Testing Technology Trends
COVID-19 primarily infects its victims via expired breath – so why are we not using breath to detect it?
One hundred-fifty years ago, Francis Anstie, an English doctor and first editor of the medical journal The Practitioner, documented that expired breath contained significant and measurable quantities of alcohol —probably not a surprise to his peers who visited a pub with him. Seventy years later, in 1938, a portable Drunkometer was introduced, relying on a color change when alcohol and acidified potassium permanganate were combined. By the late 1960s the more politically correctly named Breathalyzer entered the electronic age, opening up widespread use by traffic police to detect incapacitated drivers.
Foreign substances and unneeded metabolites, also known as VOCs, or volatile organic compounds, have to be voided by the body through one or both of two possible pathways—via the liver and kidneys in urine, or by gas exchange in the alveoli of the lungs from the bloodstream to expired air. To date, some 3,000 unique VOCs have been detected in breath, and detecting them is non-invasive, quick and cheap.
Early breath devices depended on finding a convenient chemical reaction in which a compound of interest, when added to all the other required ingredients, would generate a color change. As the tools of chemistry became more sophisticated (e.g. chromatography, spectroscopy, sequencing, immunoassays) detecting a wider range of foreign compounds has become straightforward. Over the past 20 years, the miniaturization of innovative sensors and electronics led to the development of numerous hand-held breath devices. Devices exist for an almost infinite range of compounds found in breathable air, whether ingested environmental pollutants, toxic substances or metabolic byproducts that are subsequently expired in breath. The Human Breathomics DataBase (HBDB) created and maintained by National Taiwan University, lists 913 VOCs found in 60 human diseases that are detectable in expired breath.
Significant but slow progress is being made using breath to diagnose and differentiate among classes of disease, such as molecularly delineated cancers of various organ systems, chronic inflammatory conditions, auto-immune disorders and a few pathogen-invasion diseases. Each of these categories has a relatively unique VOC signature. Unfortunately, however, there has been little progress in using breath within the infectious disease arena—there are only three of the 60 diseases in HBDB that are infectious. All of these are bacterial not viral.
Why are viruses harder to identify in breath than bacteria? Bacteria are fully functioning life forms that have their own unique VOC signature which increases as the infection takes hold. Tuberculosis is an interesting example. Its bacteria emits a well recognized VOC signature (benzene, naphthalene and various alkanes) but in spite of this, the accuracy of breath analysis to correctly identify tuberculosis is still quite limited. One study in the journal Tuberculosis, see below, developed a technique with 98% Negative Predictive Value allowing the ruling-out of a tuberculosis diagnosis, but the more interesting positive result was wholly inadequate (13% Positive Predictive Value). Viruses have no metabolic processes of their own. They hijack the patient’s own cell machinery for this, so there are no unique VOCs to detect, unlike tuberculosis.
Yet dogs seem able to sniff-out COVID-19, and we know infection occurs from inhaling infected expired air, so it seems that it has to be possible to identify COVID-19 this way. The active virus and its fragments are found in microscopic breath droplets, but the problem is their concentration is 1,000 to 10,000 times lower than in saliva. So the question we are left with is this: Is there a way forward to use breath as an accurate enough, fast and cheap diagnostic, when it can be more convenient to focus on saliva for an easy and low-cost screening method?
The most likely opportunity being pursued is to discover a more subtle and complex VOC signature, like a fingerprint or breathprint, involving measurement of hundreds of VOC candidates all at once. Typically trained artificial intelligence (AI) principal component analysis (PCA) is used for this purpose. AI/PCA is a very powerful technique, but the diagnostic result is provided from a computational black-box, an algorithm similar to those used in block-chain transactions, and it is quite hard to disaggregate the contributing factors in any intuitive way. Validation therefore requires observing accuracy in clinical cases. This means very large data sets to rule out bias, but none have yet been published to allow independent evaluation. Several companies and institutions are exploring breath diagnostics this way for diseases, including COVID-19, using proprietary AI fingerprinting e.g. Owlstone (UK), Canary (US), Deep Sensing Algorithms (Finland), Ben Gurion University (Israel). No clinical data is yet publicly available to evaluate possible success.
Twenty-five years ago, perhaps the first clinical application of breath diagnosis was the use of expired nitric oxide, or NO, to evaluate asthma. NO is highly specific for this purpose because it is a signaling molecule that expands blood flow to compensate for the reduced oxygen availability that asthma generates. Many other diseases are amenable to a similar approach, but despite a long history of investigation of expired breath, to date there is no clinical adoption of this technique for viral disease diagnosis. Over the next 12 months, as small trials are initiated by breath pioneers, we may discover reliable fingerprints, but any broad clinical usage will likely be beyond the 2021 forecast of the end of this particular pandemic. So, probably not a good idea to hold our breath for its arrival.
About the author
Mara G. Aspinall co-founded the biomedical diagnostics program at Arizona State University’s College of Health Solutions, which is the first program dedicated entirely to diagnostics as an independent discipline. She is Managing Partner of BlueStone Venture Partners and CEO of the Health Catalysts Group, a research and consulting firm for new healthcare companies.
Saja EL Yaacoub is a bioengineer with a background in genetics. She is currently pursuing a Master of Science in biomedical diagnostics at Arizona State University and is a biocurator in the Limb Girdle Muscular Dystrophy Variant Curation Expert Panel at ClinGen, an initiative to build a central database of clinically relevant genes and variants for use in precision medicine and research funded by the National Institutes of Health.
About the T3 blog
Given the unparalleled threats and uncertainty brought on by COVID-19, sharing information is more important than ever. Here you’ll gain insights, knowledge and strategies to help you navigate the challenges, improve decision-making and take control of your workplace and your life.