In the autumn of 1854, a physician named John Snow removed the handle from a water pump on Broad Street in the Soho district of London. The gesture was modest — almost anticlimactic — but it marked a turning point in the history of public health. Snow had spent weeks mapping the distribution of cholera deaths in the neighbourhood and had identified the pump as the most probable source of contamination. His map, which plotted each death as a small bar on the relevant address, demonstrated a striking concentration of cases around the Broad Street well. It was one of the earliest and most celebrated examples of what would later become known as epidemiology: the systematic study of disease patterns in populations.
What made Snow's achievement remarkable was not merely the accuracy of his conclusion but the method by which he reached it. At a time when the dominant theory attributed cholera to foul air — the so-called miasma theory — Snow relied not on microscopic observation or laboratory experiment but on the careful collection and analysis of numerical data. He counted deaths, plotted their locations, examined the water sources used by the affected households, and compared mortality rates between areas served by different water companies. The numbers, arranged and interpreted with rigour, told a story that no amount of theoretical speculation could have produced. Snow did not need to identify the cholera bacterium, which would not be isolated for another three decades, to establish how the disease was transmitted.
This episode is frequently invoked as a founding moment of evidence-based medicine, and it deserves the honour. But it also illustrates a tension that continues to animate public health debates to this day: the relationship between statistical evidence and mechanistic understanding. Snow's data pointed overwhelmingly to contaminated water as the vehicle of transmission, yet his contemporaries were reluctant to accept the conclusion precisely because no biological mechanism had been identified. The absence of a satisfying causal explanation made the statistical pattern, however compelling, feel incomplete — an objection that, in various forms, recurs whenever data-driven conclusions challenge established theoretical frameworks.
The tension has resurfaced with particular intensity in the age of big data. Modern epidemiologists have access to datasets of a scale and granularity that Snow could not have imagined, and the analytical tools available to them — machine learning algorithms, geospatial modelling, genome-wide association studies — can detect correlations with extraordinary sensitivity. Yet the capacity to detect patterns far outstrips the capacity to explain them. A machine learning model may identify a previously unsuspected correlation between two variables without offering any insight into why the correlation exists, leaving researchers in a position not unlike Snow's: armed with persuasive evidence but lacking a mechanistic account of how the effect is produced.
The pragmatic response to this situation has generally been to act on the evidence while continuing to investigate the mechanism. Snow's removal of the pump handle did not wait for Robert Koch's identification of Vibrio cholerae in 1883; similarly, contemporary public health interventions are regularly implemented on the basis of strong statistical associations before the underlying biology is fully understood. This approach is defensible — and in urgent situations, essential — but it carries risks. Correlations can be misleading, confounded by hidden variables that distort the apparent relationship between cause and effect. Without mechanistic understanding, it is difficult to distinguish genuine causal links from statistical artefacts, and policies built on spurious correlations can do more harm than good.
The legacy of John Snow is not simply that he solved a particular puzzle about a particular disease. It is that he demonstrated the power of disciplined numerical reasoning to reveal truths that are invisible to unaided intuition, while simultaneously exposing the limits of that reasoning when it operates in the absence of deeper understanding. The best public health science, then and now, requires both: the rigour of the statistician and the curiosity of the biologist, each checking and enriching the other.