The platform
NexaMed — precision medicine for inflammatory bowel disease.
NexaMed takes a sequencing laboratory’s VCF for a patient with known or suspected IBD, interprets it against a curated gene panel and published pharmacogenomic evidence, and produces a report that a named clinician reviews and signs.
It is deployed and running as a multi-tenant hospital system.
Regulatory notice
Under development toward CDSCO Class C. Not an approved or certified medical device.
Intended use
NexaMed is built for hospital gastroenterology and clinical genetics teams managing patients with inflammatory bowel disease — particularly very-early-onset and infantile-onset presentations, where a monogenic cause is materially more likely and changes management.
That patient definition is not ours. It follows the same clinical criteria that NHS England commissions its monogenic IBD test under, and those criteria are close to the ESPGHAN/NASPGHAN consensus statements. Two independent bodies describe the same patient, and we describe that patient too.
This is a statement about which patient the test is for, not a claim of equivalence to any NHS test. Theirs is whole-genome sequencing across a far larger gene set; ours is a targeted panel. Anyone who reads it the other way has been misled, so we say it plainly here.
The panel
- 80 targets in total, of which 64 are monogenic IBD genes, plus the HLA-DQA1*05 pharmacogenetic marker.
- 62 of the 64 monogenic genes are GREEN on NHS Genomic Medicine Service signed-off panels.
- 56 of the 64 are on the ESPGHAN/NASPGHAN international consensus list of 75 monogenic IBD genes.
- Genes are added or removed on clinical instruction only. A retired gene is retired, never deleted — reports already signed name it in their examined-genes snapshot, and that snapshot must keep resolving.
Capabilities
Each card carries its real state. One of the three is not built, and it says so here rather than in a footnote.
Monogenic screening
Built and deployed — clinical sign-off in progress
Variants in the panel genes are classified against published evidence and graded by tier. Findings are grouped by gene and carry the gene’s mode of inheritance, so a single finding in a recessive gene reads as what it is rather than as a diagnosis.
Pharmacogenomics
Built and deployed — clinical sign-off in progress
Thiopurine metabolism and biologic-response markers, called with PharmCAT v3.2.0. The container is pinned by digest rather than by tag, because allele definitions change upstream and the same VCF must not yield a different diplotype between two runs with nothing in our code having changed.
Endoscopy scoring
In development — the scoring model is not built
Capture, review, provenance and the clinician’s own scoring workflow are wired end to end. The image classifier is not trained and is not in use. We are working with clinicians on labelled data first, and until that exists the feature produces no score.
Reporting
Most of the engineering in this product is not in producing findings. It is in making sure the absence of a finding cannot be read as reassurance. These are the rules the report follows:
The report names the genes that were examined, taken from a snapshot frozen at the time of the run — not from today's configuration, which may have moved since.
A gene whose target regions were not fully covered is reported, and flagged with the risk of a false negative. It is never quietly omitted.
A position that was not assayed is a no-call. It is never filled in with the reference allele, because that would turn “we did not look at TPMT” into “TPMT is normal” for a patient about to be dosed a thiopurine.
An undetermined genotype produces an explicit finding stating that carrier status is unresolved, and saying in terms that this is not a negative result.
Where a finding rests on a computational prediction, the report prints the predictor’s own statement about clinical validation alongside it, rather than presenting the score bare.
Boundaries
Defining the scope boundary. It is easier to rule this out here than after a procurement review.
It does not diagnose, and it does not recommend or select a treatment.
It states a metabolizer phenotype and never a drug dose. That boundary is enforced in code and covered by a test that fails the build if it is crossed.
It does not assess splice-affecting variants. The tools that do this well are not licensable for commercial use, and we would rather say so than imply coverage we do not have.
It does not call variants. We interpret a VCF produced by an accredited sequencing laboratory; calling is the laboratory's work, not ours.
It produces nothing a clinician has not signed. An unsigned report is not a report.
Next
Security & data protection
Where patient data sits, who can reach it, what is recorded when they do, and what we deliberately do not claim.
Read it →Regulatory status
What is in place, what is not, and what we are not permitted to call this product today.
Read it →Under development toward CDSCO Class C. Not an approved or certified medical device.