The early diagnosis of upper digestive tract disease (EDIGEST) study was sponsored by Imperial College London and opened to recruitment in September 2019. The study aimed to investigate whether substances found in a person’s biological samples (breath, urine, blood, etc) could be used to distinguish between people with mouth cancer (oral cancer) or cancer of the food pipe (oesophageal cancer) from people without cancer.
The planned sample size was 300 participants: 100 with oral cancer, 100 with oesophageal cancer, and 100 without cancer (controls). Between March 2020 and March 2022, recruitment was substantially affected by the COVID-19 pandemic and the associated disruption to clinical services and research activity. As a result of these circumstances, the study was modified to focus on the collection of breath samples only.
A total of 328 participants contributed to the scientific objectives of the eDIGEST study. Of these, 154 participants (oral cancer = 51, oesophageal cancer = 44 and controls = 59) were directly recruited under the eDIGEST study ethics, 145 participants (oesophageal cancer = 44; controls = 101) were recruited through a parallel CRUK-funded study (CPMS 50774) to minimize duplication of recruitment and research activities, and 29 participants with oesophageal cancer were recruited through an international collaboration with the Chinese University of Hong Kong.
Between September 2024-September 2025, an interim analysis of the breath samples was undertaken. The purpose of this analysis was to determine whether the samples collected to date provided sufficient evidence of proof-of-concept to support an application for funding for large-scale studies in oesophageal and oral cancer respectively.
Breath samples were analysed using a laboratory technique called gas chromatography-mass spectrometry (GC-MS). This technique allows researchers to detect very small molecules present in breath samples, and these molecules are known as volatile organic compounds (VOCs). VOCs are produced through normal processes taking place throughout the body and can eventually be released in exhaled breath. Cancer may change the amounts or patterns of some of these VOCs. We therefore investigated whether participants with oral or oesophageal cancer had a different overall pattern of VOCs in their breath compared with participants without cancer
All 154 samples collected under the eDIGEST ethics were subjected to quality checks using the in-house standard operating procedures. Sixty-three samples (oral cancer=28; oesophageal cancer=17; non-cancer controls=18) could not be included in the final analysis for the following reasons: (i) sampling errors during the research visit (n=4), (ii) participants subsequently found not to meet the eligibility criteria (n=8), (iii) failure at the stage of laboratory or GC-MS analysis (n=47) and (iv) variation within the breath sample above the acceptable quality threshold (n=4). The remaining 91 samples were analysed.
From these 91 samples, 571 different VOCs were initially detected. Exploratory statistical methods were used to identify patterns between the three groups as well as to investigate whether the breath samples from participants with cancer differed from those of participants without cancer. The initial analysis showed a separation between the cancer and non-cancer samples. It also showed that the breath profiles of participants with oral cancer and oesophageal cancer were similar, with considerable overlap between the two cancer groups. Further statistical analysis was therefore carried out to identify the VOCs that contributed most to the differences between the groups. The ten most informative VOCs were then used to develop diagnostic models to assess how well they distinguish between the groups. The performance of these models was assessed using a measure called the area under the curve (AUC). This provides an indication of how well a test can distinguish between two groups. An AUC of 0.5 would indicate that the test performs no better than chance, while an AUC of 1.0 would represent perfect discrimination between the groups. Four models alongside their AUC are reported here:
Although limited by a small number of participants in each group, these results provide proof of concept that patterns of VOCs in breath differ between participants with oral or oesophageal cancer and participants without cancer. In particular, the promising performance observed for oral cancer (AUC = 0.817) provided a sufficient signal to support further investigation. On this basis, it was considered that recruiting additional oral cancer participants within eDIGEST would provide limited additional value, and recruitment to this group was therefore stopped. Oral and oesophageal cancers can be difficult to detect early because symptoms may be vague and similar to those caused by non-cancer conditions. GPs currently do not have a simple triage test to help identify which patients may be at a greater risk and require urgent investigation. The eDIGEST study provided early evidence that a breath test could potentially help address this gap, supporting progression to larger studies in oral and oesophageal cancer. The study also supported the successful completion of a PhD and helped secure £530,000 in National Institute for Health and Care Research (NIHR) funding to evaluate the oesophageal cancer breath test in a multicentre setting.