Two patients on the same ward, carrying the same organism at the same time, have not necessarily infected each other. That distinction changes what an Infection Prevention and Control (IPC) team should do next, and getting it wrong carries a cost either way.
NEX Health Intelligence
Tim works with Avi and Ash on the models that identify possible transmission within a hospital. He puts the problem plainly.
"You can have two people who've been in contact with each other who both have similar symptoms. They would have the same MDRO, but that doesn't actually mean it's been transmitted from one person to another."
An MDRO is a multi-drug resistant organism. Shared shared time on a ward can point to transmission. They can also reflect a coincidental overlap between two patients who happen to be unwell at the same time in the same place.
Waiting for the Answer Is Itself a Risk
Whole genome sequencing can settle the question. Comparing the number of single nucleotide polymorphisms (SNPs) between two isolates reveals their genetic relatedness, while analysing their shared plasmid content (Pling) adds a secondary layer of evidence that can point to a transmission link. A distance below an agreed threshold indicates little mutation between the two samples, which supports a genuine transmission link.
Sequencing is slow and expensive in routine practice. "We're talking days, sometimes weeks, depending on hospital capacity," Tim says. Many hospitals have no rapid sequencing in-house and send samples away.
Decisions cannot wait that long, so IPC teams act on incomplete information:
Cohorting. If two patients are isolated together and have not in fact transmitted to one another, that decision exposes them to more risk than keeping them apart.
Environment and practice. A contaminated sink area or a lapse in IPC practice on a ward needs time, money and investigation. Committing those resources to a coincidence takes them away from somewhere they were needed.
What We Did
NEX has developed algorithms that identifies likely transmission links between patients from routinely collected hospital data, without waiting on the laboratory.
To test whether those links were real, the team generated predictions with the model, sequenced the relevant patients, and compared the SNP and Pling distances between the isolates. Where that distance fell below the threshold, the predicted link was supported by the genomic evidence, and this held across enough cases to give statistical confidence rather than a handful of examples.
This matters because it tells us something about the alerts themselves. A prediction validated against sequencing is a prediction an IPC team has grounds to act on before the sequencing would otherwise have come back.
"It means NEX is able to predict genomically verified transmissions," Tim says. "I think that makes us one of the only medical devices which is genomically validated.”
What This Does Not Mean
NEX does not diagnose infection or confirm transmission on its own. The model identifies patients and links that may warrant review, and clinical judgement remains with the team.
The validation work described here shows that the model's predictions agree with sequencing results in the cases studied. It does not on its own establish what happens to infection rates when a hospital acts on those predictions. That is a separate question, and one we are continuing to work on.
Why It Matters
Earlier information is only useful if teams can trust it. Checking our predictions against the genome is a small but important step towards giving IPC teams a reason to isolate the right patients, investigate the right ward and leave the rest alone, at the speed those decisions actually have to be made.
For Tim, that is the point of the work. Small gains in model performance, he says, "equate to lives being saved and resources being freed up," with clinical staff "spending less time searching for patients and more time treating them."
