Prevention of ventilator-associated pneumonia, mortality and all intensive care unit acquired infections by topically applied antimicrobial or antiseptic agents: a meta-analysis of randomized controlled trials in intensive care units

A comprehensive systematic bibliographic search of medical literature published until June, 2010 was conducted to identify RCTs that assessed the effect of any type or combination of topical antibiotics or antiseptics on the prevention of pneumonia, all ICU-acquired infections and mortality in adults requiring mechanical ventilation in ICU.

The U.S. National Library of Medicine (MEDLINE), Embase and The Cochrane Library computerized bibliographic databases were used. In addition, we checked the references lists from all retrieved studies and meta-analyses or systematic reviews already published, to ensure that all studies could be identified. The following key words in different combinations were used: "aerosolized antibiotics", "airway colonization", "antibiotics", "antimicrobial prophylaxis", "antiseptic decontamination", "chlorhexidine", "critical care", "digestive decontamination", "healthcare-associated infections", "infection control", "intensive care units", "lower respiratory tract", "mechanical ventilation", "meta-analysis", "mortality", "nosocomial infection", "oropharyngeal decontamination", "pneumonia", "povidone-iodine", "prevention", "randomized controlled trials", "respiratory infection", "selective decontamination", "topical", "ventilator-associated pneumonia".

Articles that met the following criteria were included: (a) RCTs; (b) primary studies, not re-analyses or reviews; (c) examining the reduction of VAP and/or mortality and/or all ICU-acquired infections in mechanical ventilated ICU patients by prophylactic use of one or more of following topical treatments: 1) oropharyngeal decontamination using antiseptics or antibiotics, 2) gastrointestinal tract decontamination using antibiotics, 3) oropharyngeal plus gastrointestinal tract decontamination using antibiotics, 4) respiratory tract decontamination using antibiotics; (d) reported enough data to estimate the odds ratio (OR) or risk ratio (RR) and their variance; (e) English language; (f) published through June 2010. Trials that used systemic antibiotic prophylaxis were excluded.

Two of the authors independently reviewed the studies included in the meta-analysis to appraise the quality of the individual trial using criteria developed for Study Protocol and Data Analysis and Presentation by Chalmers et al. [39] and the method of Jadad et al. [40]. The Chalmers et al. scale assigns a weighting factor to each item according to whether it has been addressed completely (full score), partially (half score) or not at all (no score). If an item in the protocol was not applicable, the number of possible points was reduced. The final score of each paper was calculated as the total points scored divided by the total number of points thought applicable to that study, and two sub-scores regarding quality of Study Protocol and Data Analysis and Presentation were also calculated yielding a range from 0 to a full score of 1. The Jadad score, ranging from 0 to 5 points, was assigned to the included trials according to whether the investigators described the study as randomized and double-blind, reported the methods used to randomly assign patients and blind the intervention, and reported the number of withdrawals and dropouts and the reasons.

The readers discussed their evaluation and any disagreements were resolved through discussion and re-reading.

The following items were collected from each clinical trial selected: a) study characteristics (authors, year of publication, design, length of follow-up); b) patients' characteristics (sample size in intervention and control groups, duration of mechanical ventilation, disease severity); c) VAP definitions and the incidence of VAP in treatment and control groups; d) all ICU-acquired infections' definitions and the incidence in treatment and control groups (we considered the number of patients who have developed at least one nosocomial infection including VAP); e) mortality in the ICU and/or in-hospital; f) type, concentration and mode of delivery of antimicrobial/antiseptics used, and control therapy.

Two independent reviewers extracted relevant trial characteristics and interobserver agreements were checked using the unweighted-kappa score [41]; differences between reviewers' data were resolved by discussion until a consensus was reached.

The pooled effects estimates were used to combine the values from the single studies and were expressed as RR and the related 95% confidence intervals (CI). RR and CI were obtained using the Mantel-Haenszel fixed effects model [42], if the studies were homogeneous, and the DerSimonian and Laird random effect model [43] in cases with heterogeneity. Statistical heterogeneity was assessed using Cochran Q and I² measure; a I² value above 25% may be considered low heterogeneity, a value above 50% and 75% were predefined as moderate and high heterogeneity [44, 45]. We preferred applying random effects models' results in case of I² equal or higher than 50%.

The trials included in the meta-analysis differed considerably in several factors such as study design (method of randomization, blinding technique, modes of patients recruitment), clinical heterogeneity of patients (characteristics of participants, baseline disease severity), details of intervention (type and mode of administration of drugs, duration of treatment), and follow-up period, and subgroup analyses were used to explore eventual heterogeneity. We performed separate sensitivity analyses by grouping studies that had similar characteristics, such as patient population (medical or surgical or trauma or mixed critically ill patients); topical selective decontamination of the digestive tract or of the respiratory tract; associations of antimicrobial agents; antiseptic decontamination only with chlorhexidine; using placebo as comparator agent; double-blind studies. At last, we performed a meta-analysis to determine the potential impact of the quality of the studies on the results, by pooling only studies with Jadad scores greater than or equal to the median.

Finally, publication bias was explored by Egger's test [46, 47] and Begg's rank correlation test [48]. All statistical analyses were performed using Stata software, version 10 (Stata Corporation 4905 Lakeway Drive, College Station, Texas 77845 USA) [49]. The reporting of the study's findings was in accordance with the PRISMA statement [50].

A total of 333 publications were identified as potentially eligible for inclusion. Of these, 28 articles met all inclusion criteria and were included in the meta-analysis. A flow diagram providing the reasons for excluding the articles from the meta-analysis is reported in Figure 1. The agreement between the two researchers in the first comparison was 89%, with a kappa score of 0.88, and after discussion and detailed review of the articles was complete (k = 1).

Tables 1 and 2 summarize information on the patients and design of the included trials [32‐38, 51‐71]. All trials assessed mortality as an outcome, 25 VAP [32‐34, 36, 37, 51‐63, 65‐71] and 13 all ICU-acquired infections [37, 38, 51, 52, 54, 56, 58, 60, 64‐68].

The mean quality scores of the individual studies using the Chalmers et al. scale was extremely variable, ranging from 0.19 to 0.9 (mean = 0.64), for the Protocol from 0.25 to 0.97 (mean = 0.61) and for the Data Analysis and Presentation from 0.13 to 0.88 (mean = 0.58). It should be noted that none of the studies had a full score, both for Protocol and Data Analysis and Presentation. Almost all trials received full credit for description of daily amount and timing of therapeutic regimens (93%), presentation of test statistic and P-value (93%) and number of patients who withdrew and the reasons why (89%), most trials reported description of relevant variables in experimental and control group (86%), criteria for patient selection (82%), methods for assuring masking of randomization and to evaluate success of masking (75%) and again start and stop dates (71%) and analysis of comparability of the study groups (61%). A few studies reported criteria for stopping the trial (39%), CI (36%) and only 18% and 21% of the studies discussed beta error and side effects of treatment, respectively (Table 3). With regard to the Jadad et al. criteria, the mean score was 3.39 (median 4), most trials addressed adequately the problems of withdrawals or dropouts after randomization (74%) and were classified as double-blinded (63%). Only eight trials [33, 36, 56, 57, 64, 66, 68, 71] had a full score on the Jadad et al. scale. It should be noted that there has been improvement in study design and reporting, since findings published more recently tended to receive a higher quality rating.

Results of the meta-analyses that explored the effects of topical use of antiseptics and antibiotics on the prevention of VAP, mortality and all ICU-acquired infections are shown in Table 4.

The major finding of the present meta-analysis provides support to the observation that topical antiseptic or antibiotic SDRD plays a significant role as a protective factor against the development of VAP.

Chlorhexidine has been used as a degerming agent in all but one antiseptic trial, because it has a high level of antibacterial, antiviral and antifungal activity, it is virtually free of adverse effects [72] and it is an inexpensive solution. Seguin et al. used povidone-iodine, another antiseptic agent that has high, rapid and persistent activity on Gram-negative and Gram-positive bacteria [73, 74], and it is simple and safe to use [75]. Our results do not allow us to determine which of the two antiseptics has a greater effect, but confirmed the preventive role of antiseptics against VAP. To find the best disinfectant further studies are needed comparing povidone-iodine and chlorhexidine in preventing VAP in "head to head" trials.

The findings of this pooled analysis are consistent with the relatively strong protective association observed between antiseptic oropharyngeal decontamination and the risk of VAP observed in previous meta-analyses [25‐28].

Since the pathogenesis of VAP is related to contamination of the aero-digestive tract, our meta-analyses on antibiotics were conducted combining data on topical digestive and respiratory tract decontamination and our findings showed a significant protective effect of the antibiotics. This result persisted when we considered separately the digestive and the respiratory tract decontamination, suggesting that any of these two ways of decontamination may be used. Only two studies [57, 70] classified VAP according to time of onset in early and late-onset [76]; therefore, we have not performed a sensitivity analysis on this issue.

Comparisons with the results of previous meta-analyses are very difficult since the inclusion criteria used were different. Indeed, Liberati et al. [31] did not make any restriction on the type of respiratory tract infections (RTIs), combining trials on pneumonia and tracheobronchitis in ICU patients, nor on unpublished RCT or on language, whereas they restricted inclusion to digestive decontamination by antibiotics only. The results of Liberati et al. showed a significant protective effect on RTIs.

Chan et al. [27] did not reach our conclusions since their results suggested no significant protective effect of topical antibiotic prophylaxis on VAP. Indeed, the meta-analysis of Chan et al. had only two studies [68, 71] in common with ours, and pooled 1,098 patients, whereas the present meta-analysis could take into account a larger sample size (2,463 patients). Finally, Falagas et al. [30], who took into account only the effect of administration of antimicrobial agents via the respiratory tract, concluded that pneumonia occurred significantly less often in the prophylaxis arm compared to the comparison arm. Our sub-analysis on trials that tested respiratory tract decontamination confirms these results, even if the sample analyzed is different because we included only pneumonia occurring in patients assisted by mechanical ventilation; therefore, we excluded the trials of Greenfield et al. [77] because data on colonization only was reported and of Klick et al. [78] for incomplete data on outcomes. Moreover, we excluded non-randomized trials [79] and abstracts [80], and included one study in which the preventive strategy was topical decontamination in the subglottic area [70], and a recent trial that tested prophylactic administration of aerosolized ceftazidime [37].

In the present meta-analysis neither antibiotic nor antiseptic topical decontamination influenced overall mortality, and the results did not substantially differ in the separate meta-analyses performed according to different quality and study design or details of intervention. Liberati et al., in previously published meta-analysis, examining the prophylactic use of antibiotics, have shown efficacy in reducing mortality when topical decontamination was combined with intravenous antibiotic administration, whereas, in line with our results, topical prophylaxis alone was not effective. No former meta-analyses on antiseptics or antibiotics [27‐29, 31] showed a significant beneficial effect on mortality.

These results could be due to lack of an effect of administration of topical agents on mortality since in only one trial that used antiseptic decontamination [51], and in no antibiotic trials [37, 38, 57‐71], was a significant reduction in mortality found. Another possible explanation was that in most of the trials mortality was a secondary outcome and data accuracy could be lower than primary outcome; therefore, analysis combining these data could have failed to show an effect of experimental treatment. Also, in our case the overall sample size was small and, therefore, could have limited the interpretation of the effect on mortality. Another possible explanation can be related to our inability to distinguish the role of the topical SDRD on the occurrence of early and late-onset VAP. It has been demonstrated that VAP related mortality is restricted mainly to patients with late-onset VAP; it is only marginally reduced with appropriate empirical antibiotic treatment [14]. In the same study the authors report that the prophylactic regimen they used, that is, the universal use of continuous aspiration of subglottic secretions, was recognized to be a reducing and delaying factor for VAP. Therefore, we may hypotheses that if the topical SDRD is also particularly effective only on the reduction of early-onset VAP, its role on late-onset VAP related mortality may be marginal. Trials allowing data extraction on the occurrence of early and late-onset VAP separately are strongly needed.

To the best of our knowledge, this is the first meta-analysis aimed at the assessment of the effectiveness of topical SDRD on the incidence of all ICU-acquired infections.

Our results have shown the effectiveness of topical SDRD both with antiseptics, although only in the sensitivity analyses restricted to high quality studies and to those involving only specialty surgery ICU, and antibiotics.

These achievements confirmed what we expected because colonization is a prerequisite for the development of infections that frequently arises from the endogenous flora in the oropharyngeal and intestinal tract; therefore, healthcare-associated infections were potentially preventable through the suppression of colonization of the digestive and/or the respiratory tract. Also, nosocomial pneumonia is the second most common healthcare-associated infections (HAI) and the most frequently acquired infection in the ICU, so all ICU-acquired infections could be reduced secondarily from a VAP prevention.

Considering the importance of antibiotic-resistance in ICUs, we focused our interest on topical administration since it has been reported that it is more frequent when a combination of topical plus systemic antibiotics was used [81, 82] rather than only topical antibiotics [38, 83]. Anyway, the biggest criticism even against topical SDRD is the emergence of resistant strains. This issue has been only marginally investigated in most of the studies included in the meta-analysis. The majority of the trials using antibiotics [37, 38, 57, 59, 61, 62, 64, 68, 69] found no increase in infections caused by antibiotic-resistant organisms, at least during the relatively short period of studies, whereas a trend towards increased colonization of patients by resistant microbial strains was reported [58, 66, 67]. This increase was not found in studies using antiseptics [37, 51, 52, 57]. On the basis of these results, a long-term increase in the occurrence of HAIs sustained by resistant strains as a result of topical SDRD with antibiotics cannot be excluded, and this issue warrants cautious attention in further studies. However, it should be pointed out that routine five-year use of selective digestive decontamination was not associated with increased antimicrobial resistance rates [83]. This finding allows us to suggest an analogous result with the use of long-term topical SDRD at least in ICU with low baseline resistance rates.

The strengths of the present meta-analysis include the considerable number of studies and subjects included as well as the acceptable methodologic quality of the studies on which the analysis is based. It is well-known that the quality assessment of the primary studies has been identified as one of the most important steps of the peer-review process [84]. This comes from the consideration that studies of poor quality may yield information that is not valid; therefore, the inclusion of studies with invalid information in a meta-analysis can make the conclusion of the meta-analysis invalid. Therefore, taking the quality of studies into account in a meta-analysis has the potential to enhance the validity of a meta-analysis because quality is implicitly a measure of validity [85]. Moreover, the evaluation of the quality of the studies in a meta-analysis may contribute to point out limitations in published studies and suggest ways to improve the methodology of studies in further research. In this meta-analysis the quality of the RCTs was good with regard to the various methodological aspects of the research protocol (for example, description of therapeutic regimen, criteria for patient selection, randomization, blinding) while the main shortcomings were related to the overall assessment of the statistical methods and presentation of data. Limitations of the present meta-analysis study include the heterogeneity between the studies with respect to patient populations that had a different profile of risk factors; different medications used, particularly in antibiotic trials, in the choice of the antimicrobial agents or their associations or the delivery mode; different approaches for the control arms and different outcome definitions. Moreover, among our inclusion criteria there was restriction to studies published in English. This is a controversial issue, since some argue that authors are more likely to report positive results in international journals and negative results in local journals, as demonstrated by Egger et al. [86]. Language restriction could, therefore, introduce bias in the results of the meta-analysis [87, 88]. However, there has also been evidence that studies published in local journals may be of lower methodological quality, as reported by Sterne et al. [89], and this would be in favor of their exclusion. We have also performed a search without language restriction and it would have led to the inclusion of two more papers [90, 91]. Vogel et al. [90] was excluded because it was not a RCT, while we have repeated the meta-analyses of topical SDRD by antibiotics on VAP prevention and on mortality by entering Rathgeber et al. [91] and the results did not alter our conclusions. Indeed, the overall estimate of efficacy of antibiotics on VAP prevention was 37% (95% CI of efficacy = 20% to 51%) and the meta-analysis on mortality showed no significant effect (efficacy = -2%; 95% CI of efficacy = -12% to 7%). These results were confirmed in the sensitivity analyses involving only upper respiratory tract decontamination with antibiotics (data not shown). Therefore, we believe that our strategy is not prone to substantial bias or to low robustness in the overall results. Finally, the findings are affected by the limitations of the individual trials included. According to the mentioned limitations, the results of the meta-analysis must be interpreted with caution; however, the careful examination of possible sources of heterogeneity contributed to assess the methodologic quality of research, and to identify potential biases, data gaps, and suggestions for future research.