shorter course of antibiotic therapy was associated with a significant reduction in VAP recurrence due to multi‐resistant organisms (odds ratio (OR) 0.44; 95% CI 0.21 to 0.95) (Chastre 2003a).

Clinical resolution of VAP was non‐significantly lower (Capellier 2012; Kollef 2012; Medina 2007) (OR 0.75; 95% CI 0.49 to 1.15) and recurrence was non‐significantly greater (Capellier 2012; Chastre 2003a; Fekih Hassen 2009; Medina 2007) (OR 1.41; 95% CI 0.94 to 2.12) for patients treated with a short course of therapy. A significant increase in recurrence was observed in the subgroup of patients with VAP due to non‐fermenting Gram‐negative bacilli (NF‐GNB) (Chastre 2003a; Medina 2007) (OR 2.18; 95% CI 1.14 to 4.16), though not for those with VAP due to methicillin‐resistant Staphylococcus aureus (MRSA). Despite this, short‐course therapy for VAP due to NF‐GNB was not associated with other significantly poorer outcomes, in terms of mortality, mechanical ventilation‐free days and length of stay (summary of findings Table for the main comparison).

For critically ill patients with suspected HAP (whether mechanically ventilated or not) but a low probability of pneumonia (according to a day one and day three Clinical Pulmonary Infection Score (CPIS) below seven; Singh 2000), discontinuing antibiotic therapy at day three was associated with significantly shorter duration of antibiotic therapy (mean three days versus 9.8 days; P value = 0.0001 reported), a significantly reduced composite rate of superinfection and antimicrobial resistance (OR 0.29; 95% CI 0.09 to 0.92) and a significantly reduced duration of intensive care unit (ICU) stay.

Overall completeness and applicability of evidence

In this review, we have included only those studies using fixed durations of antibiotic therapy for hospital‐acquired pneumonia (HAP) in the critically ill; thus (unlike the previous version of our review), we have excluded studies that evaluated clinical response (e.g. Micek 2004) or a biomarker (procalcitonin; e.g. Stolz 2009a) as a guide to variable courses of antibiotic therapy. The included studies enrolled 1088 medical and surgical (including cardio‐thoracic and neuro‐surgical) critically ill adults with HAP from ICUs in Africa, Asia, Europe, and North, Central and South America. However, studies were few in number, and participants, interventions and reported outcomes varied considerably. Furthermore, we identified only one study presenting outcome data for patients with HAP who were not receiving mechanical ventilation; separate outcome data are not presented for those subgroups of patients who were and were not receiving mechanical ventilation, and indeed all of the enrolled patients were required to have 'low probability' of pneumonia according to CPIS (Singh 2000). All other included studies focused entirely on patients with VAP. This limitation is important since there may be differences in patient characteristics and in bacterial aetiology between ICU cases of nosocomial pneumonia acquired in the presence or absence of assisted mechanical ventilation (Esperatti 2010; Kohlenberg 2010); therefore, guidance that attempts to extrapolate from studies of patients with VAP to non‐ventilated patients with HAP may potentially be flawed.

For critically ill patients (not necessarily requiring mechanical ventilation) with new radiographic infiltrates and a suspicion but low probability of pneumonia (for example, according to CPIS on day of diagnosis and 72 hours later), a short course (three days) of appropriate antibiotic therapy appears to be more suitable than a prolonged (10‐ to 21‐day) course (Singh 2000), though the generalisability of empiric ciprofloxacin monotherapy may be debated. Current American (ATS/IDSA 2005), British (BSAC 2008), and Canadian (Rotstein 2008) guidelines do not make any specific recommendations for this particular scenario.

A fixed seven‐ to eight‐day course (rather than a 10‐ to 15‐day course) of antibiotic therapy appears to reduce overall antibiotic exposure and, on the basis of one study, to reduce the risk of further infection associated with resistant organisms. However, pooled data from two studies enrolling a total of 176 patients with VAP due to NF‐GNB, indicated that a fixed, short course of antibiotic therapy (seven or eight days) was associated with significantly increased risk of recurrence compared with a more prolonged course (12 or 15 days) (Chastre 2003a; Medina 2007). Outcome data for a subgroup of patients with VAP due to NF‐GNB were presented in only one other study; a trend to lower clinical resolution following short‐course therapy was seen among patients with and without VAP due to NF‐GNB, and a higher CPIS was observed at day 10 after short‐course therapy (Kollef 2012).

Given the (non‐significantly) lower clinical resolution among patients treated with short‐course therapy and significantly higher recurrence among patients with VAP due to NF‐GNB, data regarding the continued presence of bacterial pathogens (e.g. Pseudomonas sp.) in respiratory specimens during and after fixed‐duration therapy may also have been of interest but were not routinely reported in the included studies.

Diagnostic criteria for defining an episode of pneumonia were not uniform, and no study reported using the new Centers for Disease Control and Prevention (CDC) classification of ventilator‐associated events, including possible VAP (PVAP).

Unfortunately, insufficient outcome data were available to allow exploration of other subgroups of interest, for example: presence of chronic respiratory illness, illness severity, prior administration of antibiotic therapy, prior duration of mechanical ventilation (i.e. early‐ or late‐onset VAP) and class of antibiotic (including combination therapy). Durations of therapy incorporating more novel modes of administration of potential interest were confounded by multiple interventions (e.g. extended infusion versus standard infusion, doripenem versus imipenem, short course versus prolonged course) or did not meet the inclusion criteria for this review (e.g. use of nebulised antibiotics, such as Niederman 2012).

Subsequent infection due to resistant organisms was reported in only two studies (one of which reported as a composite measure only). Emergence of extra‐pulmonary infection, in particular, C. difficile‐associated disease, was surprisingly not reported at all.

Both American (ATS/IDSA 2005) and British (BSAC 2008) guidelines recommend that efforts should be made to shorten antibiotic therapy for HAP (including patients with VAP) to a duration of seven to eight days, where there has been clinical response and, in the case of American Thoracic Society/Infectious Diseases Society of America (ATS/IDSA) guidelines, where the causative organism is not P. aeruginosa. Overall, our review was broadly consistent with these guidelines; nevertheless, in the context of VAP due to NF‐GNB, short‐course therapy is unlikely to be appropriate for all patients and the optimal duration remains uncertain.

Quality of the evidence

Differences in participants, interventions and reported outcomes between studies meant that the number of studies contributing to a particular summary statistic was very small. We considered selection bias to be at low risk in three of the six included studies. Blinding occurred in only one of the six studies. We felt that there was significant other bias in two of the studies, because of multiple differences between intervention groups (Kollef 2012), and because of contamination of the control group with early termination of the study (Singh 2000). We therefore assessed the quality of evidence according to the GRADE classification as low to moderate for the main outcome measures (summary of findings Table for the main comparison).

Potential biases in the review process

We attempted to minimise bias in a number of ways. Our search strategy utilised the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE. In addition, we searched review articles and the 'grey literature' to minimise reporting bias. We established inclusion criteria prior to conduct of searches and, where necessary, we contacted trial authors to clarify study eligibility. We collected data using a standardised data collection form; when not available, we contacted trial authors to provide additional relevant data. We systematically assessed internal validity using the Cochrane 'Risk of bias' tool.

The potential for study heterogeneity was evident at the planning stage of this review, for example, in terms of methodology, diagnostic criteria, illness severity and antibiotic class. We identified relevant subgroups a priori. However, outcome data were typically not presented in sufficient detail to enable adequate subgroup analysis (e.g. according to prior administration of antibiotic therapy, prior duration of mechanical ventilation, classes of antibiotic including combination therapy) and given methodological differences we used a random‐effects model for meta‐analysis.

With a very small number of included studies, we were able to meta‐analyse data from a maximum of four (and more frequently only two or three) studies per outcome. We included studies that focused on early‐onset VAP (Capellier 2012), and those that primarily recruited patients with later‐onset VAP (e.g. Chastre 2003a; Kollef 2012), with the intention of analysing according causative organism (e.g. NF‐GNB). We felt that this approach was appropriate, particularly given that the relationship between early‐ versus late‐onset VAP and the likelihood of pneumonia being due to a resistant organism has been questioned in recent reports (Gastmeier 2009; Martin‐Loeches 2015; Restrepo 2013).

Due to marked differences in participants, interventions and reported outcomes between Singh 2000 and other studies, we considered outcomes for that study separately.

We performed sensitivity analysis to exclude the results of Kollef 2012 from 28‐day mortality data, which did not produce different findings. We also performed sensitivity analysis to exclude Capellier 2012 (given its focus on early onset VAP) and Medina 2007 (since it has been published only in abstract form) from the recurrence results, which led to recurrence among the subgroup of patients with VAP due to NF‐GNB no longer being statistically significant.

Agreements and disagreements with other studies or reviews

Three systematic reviews specifically focusing on duration of antibiotic therapy for VAP have been published (Dimopoulos 2013; Dugan 2003; Grammatikos 2008). The two earlier of these reviews included observational studies and studies focused on diagnostic strategies (for example, invasive versus noninvasive diagnosis of VAP, Fagon 2000). Notably, the Dugan 2003 review also appears to have been written before publication of the Pneuma study, the largest randomised controlled trial (RCT) to date to investigate duration of therapy for VAP (Chastre 2003a). The most recently published systematic review, Dimopoulos 2013, restricted itself to RCTs as did we, but excluded Singh 2000 (presumably on the basis that it included non‐ventilated patients and all patients had a low probability of pneumonia) and Medina 2007 (since it is currently published in abstract form only). In their review, relapse was chosen as a primary outcome measure rather than recurrence.

The conclusions of Grammatikos 2008, namely that short‐course antibiotic therapy for VAP not caused by NF‐GNB does not appear to adversely affect mortality, recurrence or length of stay (Grammatikos 2008), are consistent with the findings from our review and those of Dimopoulos 2013. Dugan 2003 concluded that shorter‐course therapy may be associated with reduced antimicrobial resistance, which is consistent with Chastre 2003a. However, studies published subsequent to their review have been less supportive of the benefits of short‐course therapy in reducing ICU and hospital length of stay. Having excluded Medina 2007, the systematic review published by Dimopoulos 2013 reported a non‐significantly greater relapse in patients with VAP due to NF‐GNB treated with short‐course therapy, rather than the significantly greater recurrence that we found upon meta‐analysis.

Our finding that short‐course antibiotic therapy for VAP in the absence of NF‐GNB may be adequate, without increased risk of recurrence or mortality, is consistent with data from observational studies. Dennesen 2001 demonstrated in 27 cases of VAP treated with appropriate initial antimicrobial therapy that clinical resolution occurred within the first six days of treatment and that S. pneumoniae, H. influenzae and S. aureus were eradicated within this period. In contrast, in Dennesen's study P. aeruginosa was isolated from endotracheal aspirates in all patients 15 days after the initiation of therapy, despite a mean 12.7 days of antibiotic treatment. Similarly, after seven days' appropriate antibiotic therapy for VAP, El Solh 2007 observed a disparity between clinical response and persistent respiratory colonisation, with P. aeruginosa continuing to be cultured from broncho‐alveolar lavage (BAL) despite improvements in CPIS. This persistence may be clinically significant. Zhuo 2008 demonstrated that the bacterial load in mechanically ventilated patients colonised with P. aeruginosa but without meeting the clinical criteria for VAP may predict outcome; in their study, high P. aeruginosa load was associated with a significantly greater 28‐day mortality.

A reliable measure of clinical or microbiological response to treatment has appeal as a means of identifying those for whom a shorter course of therapy may be appropriate, particularly among those with VAP due to NF‐GNB. Kollef observed 10 days after initiation of antibiotic therapy for VAP due to Gram‐negative bacteria, that CPIS was lower among patients treated for 10 days rather than seven days (Kollef 2012). In a non‐randomised study, use of repeat BAL to guide duration of therapy for VAP in surgical/trauma patients led to a reduction in duration of therapy among patients with VAP due to NF‐GNB compared with a control group (mean 10.7 versus 14.4 days respectively) without there being any overall increase in ventilator‐free ICU days, in‐hospital mortality and VAP relapse (Mueller 2007). Bouadma 2010, Pontet 2007, and Stolz 2009a reported reductions in antibiotic exposure in patients with VAP treated according to a procalcitonin‐guided algorithm compared with standard treatment in RCTs, without observing an increase in mortality. Recurrence rates did not differ between groups in one of these studies (Pontet 2007), though outcomes for patients specifically with NF‐GNB VAP were not reported at all.

To summarise, there are very few data for non‐ventilated patients with HAP. For patients with VAP, although short‐course (seven to eight days) therapy for VAP generally appears to be safe, the increased rate of recurrence of VAP due to NF‐GNB among patients administered a fixed, short course of antibiotic therapy demonstrated in our review may reflect inadequate microbiological and/or clinical response to treatment. It would therefore appear that further work is warranted to establish the optimal duration of therapy particularly for these patients, perhaps involving strategies that enable individualisation of treatment.