Analysis of exhaled breath to identify critically ill patients with ventilator‐associated pneumonia

Ventilator‐associated pneumonia commonly occurs in critically ill patients. Clinical suspicion results in overuse of antibiotics, which in turn promotes antimicrobial resistance. Detection of volatile organic compounds in the exhaled breath of critically ill patients might allow earlier detection of pneumonia and avoid unnecessary antibiotic prescription. We report a proof of concept study for non‐invasive diagnosis of ventilator‐associated pneumonia in intensive care (the BRAVo study). Mechanically ventilated critically ill patients commenced on antibiotics for clinical suspicion of ventilator‐associated pneumonia were recruited within the first 24 h of treatment. Paired exhaled breath and respiratory tract samples were collected. Exhaled breath was captured on sorbent tubes and then analysed using thermal desorption gas chromatography–mass spectrometry to detect volatile organic compounds. Microbiological culture of a pathogenic bacteria in respiratory tract samples provided confirmation of ventilator‐associated pneumonia. Univariable and multivariable analyses of volatile organic compounds were performed to identify potential biomarkers for a ‘rule‐out’ test. Ninety‐six participants were enrolled in the trial, with exhaled breath available from 92. Of all compounds tested, the four highest performing candidate biomarkers were benzene, cyclohexanone, pentanol and undecanal with area under the receiver operating characteristic curve ranging from 0.67 to 0.77 and negative predictive values from 85% to 88%. Identified volatile organic compounds in the exhaled breath of mechanically ventilated critically ill patients show promise as a useful non‐invasive ‘rule‐out’ test for ventilator‐associated pneumonia.


Introduction
Ventilator-associated pneumonia is the most common cause of nosocomial infection occurring in critically ill patients and is associated with significant morbidity, mortality and healthcare cost [1,2]. Prompt use of broadspectrum antimicrobial drugs is recommended due to the wide range of potential causative organisms [3,4]. However, diagnosis of ventilator-associated pneumonia is complex and pneumonia is confirmed only in approximately onethird of patients suspected of ventilator-associated pneumonia [5]. Overuse of antimicrobial drugs is associated with drug-induced adverse events and drives emergence of antimicrobial resistance [6,7]. Antimicrobial resistance is increasingly identified in critically ill patients with ventilatorassociated pneumonia and is associated with poor clinical outcomes [8].
Antimicrobial stewardship aims to preserve the efficacy of antimicrobial drugs [9]. Improved diagnostic strategies are critical to an effective antimicrobial stewardship programme, allowing antibiotics to be stopped in patients without infection and narrowing the focus of antimicrobial therapy once a causative organism has been identified [10].
Soluble biomarker-led invasive diagnostic approaches to patients with suspected ventilator-associated pneumonia, collected via bronchoscopy, have been shown to outperform scoring systems as a`rule-out´test, but have not led to a reduction in antibiotic use [11].
Measuring volatile organic compounds may offer noninvasive biomarkers that can be used to rule out ventilatorassociated pneumonia without the need for bronchoscopy.
Volatile organic compounds can be measured in human breath and diagnostic utility has previously been demonstrated in patients with asthma and chronic obstructive pulmonary disease [12,13]. We have previously shown that volatile organic compound capture and off-line analysis of exhaled breath of critically ill patients is feasible [14], and further refined and validated the sampling method ex vivo and in vivo [15]. Here, we aim to establish`proof of concept´for volatile organic compound capture and analysis as a potential`rule-out´test for mechanically ventilated critically ill patients with ventilator-associated pneumonia. As secondary aims, we investigated the performance of individual volatile organic compounds and multivariable models in ruling out ventilator-associated pneumonia, and assessed changes in these volatile organic compounds following treatment.

Methods
We conducted a multicentre, prospective, observational,  [5]. A total of 120 ml of saline were instilled, aspirated and pooled during bronchoalveolar lavage.
Non-directed lavage was performed using a total of 20 ml of saline blindly instilled, and aspirated via a suction catheter passed and wedged via the tracheal tube [16]. Tracheal aspiration was achieved by blind aspiration of tracheal contents through a suction catheter passed through, and extending just beyond, the tracheal tube [17].
The analytical method for breath samples has been described by van Oort et al. [15]. Briefly, sorbent tubes were conditioned at 330°C in nitrogen (OFN, BOC Ltd, Woking, UK; 50 ml.min À1 ) using a sorbent tube conditioner (TC-20, Markes International, Bridgend, UK). After breath sample collection, samples were refrigerated at 4°C until analysis (median (range) storage time before analysis was 7 (0-42) days; volatile organic compounds have been shown to be stable within this range previously [18]). Samples were analysed by thermal desorption-gas chromatography-mass spectrometry using a thermal desorber coupled to an Agilent 7010 GC-MS (Agilent, Santa Clara, CA, USA).
TenaxGR tubes are hydrophobic and in most cases were sufficiently dry purged during the pre-purge method in the thermal desorber (1 min at 50 ml.min À1 He flow; TD-100, Markes International, Bridgend, UK). Tubes were weighed before storage to check whether water had condensed within them (this was the case for 17 tubes, which were further dry purged with a counter flow of 50 ml.min À1 nitrogen for 4 min before tube storage). Analytes were initially desorbed at 280°C onto a general purpose focusing trap before a second desorption onto the GC column (DB-   Table S1) Information Table S2).
However, none of the volatile organic compounds demonstrated a differential change over time between ventilator-associated pneumonia and non-ventilatorassociated pneumonia using a formal test of interaction.
There was considerable inter-individual variation in volatile organic compound concentrations which persist over time with intra-class correlation coefficients for individual effects between 0.68 and 0.92.
Both the Lasso and the Ridge regression identified models with modest predictive ability (AUROC 0.79 and 0.81, respectively; the Lasso is shown in Fig. 2). The negative predictive value of the two models was 0.88 for the Lasso

Discussion
Detection of volatile organic compounds in the exhaled breath of mechanically ventilated patients offers a novel, non-invasive approach to assist in managing critically ill patients with suspected ventilator-associated pneumonia [15]. Capture and analysis of volatile organic compounds appears safe. This study provides promising preliminary data to support ongoing development of a potential`rule-out´test in the management of critically ill patients.
Our approach focused on confirmation of previously  (Table 3) were alkanes or aldehydes, which are potentially fatty acid breakdown products and markers of oxidative stress [24].  Figure S1 or the volatile organic compounds listed in Table 3. This significant decrease in biomarker levels with ventilator-associated pneumonia is however consistent with findings from previous studies, e.g. dodecane and tetrahydrofuran [25], methyl isobutyl ketone [26] and acetone [25,26]. In other  cases, the direction of volatile organic compound biomarkers with ventilator-associated pneumonia has been mixed, e.g. heptane [14,26]. This observed duality may reflect the complex evolution of volatile organic compounds with the course of infection and requires further investigation in mechanistic studies. Whatever mechanisms are being suppressed due to the onset of ventilatorassociated pneumonia, there is evidence they are returning to their pre-ventilator-associated pneumonia state following treatment, as can be seen in Figure 2.
Less than half (46%) of the significant volatile organic compounds reported in Table 3   Overall, the study results demonstrate that a model can be developed utilising volatile organic compounds identified in exhaled breath of mechanically ventilated critically ill patients which could`rule out´ventilatorassociated pneumonia with clinically acceptable performance [11]. The`rule-out´test would suggest avoiding or stopping antibiotics in approximately half the patients without ventilator-associated pneumonia, thus avoiding unnecessary antibiotic treatment in these patients.

Supporting Information
Additional supporting information may be found online via the journal website. Figure S1. Correlations between the various organic compound abundances represented in a heatmap. Table S1. Previously reported sources of volatile organic compounds included in the analysis.