Breath VOC diagnostics · Upper-GI cancer
The signal exists. It just doesn't travel.
A five-compound breath signature for oesophago-gastric cancer has been published and externally validated across three sites. Decades of work like it, and still no test on a ward. The failure has never really been sensitivity. It's that a model built at one site falls over at the next. I'm building the harmonisation layer that lets breath VOC data pool across sites and instruments — so the signal survives the journey.
The wall
Decades of positive studies, and no product
Breath carries real chemical information about disease. Dogs find it, instruments find it, papers report it. What no one has managed is a test that holds up once it leaves the lab it was built in. No two studies collect breath the same way, so the models don't pool and they don't reproduce. That's the wall. It's a data problem wearing a chemistry problem's clothes.
The approach
Fix the data before the device
My bet is that the poor generalisation is mostly protocol and batch effect, not irreducible biology. So the first thing I'm building isn't a sensor. It's a sampling protocol and a differential measurement design that make multi-site, multi-instrument breath data poolable — with a within-sample baseline that cancels the ambient noise most studies never controlled for. The classifier follows the data. The device follows the classifier. In that order, deliberately.
First indication
Why upper-GI, and why now
Oesophago-gastric cancer is the one indication with a named, externally validated breath panel behind it: five compounds, three London sites, published in JAMA Oncology. The health economics are already modelled — a breath-assisted referral pathway comes out roughly £155M a year cheaper than the current one. And the signal looks anatomically local, strongest in air from the upper-GI tract rather than diluted across the whole body. Better odds than most, and someone else has already published the economics.
The honest ledger
What's established, and what's open
Established
- A five-compound OGC breath panel, externally validated across three sites. JAMA Oncology, 2018
- The compounds are named and published — not a black-box fingerprint.
- A breath-assisted referral pathway modelled at roughly £155M a year saved.
- The VOC signal is reported strongest in upper-GI endoluminal air — the source is local.
Open
- Whether the panel holds at early stage (T1/T2). The validation cohort skewed to advanced disease — this is the largest scientific risk, and I won't claim early detection until stage-stratified data says so.
- Whether the signal survives a pooled, harmonised, multi-site design. That's precisely what I'm building to test.
Who's behind this
The operator half of the problem
I'm Raghu Ram Nair, a solo operator, and I'll name the edges plainly. I'm not the analytical chemist, and I don't yet hold a dataset at scale. What I bring is the other half: a diagnostic taken through CE marking, DCB0129 clinical safety and into NHS use, product work at Genomics England and at national-health-system scale, and three years spent judging whether other people's clinical claims actually held. The science here isn't short of studies. It's short of someone who can carry it into a product.
What I'm looking for
Two people, in this order
Analytical chemist
A breath-sampling and measurement partner. GC-MS, PTR-MS, thermal desorption, and the craft of holding measurement steady across sites. This is the capability I most need beside me.
Clinical PI
A principal investigator with cohort access, particularly upper-GI or inflammatory disease, who'd want to test whether a harmonised protocol travels where single-site models keep failing.
If that's your world, or you know whose it is, I'd like to talk. A pointer to the right person is worth as much as a yes.
raghuram@symplifye.health