Epidemiological surveillance since John Snow in public health

The objection that should win

Here is the strongest case against everything that follows. Snow did not prove that data beats theory. He had a theory — waterborne transmission of cholera — before he ever drew the Soho map, and the map was built to test it, not to generate it. Turn that around and the danger becomes obvious. Feed a surveillance system continuous, unstructured intake with no hypothesis to police it, and it will manufacture clusters that mean nothing. Case counts by postcode wobble. Wastewater titres spike on rainfall, not on outbreak. Genomic surveillance flags a lineage as "concerning" because three sequences happened to arrive from the same clinic in the same week. This is the classic problem of spurious spatial and temporal aggregation, and it gets worse, not better, as more streams are added. An epidemiologist drowning in dashboards is not closer to the truth than one reading a single well-designed cohort study. More intake without theory is a machine for false alarms.

Continuous surveillance without a governing hypothesis is not vigilance. It is noise with a timestamp, and every additional feed multiplies the noise faster than the signal.

That objection is correct, and it has to be answered honestly, because it is not a debating point — it is the daily experience of anyone running a syndromic surveillance system.

What Snow actually did

Miasma theory, the dominant account of Snow's day, held that cholera rose from foul air given off by decaying matter. It was coherent. It explained the association between filth and disease that everyone could see. It was also a frozen intellectual corpus in the precise sense that matters here: a body of accepted reasoning, fixed before any particular outbreak, applied to new cases by inference within the theory rather than correction from outside it. William Farr, compiling the Registrar-General's mortality returns from 1839 onward, had already built a standing series of numbers that miasma theory could absorb without strain — higher elevation, less filth, less disease, all consistent, all wrong about mechanism.

Snow's 1854 investigation did not discard theory in favour of data. It used arriving, located cases to adjudicate between two theories already on the table. He plotted 578 deaths as stacked bars on a street map and watched them converge on the Broad Street pump. Then he went looking for the cases that should not fit if his theory were true, and the cases that would refute it if it were false. Seventy brewery workers on Broad Street drank their own brew, not the pump water, and none died. A widow in Hampstead who liked the taste of Broad Street water had it carted to her weekly, and she died, alone, far from the cluster. Those two facts did the work. They were not generated by wide intake. They were sought out because the theory said where to look.

This is the concession the argument must make in full: intake does not supply hypotheses. It bounds them. Wide, continuous, provenance-tagged intake does not replace an epidemiologist's judgement about mechanism. It narrows the field of judgements that can survive contact with what keeps arriving. That is a real function, and a much smaller one than "surveillance will find the truth for you."

The three positions surveillance actually occupies

Set against that narrower claim, the three-stage structure still holds, and public health passed through all three stages before 1900.

Reasoning from a frozen prior body of knowledge is miasma theory: internally coherent, unfalsifiable by any single new death, because new deaths were interpreted through the theory rather than against it. Direct observation of a bounded, present scene is Snow with his notebook: richly informative about one outbreak, in one parish, over a few weeks, and finished once the pump handle was removed and the investigation closed. Permanent streams that revise standing belief, with provenance attached, are the notifiable-disease apparatus that grew out of Snow's method: statutory notification after the Infectious Disease (Notification) Act of 1889, sentinel general practices, and eventually molecular typing and wastewater assay layered on top.

The third position is where the discipline has stayed, and improved, for over a century. Wastewater surveillance for poliovirus is the clean modern instance: Israel detected wild poliovirus type 1 circulating in sewage in 2013 with no paralytic cases at all, because only about one infection in two hundred causes paralysis and the clinical stream is blind to the rest. London's sewage yielded vaccine-derived poliovirus in 2022 by the same route, again with no case in front of it. Neither event was a new kind of evidence. Both were the notifiable-disease principle — a standing stream, revising belief, carrying its own provenance — extended to a medium Farr never sampled.

The failure that never quite goes away

None of this abolishes the characteristic failure of the field: an outbreak confirmed three weeks after the epidemic curve has already turned. Wastewater titres lag infection by days; genomic confirmation of a new lineage lags sequencing capacity by more; clinic load lags severe disease by the incubation period plus however long people wait before presenting. An epidemiologist assembling belief from four asynchronous streams, each with its own delay and its own failure mode, is not looking at reality. They are looking at four stale, differently stale, partial views of reality, and the job is to weight them by how stale and how partial each one currently is.

This is where the second objection worth taking seriously enters, and it is not really about evidence at all. Notifiable disease law worked because clinicians could be compelled and sanctioned, not because it revealed a new class of fact — the same clinical picture existed before 1889, unreported. The bottleneck was compliance and jurisdiction, not sensory reach. That reading of the history is largely correct, and it does not undermine the intake argument; it completes it. Once intake has reached the standing-stream position, the remaining hard problems are who is obliged to report, who is trusted to see the pooled data, and which stream's flag gets acted on when two streams disagree. Those are institutional problems, not intake problems. Naming them as the successor difficulty is part of this claim, not an escape from it.

Why there is no fourth position to reach for

The temptation, faced with a lag like that, is to imagine some further category of intake that would close it — a system that sees the outbreak the instant it starts, rather than three weeks after. No such category exists, and it is worth being precise about why. Real surveillance systems are sparse by design. Sentinel practices cover a small fraction of clinics. Notifiable disease lists run to a few dozen conditions, not every possible pathogen. Syndromic categories discard clinical detail on purpose, trading resolution for speed. Calling any of this "everything, continuously" is inflation, and a fair critic is right to say the terminal position, read literally, has never been occupied.

But the claim was never about completeness of coverage. It is about the category of evidence admitted. A sentinel network sampling one per cent of general practices, running without a stopping date, revising belief every week and recording which practice each signal came from, belongs to a different category from a bounded investigation that ends when the report is filed — even though both are, in some crude sense, "just data." Coverage is a scale variable inside the standing-stream position: it can go from one per cent to ninety per cent without changing category. The change from an investigation that closes to a stream that does not is a change in kind. An epidemiologist can always ask for more sentinel sites, faster sequencing turnaround, another wastewater catchment. None of that is a request for a new axis of evidence. It is a request for more of the same axis, arriving sooner, trusted more. That is the sense in which the third position is terminal: not finished, not complete, but the last new kind of thing intake can be.