Assessment of a Noninvasive Exhaled Breath Test for the Diagnosis of Oesophagogastric Cancer

Key Points Question What is the diagnostic accuracy of a breath test for esophagogastric cancer? Findings In a multicenter diagnostic study of 335 patients, including 172 patients with esophagogastric cancer, the breath test demonstrated good diagnostic accuracy. Meaning This study suggests the potential of breath analysis as a noninvasive tool in the diagnosis of esophagogastric cancer.


eMethods 1: VOC breath model refinement
Methods SIFT-MS data from the previous publication by Kumar et al, included 210 patients, 81 with oesophageal or gastric adenocarcinoma and 129 control patients. Given the challenges around separation of phenol and the difficulties associated with transport of phenolic based VOCs, it was considered that in a multi-centre study the analysis of phenolic based VOCs would be unreliable using SIFT-MS, and therefore they were excluded from the diagnostic model. This left ten VOCs that were taken forward to generate a new diagnostic model from this obtained dataset (eTable 1). The concentrations of all VOCs from these data were compared using univariate statistics, Mann-Whitney-U test, across cancer and non-cancer groups. These ten VOCs were taken forward into a multivariable logistic regression model with the dependent variable being the presence of oesophago-gastric cancer. Significant VOCs from this multivariable analysis were defined by statistical significance of P<0.05, and these were taken forward to another multivariable logistic regression model (stepwise regression). Results are presented as odds ratios and 95% confidence intervals. To construct the Receiver Operating Characteristic (ROC) curves, cancer status was used as the dependent variable and the sum concentrations of significant VOCs from the multivariable logistic regression model were used as the independent variable. All statistical analysis was performed using the statistical software SPSS (version 22).

Results
Univariate comparison performed for all VOCs measured in the previous dataset demonstrated significant for 15 VOCs (eFigure 1). All ten VOCs described in etable 2 were significantly dysregulated in the cancer state.

eFigure 1: Graphically illustrating changes in all VOCs between study groups. Positive deflection indicated an upregulation in the cancer group and a negative deflection indicated a downregulation in the cancer group relative to the noncancer group.
The ten VOCs described in etable 1 were then taken forward to a multivariable analysis, with the results showing significant associations between the presence of cancer and five VOCs which were butyric acid, pentanoic acid, hexanoic acid, butanal and decanal (etable 3). Where the full study protocol can be accessed Yes 30 Sources of funding and other support; role of funders Yes

eMethods 3: Optimisation of Bag materials
We conducted an experiment to optimise the bag materials to minimise losses of trace VOCs as part of a multi-centre investigation.

Liquid Calibration Unit (LCU)
The loading of the bags with known amounts of linear aliphatic aldehydes was carried out with the aid of a liquid calibration device (s-LCU from Ionicon Analytik, GmbH -Innsbruck, Austria). The calibration mixture was generated by injecting a mixture of aldehydes (propanal to heptanal, at 1·0-1·7 mg/L in water) at a flow of 50 ml/min into a heated (100°C) chamber. Upon injection, the liquid encountered a gas stream, flowing at 1,000 ml/min; this allowed for rapid evaporation of the analytes, due to the generation of micro-droplets. Knowing the starting concentration of the single aldehydes and supposing the evaporation of the liquid to be instantaneous and quantitative, this should generate C3-C7 linear aliphatic aldehydes in the lowparts-per billion volume (ppbv) range. This assumption was experimentally verified by connecting the LCU device to the SIFT-MS. C3-C6 aldehydes showed a good agreement between expected and measured values, with concentrations in the ±20% range with respect to theoretical values. In the case of heptanal, the measured concentration was repeatable, but considerably lower than the theoretical one. This was probably due to poor evaporation efficiency, also observed for higher boiling point aldehydes, which were evaluated in a preliminary experiment (C8 to C10). The relative humidity of the obtained calibration mixture was 6·2%, and therefore similar to that occurring in breath. The gas stream injection was achieved by means of pressurised gas (synthetic air, BOC gases -Guildford, UK), passed through a scrubber (Supelco -Bellefonte, PA) and connected to the LCU. The calibration mixture was conveyed to the SIFT-MS by means of a short (10 cm) section of PEEK tubing. For the multi-ion monitoring mode, selective VOCs (trace aldehydes) from breath were analysed for a total of 60s and measured concentrations were averaged over this time for each VOC.
Bag materials under investigation were Nalophan (Kalle Ltd, Germany), Tedlar (Sigma Aldrich Ltd., Poole -UK), and Steel (Gastrocheck-Bag-XL-Bedfont Scientific Limited). Bags were stored at room temperature, and were sampled at 0, 24, 48, and 72 hours. Three bags were sampled at each time point with the median and range presented for analysis. Kruksall-Wallis test was utilised to compare the concentration of the trace VOCs at different time points, with a P value of 0.05 taken to indicate statistical significance. Comparison of the three bag types showed variable performance in the ability to retain water and trace aldehydes over the up to 72-hour study period. When stored in Nalophan for 72 hours, there were significant reductions in water (57·1%), propanal (40·4%), butanal (48·7%) and hexanal (55·2%). Tedlar performed well for most aldehydes, however again there were significant reductions in water (47·9%) and heptanal (54·7%). Steel performed well in the retention of most aldehydes with the exception of pentanal, which showed a 73·9% reduction during the 72-hour study period (eTable 5 and eFigure 3).

eFigure 3: Illustrating losses over time of C3 -C7 aldehydes when stored in different bag materials
The results of this study demonstrate that there is loss of trace VOCs from breath bags that impair the interpretation of multi-centre breath studies that involve long periods of sample transport and storage. For the purpose of our investigation steel breath bags appear to have the best performance in reducing loss of trace  P ro p a n a l B u ta n a l P e n ta n a l H e x a n a l H e p a ta n a l

T e d la r T im e (h ) C o n c e n tr a tio n (p p b v )
P ro p a n a l B u ta n a l P e n ta n a l H e x a n a l H e p a ta n a l P ro p a n a l B u ta n a l P e n ta n a l H e x a n a l H e p a ta n a l aldehydes. However, the results of the study do highlight the need for minimising storage time and facilitating early SIFT-MS analysis. Therefore we amended our protocol in response to this study so that all breath samples were stored in steel breath bags and analysed within 8 hours of being taken from the patient.

eMethods 4: Effect of ambient room air upon analysis of trace VOCs
The primary objective of this study was to examine the variation in the levels of traces VOCs from the ambient air in different clinical environments where patients are commonly sampled. The secondary objective of this study was to evaluate the intra-and inter-day variability in the levels of VOCs in these four locations. Room air samples were collected in breath bags using a room air pump. Room air samples were on 5 separate days over a 1-month period in the morning and afternoon from 3 hospital environments (outpatient clinic, endoscopy and theatre waiting rooms) and the laboratory. For each VOC measurement, the syringe plunger was removed from the 1ml Luer lok syringe and the breath bag was directly connected via the syringe barrel to the sample inlet arm of the SIFT-MS instrument. For the multi-ion monitoring mode, selective VOCs from ambient air were analysed for a total of 60s and measured concentrations were averaged over this time for each VOC. The only significant variation in room air VOCs between rooms was seen for butenal, acrolein, butanol, pentanol, butyric acid, putreisceine, methanol33 and isoprene. Importantly there was no significant variation between hospital environments seen in all VOCs included in the oesophago-gastric cancer prediction model previously generated by Kumar et al [8].   Room air from different clinical environments has previously been shown to vary in terms of more abundant VOCs. The present study identifies minimal variation in trace VOCs associated with oesophago-gastric cancer from previous research. However good scientific practice will remain to sample ambient room air at the time of breath sampling to ensure, that exogenous contribution to the patient breath profile is minimal. Regular ambient room air sampling was therefore included as part of the protocol for all clinical samples taken as part of this research.

eMethods 5: Human factor analysis of breath bag sampling
Previous breath research has most commonly involved one or two well-trained researchers taking breath samples from individual patients. Single centre breath studies are of value in establishing pilot research findings, however require validation in larger scale multi-centre studies in order to demonstrate reproducibility of findings. This present study sought to utilise human factor analysis to identify potential sources of error in the breath sampling and analysis process that may lead to errors in sample study and spurious results. Clinicians and researchers undertaking breath sampling were directly observed or videoed during the first three times they performed breath sampling from patients using the 500mL Steel breath bag (Gastrocheck-Bag-XL-Bedfont Scientific Limited). Human factors and Ergonomic (HFE) analysis was employed to identify potential errors and the consequences of these errors associated with the breath sampling technique. HFE is a multidisciplinary science in which human behavior, capacities, and engineering principles are used to explore why errors occur, and how to reduce the likelihood or preventable harm to individuals, with the specific aim to support human performance and safety. The observation of 3 clinicians and 2 researchers during the first 3 episodes of breath sampling identified 10 tasks with associated errors and consequences associated with breath bag sampling. From this, a task analysis was developed ( Table S1) that allows assessment of researchers before permitting them to enroll patients in multicentre breath studies. This task analysis was taken forward and used in practice as part of the multi-centre trial to ensure all researchers were adequately trained to take breath samples, and reduced any previously demonstrated variability in performance of breath sampling.
Correlation plot demonstrating a good correlation between methyl phenol measured on NO+ and H3O+, up a value of log 0 or 1ppbv.  A previous study [1] depicts that volatile constituents consisting hydrophilic -OH or -COOH functional group tend to adhere to the non-polar column through Van der Waals forces and to each other, resulting lower vapour pressure and undesired chromatograph behaviour such as peak tailing. Non-polar column phase such as ZB-624 column exhibits less suitable for analysis of polar constituents. For a successful GC determination of underivatised free fatty acids, application of polar FFAP column phase would be preferred [2]. Ethylphenol is a phenolic compound produced in wine and beer by the spoilage yeast Dekkera bruxellensis [4] but its presence in breath is yet reported.

Methods
Decision conferencing is a series of workshops attended by key players who are facilitated by an impartial specialist in decision theory and group processes, to resolve important issues of concern to the participants. The purposes of decision conferencing are to achieve in the group of key players a shared understanding of the issues, to create a sense of common purpose despite difference of opinion, and to achieve commitment to effective policy and to best practice guidelines. Decision conferencing has been developed over the past 28 years at the London School of Economics by Professor Larry Phillips and his colleagues, and is now used world-wide by hundreds of organisations in all sectors. It is an effective way to tackle difficult problems quickly and thoroughly, and it produces outputs that are readily understood by others because the process of arriving at recommendations is totally transparent.
In applying decision conferencing to this project, we did convene a panel of chairs of multidisciplinary oesophago-gastric teams, gastroenterologists, general practitioners, surgeons and patient members from the oesophageal patient association. The meeting explored the issues, informed participants regarding diagnostic accuracy and cost-effectiveness, and identified factors that affect uncertainty about the location of the test in patient pathway, discussed possible consequences and the key attributes of those consequences, and came to a decision.

Results
On 22 January 2015, 18 people (including five patients) gathered at the Royal College of Surgeons to provide guidance for the developers of a breath test for oesophagogastric cancer that will ensure adequate patient uptake and provide early diagnosis.
Following introductions around the table, data were presented about oesophagogastric cancer, noting that the incidence of oesophageal adenocarcinoma in the UK is the highest in the world. It was explained that patients in the UK take too long to present to their GP, partly because symptoms of heartburn and indigestion are not widely recognised as potentially indicative of oesophago-gastric cancer. Endoscopy is the gold standard for diagnosis, but it is costly and invasive and experienced as a very unpleasant investigation by many patients.
Furthermore it was explained the status of the breath test approach, with data suggesting that the risk prediction model has a good sensitivity and specificity. We stated that the goals for implementing the breath test are, for patients, earlier diagnosis at an early stage of the disease, better survival, and more patient satisfaction, and, for the NHS, tailored referral and lower cost.
A demonstration of the breath test, using the currently-available technology and software, followed.
Questions were asked about the current status of the breath test, which led to an extended discussion and further questions and answers, as follows: • Greater general awareness of the disease is needed, and will be stimulated by a programme in the last week of January 2015. If the result is more endoscopies, than that would help to support need for the breath test.
workshop participants did also suggest that there is a portion of patients who fail to seek medical attention for long-standing symptoms of heartburn, and would typically self-medicate with medications purchased over the counter in a pharmacy.
Therefore a secondary position for the breath test may be in the pharmacy in order to reach this population of patients who typically would not seek medical attention.
A further important finding from this workshop was that the majority of participants were unsure as to the optimal patient interface for any breath test in the future. This is clearly needed to plan in the next stage of the breath test development as a robust breath test in clinical practice is unlikely to gain widespread dissemination using breath bags for sample collection. Sensor based technology may allow for the development of disease-specific hand-held devices capable of utilisation in the primary care setting. However participants in the workshop felt that delivery of the results of the test would be critical to reduce the psychological stress upon the patient and therefore a sensor based hand-held device would need to be performed by a trained medical professional capable of delivering the results of the test in a balanced manner. The alternative would be to utilise thermal desorption tubes for breath sample storage, and transport for analysis to a central laboratory [289]. A possible advantage of thermal desorption tubes would be the potential to allow breath profiling of a number of diseases and therefore the test could be more costeffective and utilised for multiple purposes.

Conclusions
• The ideal position of the breath test in the diagnostic pathway would initially be in the GP surgery and community pharmacy as a secondary location.
• The exact device for breath sampling as a hand-held device to allow for point-ofcare testing or thermal desorption tubes to allow for laboratory testing, remains undetermined by the present workshop.