Treatment of severe hospital-acquired and ventilator-associated pneumonia: a systematic review of inclusion and judgment criteria used in randomized controlled trials

We searched PubMed, Embase, ICAAC and ECCMID for publications from 1994 to 2016. The search string used was:

1-(Pneumonia and (hospitalized or acquired or ventilatory-associated or nosocomial or health-associated)) and (antibiotic treatment or (antibiotic therapy) or antibiotic or (antibacterial agents) or (anti-infective agents) or antimicrobial)

It was complemented by:

2-Pneumonia and (hospitalized or acquired or ventilatory-associated or nosocomial or health-associated)) and (doripenem or meropenem or impenem or linezolid or vancomycin or piperacillin or (piperacillin and tazobactam) or ceftazidim or ceftriaxone or cefotaxime or ceftobiprole or ciprofloxacin or levofloxacin or moxifloxacin or piperacillin or teicoplanin or cefepime or ticarcillin or (ticarcillin and clavulanic acid) or amikacin or tobramycin or gentamicin) and (randomized controlled trial).

Of note, some studies were identified through sources other than a database search, including contact with researchers.

All RCTs evaluating the efficacy of antimicrobials in HAP/VAP in adult patients in the ICU from 1994 to 2016 were considered as eligible for this systematic review. The study selection process is detailed in a flow diagram (Fig. 1). The search period was limited to twenty years because of the evolution of publication requirements and treatment concepts. All articles were independently screened by two authors (J-RZ and CA). Full texts (and clinical trial registry data (ClinicalTrials.gov website)) of all potentially relevant studies were analyzed to assess eligibility. Disagreements on the inclusion/exclusion of studies were resolved in consultation with other co-authors (EW and J-FT). We included studies comparing one antibiotic agent to another, monotherapy to combination therapy, inhaled to systemic route or extended/continuous infusion to intermittent administration. Finally, in January 2017 we also searched the ClinicalTrials.gov website for characteristics of currently unpublished RCTs comparing HAP/VAP antimicrobial treatment strategies.

The following data were collected for each study included: inclusion and exclusion criteria, design, primary endpoint and population in which it was analyzed, definition of clinical cure, time point used for the test of cure (TOC) visit, and the statistical hypothesis. Data extraction from the studies was confirmed by an independent reviewer. At the end, a global quality assessment of each RCT was performed using an 8-item scale allowing calculation of a global quality score adapted to the focus of this systematic review.

The main characteristics of the 27 studies included in the systematic review (see Additional file 1 for references) and of their enrolled populations are described in Table 1.

The number of patients enrolled in the intention-to-treat (ITT) population was highly variable and ranged from 35 to 945. Patients were predominantly male and the mean age across trials was 50.5–70.0 years. In most studies (92%), severity of illness was reported using the APACHE II score and its mean value varied from 13.8 to 24.0 across trials. The rate of comorbidity was rarely reported.

Sixteen studies co-enrolled patients with HAP and VAP and 11 studies exclusively enrolled patients with VAP. Among the former, the proportion of patients with VAP greatly varied from 27 to 91%. In some studies, randomization was stratified by patients’ severity of illness (three trials) and on the rate of VAP (two trials). Whereas 18 studies (67%) included all HAP/VAP episodes regardless of the causative pathogen, some only enrolled patients with HAP/VAP related to specific pathogens. The duration of hospital stay before the onset of pneumonia was required to be longer than 48 hours and 72 hours in 18 studies (67%) and 7 studies (26%), respectively (not mentioned in the remaining study). However, the duration of the mechanical ventilation was an inclusion criterion in only 11 VAP studies (41%).

The diagnosis of pneumonia was consistently based on clinical and radiological (new lung infiltrate on chest radiography) findings but microbiological confirmation was required in only 15 studies (55%). Among the remaining studies, the proportion of patients analyzed who had positive respiratory samples varied from 62 to 82%. Clinical signs leading to suspicion of HAP/VAP were highly variable across the 27 studies. These clinical findings and the percentage of studies using them as inclusion criteria are shown in Fig. 2a. In 73% of studies, a predefined number of respiratory symptoms and signs of sepsis were required for pneumonia diagnosis and subsequent inclusion (Fig. 2b).

Various exclusion criteria were also reviewed. In five studies (20%), patients with severe disease, as reflected by high Acute Physiology and Chronic Health Evaluation (APACHE) score (four studies) or Simplified Acute Physiology Score (SAPS) (one study), concomitant septic shock (two studies) or acute respiratory distress syndrome (one study) were excluded. Neutropenia and immunosuppression (various definitions) were considered as exclusion criteria in 13 studies (48%) and 7 studies (26%), respectively. Finally, patients with chronic respiratory diseases were excluded in nine studies (33%).

Only a few studies tried to balance heterogeneity by stratifying the randomization according to the severity of the acute disease (APACHE in two studies and partial arterial oxygen pressure/fraction of inspired oxygen (PaO2/FiO2) ratio in one study) or the characteristics of the pneumonia (i.e., HAP or VAP, in three studies).

There were 21 multicenter studies (78%), 14 double-blinded studies (52%) and 13 open-labeled studies (48%). The duration of study therapy varied from one study to another (from 5 to 21 days) and was left at the discretion of the investigators in 37% of studies.

Nine trials (33%) used a non-inferiority design with non-inferiority margins (i.e., absolute percentage difference in the primary outcome acceptable for non-inferiority to be established) mentioned in eight of them (20%, 15% and 10% in three studies, four studies, and one study, respectively), but the scientific evidence leading to calculation of the non-inferiority margin was missing in two of them. Six studies (22%) and three studies (11%) were equivalence and superiority trials, respectively. Of note, a pre-specified statistical hypothesis was lacking in nine studies (33%) and sample size calculation was not mentioned in nine studies (33%).

Our work revealed marked heterogeneity among the primary endpoints reported in HAP/VAP trials. A single primary endpoint in one population was reported in 13 studies (48%): a clinical cure in 12 trials (44%), CPIS decrease in 1 trial (4%), and 28-day mortality in 1 (4%).

In seven studies, the clinical cure was analyzed in several populations (two populations in six studies, and three populations in one study) without specifying which one was the primary endpoint. In four studies, the clinical cure and microbiological cure were both reported as the primary endpoint and safety was also added in two studies. Altogether, clinical cure was the primary endpoint (or one of the “primary” endpoints) in 24 studies (89%) and mortality in only 2 studies. The various populations used to analyze clinical cure are displayed in Fig. 3a.

For a global assessment of study quality, we calculated an 8-point quality score based on 8 systematically evaluated items, reflecting enrollment criteria and methodology (Table 2). Altogether, this score ranged from 2 to 7, thereby confirming the heterogeneity between studies.

Interestingly, elements included in the definition of clinical cure greatly differed among studies (Fig. 3b). Remission of signs and symptoms of pneumonia were included in the definition of clinical cure in all studies, but was only partial in 11 (41%) studies. Chest radiography findings were not systematically taken into account (22 studies, 82%) and they qualified for clinical cure whether the improvement was complete, partial or only a lack of progression in 18 studies (67%). Finally, no need for additional antibiotics during follow up and decrease in the Clinical Pulmonary Infection Score (CPIS) [8] were used in 26% and 11% of studies, respectively. The timing of the TOC visit was different across trials. The time point was between day 7 and day 21 after the end of therapy in 56% percent of the studies; in the remaining trials, it ranged from the end of treatment (4 studies, 15%) till day 30 after treatment (3 studies, 12%). These differences in the definition of clinical cure and in the population and time points used to assess it led to wide variability in reported clinical cure rates that ranged from 23 to 77% (Table 1).

All-cause mortality was reported in 22 studies (81%) (primary endpoint in one study) and HAP/VAP-related mortality was reported in 2 studies (Table 1). The time points used to assess mortality were highly variable: day 28 in nine studies, at the end of the study in eight studies, and day 30 and day 60 in four studies and two studies, respectively. At day 28, the mean mortality rate was 23%, with broad variability of the reported all-cause mortality rates from one study to another (range 10–67%; Table 1).

The methodological quality of studies improved over time. We specifically analyzed the six studies published after the EMA workshop on antimicrobials held in 2011, i.e., between 2012 and 2016. Among the six studies, four had a non-inferiority design (with anticipated statistical approach and non-inferiority margin), were double-blinded, and assessed clinical cure in one population as the single primary endpoint. Clinical criteria for diagnosis of HAP/VAP were more homogenous and at least two clinical signs were consistently required. Furthermore, all cases of pneumonia that were analyzed were microbiologically proven in four of the six trials. Finally, the only items that did not improve were the definition of clinical cure and the time points used to assess it. Calculation of the mean quality score confirmed this global improvement: the mean score of studies published after 2011 was higher than that of older studies (5.3 vs 3.9) (Table 2).

Thirteen currently unpublished RCTs that tested new antimicrobial treatment strategies in patients with HAP/VAP or with severe infections due to carbapenem-resistant bacteria including HAP/VAP are registered on ClinicalTrials.gov (Table 3). Interestingly, the primary endpoint remained highly variable: 28-day all-cause mortality in four studies (all comparing systemic molecules) and clinical cure (with a variable definition, and using variable timing of assessment) in nine studies (four of them comparing systemic molecules and five testing the efficacy of nebulized antimicrobials) and a favorable clinical response. Only five studies described which population was analyzed.

In this systematic review comprising 27 published RCTs addressing the efficacy of antimicrobials for treatment of HAP/VAP in critically ill patients, we found that the enrolled populations, clinical trial design, and endpoints assessed vary greatly between studies. This variability is intuitive but its quantification is of importance because, as shown by the variable mortality and clinical cure rates that were reported, it impacted the results of RCTs. This heterogeneity may therefore influence the ability of studies to demonstrate differences between investigational drugs and comparators. These results are not surprising as far as there are even differences in the guidance from regulatory agencies on how to evaluate treatments for HAP/VAP [2, 3], especially with regards to the study design, population and endpoint.

In parallel with the inconsistent recommendations of the regulatory agencies on methodology for the diagnosis of HAP/VAP, we found great variability in the clinical criteria used for clinical diagnosis of HAP/VAP. These discrepancies were probably increased by the co-enrollment of patients with HAP and VAP in the majority of studies. Indeed, in agreement with the FDA, but not with the EMA guidelines, no distinction was made between clinical signs and symptoms used for the diagnosis of these different diseases; a minimal duration of mechanical ventilation was required as an inclusion criterion in only 41% of studies. Such a merge of populations with HAP and VAP could mask differences between groups in drug activity, and lead to false conclusions [9]. According to several authors, a combination of clinical and radiological criteria should be used to increase the pretest probability of disease before confirmatory microbiology culture for diagnosis of HAP and VAP [8, 10]. However, in our review, the number of required clinical and biological symptoms greatly varied, and scores combining clinical, biological, and microbiological data such as CPIS [8] were almost never used.

Similarly, in the context of HAP/VAP treatment trials, microbiological confirmation of pneumonia is left optional by the EMA; it was used as inclusion criterion in only 55% of studies included in our review, and among the remaining studies the proportion of microbiologically documented pneumonia varied between 60 and 80%. As a consequence, some patients without pneumonia may have been enrolled in some studies and may have led to false negative results in superiority trials and false conclusions of similarity of drugs in non-inferiority trials. Finally, the severity scores of enrolled patients were very different from one study to another and, in some studies the most severe patients were excluded. The lack of inclusion of some subgroups of patients may be questionable as far as they may specifically receive the maximum benefit of the tested antibiotic [11, 12]. Stratification of randomization of patients based on the severity of illness unfortunately was rarely used.

We identified a large proportion of studies with methodological issues. Although drug dose monitoring or use of multiple antibacterial agents may complicate double-blinding in HAP/VAP studies, it remains crucial to minimize post-randomization bias. In our review, double-blinding was performed in only half of the studies. Furthermore, a pre-specified statistical hypothesis was lacking in more than one study out of four, and the sample size calculation was not mentioned in more than one out of three. Of note, only three studies had superiority design. A statistical hypothesis of non-inferiority was three times more frequent and increased over time (four out of the five most recent studies used it). The use of this design may be the result of FDA and EMA recommendations that describe this design as “acceptable” [2, 3]. Nevertheless, non-inferiority trials are complex to conduct. They require an appropriately selected narrow non-inferiority margin and consequently a large sample size to prevent false conclusions of non-inferiority of the tested drugs. In this context, according to the FDA, mortality non-inferiority margins should be less than 10% [2]. According to the EMA, clinical cure non-inferiority margins should be less than 12.5% [3]. In our study, clinical cure non-inferiority margins were frequently mentioned but they were larger than those recommended in almost all studies, which is clinically unacceptable.

The question of the best primary efficacy endpoint to use in HAP/VAP trials remains highly debatable [13]. Proof of this is provided by the conflicting guidelines provided by the regulatory agencies, with the EMA recommending the clinical outcome at the TOC visit (ranging from 14 to 21 days after the end of therapy) and the FDA recommending all-cause mortality at day 28 [2].

Our results show that the assessment of clinical response (clinical cure) by investigators was used in the vast majority of HAP/VAP RCTs. The main advantage of this criterion is its routine use by clinicians to assess patient response to an antimicrobial treatment. However, as demonstrated by the wide range of clinical cure rates showed in this review, the variable definitions of clinical cure that are used are problematic and may impact the reliability of this endpoint. These results suggest that a consensual definition of clinical cure is urgently needed. Such consensus should also homogenize the timing used to assess the primary endpoint, in order to limit its variability that may also account for variable clinical cure rates. In our review, whatever the endpoint used (i.e., clinical cure or mortality), there was wide variability in the time point used to assess it. Finally, the documentation of the magnitude of treatment effect on clinical cure should also be established. Indeed, a previous demonstration of the superiority of the comparator drug to a placebo or no therapy is necessary before conducting non-inferiority trials [14], and in HAP/VAP such benefit of effective antibacterial therapy has only been established for all-cause mortality [15] and not for clinical cure.

For this reason, mortality is the only endpoint that should be used in non-inferiority trials. However, while mortality is the endpoint that reflects the strongest outcome criteria, its choice may also offer some disadvantages. First, the mortality attributable is difficult to determine; all-cause mortality may be related to underlying comorbidity, affecting the relationship between the efficiency of the antimicrobial treatment and death. Second, low mortality rates reported in HAP/VAP trials (except for those enrolling patients with difficult-to-treat pathogens) and the large non-inferiority margins that are recommended may prevent rejection of the non-inferiority of the tested drug and biased the results.

Nevertheless, the latest FDA recommendations [2] may promote the use of mortality as a primary endpoint. Indeed, among the five currently unpublished RCTs of new systemic molecules that are reported on the ClinicalTrials.gov website, two use 28-day all-cause mortality as the primary endpoint, and one includes death in the definition of clinical cure. Clinical cure was used in seven studies, including one addressing the effect of inhaled antimicrobials as adjunctive therapy. As expected, as a clear consensus is still lacking, the variability in the definition of clinical cure and in the time point used to assess it remain significant.