An Epidemic Radar Built from Ordinary Doctor's Notes
How DeepSensi turns the exhaust of everyday medicine into a privacy-preserving early-warning system for the next outbreak
Feature · DeepSensi PBC, Dover, Delaware · July 2026 · [email protected]
The most valuable early-warning signal in a pandemic is also the most perishable. In the first days of an outbreak, before the laboratory confirmations, before the case counts, before the name of the pathogen is even known, the disease is already walking into clinics. Doctors are already seeing it, describing it, writing it down. And then, almost everywhere on Earth, that information evaporates into a thousand disconnected records, and the world waits weeks for the laboratories to catch up to what the clinics already saw.
DeepSensi, the verified clinical-intelligence system built by Tomasz Jan Gomoła, was not designed to be an epidemiology tool. It was designed to make a single physician safer on a single difficult case. But in a new technical paper, the company describes a capability that falls out of its architecture almost for free: a syndromic-surveillance layer called the Global Emergency Response Network, or GERN, that turns the ordinary documentation of medicine into an always-on radar for emerging disease.
The trade-off everyone in public health knows
Outbreak detection has been stuck between bad options. Laboratory-confirmed reporting is authoritative but slow: the pathogen has to be suspected, sampled, cultured or sequenced, and reported up a chain before it shows up in the numbers. The fast alternatives are proxies: search queries, social-media chatter, pharmacy sales. Those can move days earlier, but they are treacherous. The cautionary tale that every data scientist in the field has memorized is Google Flu Trends, a system that tracked influenza from search behavior brilliantly, until it didn't, drifting and overshooting until it was quietly retired. A proxy that correlates with disease one season can decouple from it the next.
GERN attacks the trade-off from an angle the older systems couldn't reach. Its signal doesn't come from what people search for, or even from a hurried chief complaint typed into a triage field. It comes from clinical documentation that has already passed DeepSensi's verification pipeline: a multi-specialist deliberation in which every assertion was checked against verified medical evidence before it was written down. The raw material of the radar isn't a keyword. It's adjudicated clinical reasoning.
Reading the population without reading a person
The obvious objection writes itself: clinical records are the most sensitive data there is. GERN's answer is that it never touches identity at all. Patient identifiers are stripped irreversibly at the moment of documentation: not protected downstream, but simply never present downstream. Nothing is ever surfaced below a floor of at least five similar cases clustered in space and time, so no individual can ever be singled out. Every statistic released to a health authority is mathematically perturbed under a fixed privacy budget, so the presence or absence of any one patient cannot be inferred. And the data never has to leave the country it came from: detection can run locally, so a nation's clinical records stay within its borders even as its outbreaks become visible.
What crosses a boundary is never a record. It is a number (an unusual cluster of this kind of presentation is forming in this region), and even that number is checked, twice. Before any signal is escalated, an independent panel of analytic agents has to agree it isn't an artifact of a coding change or a shift in who's seeking care, precisely the false alarms that have historically taught public-health officials to distrust syndromic systems. And a qualified human being always stands between a statistical signal and any alert to a government. GERN compresses the time it takes to detect; it never presumes to declare.
An honest radar
There is a discipline in the paper that is worth noticing, because it is rare in this field. DeepSensi claims no outbreak-detection results at all. It describes the architecture, specifies detection mathematics drawn entirely from decades of peer-reviewed epidemiology (the same aberration-detection and spatial-scan statistics public-health agencies already trust), and then lays out, in the open, exactly how it intends to measure whether the whole thing works: prospectively, pre-registered, against gold-standard confirmed outbreaks, reporting the lead time it actually delivers rather than the lead time it hopes for. It is the same posture the company takes toward its autonomous-research program, where capability is described but findings are claimed only when earned. In a domain littered with big-data hubris, a surveillance company saying we haven't proven this yet, and here is precisely how we'll try is a form of credibility in itself.
Why it matters beyond one company
The quiet radicalism of GERN is economic. A conventional surveillance system is something public health must fund, staff, and maintain as a thing apart. GERN is a by-product. The documentation is happening anyway; the radar is what you get when a privacy architecture careful enough to be trusted is pointed at the coded residue. That inverts the cost structure of early warning, and it points hardest at exactly the places the current system serves worst: the low- and middle-income regions where laboratory capacity is thin but patients are still seen, still described, still written down. DeepSensi flags that reach as a question it wants to test, not a result it has proven, but it is the question that matters most, and the company has offered the signal to the global health community as a public good, royalty-free, and asked the World Health Organization to help measure it.
Medicine has always generated this signal. Every outbreak in history announced itself first in a doctor's notes. What has never existed is a way to read those notes at the speed and scale of a population without reading a single patient. That, and not a diagnosis, may turn out to be one of the most consequential things a verified clinical system quietly does.
Technical documentation: WP-007, www.deepsensi.com/papers · Press and auditor access: [email protected] · DeepSensi PBC is a Public Benefit Corporation; public-benefit access to the GERN early-warning signal is written into its charter.
