Run a measured cohort study of sleep and cardiovascular outcomes in populations adjacent to operating data centres

The proposal

An epidemiological study that measures exposure and outcomes in the same population. Exposure: continuous octave-band and dBC monitoring at a sample of residences at varying distances from operating facilities, indoors and outdoors, over at least one full seasonal cycle. Outcomes: sleep actigraphy and validated sleep questionnaires, ambulatory blood pressure, and where available linked primary-care or hospital records for hypertension and cardiovascular endpoints.

Design: a distance gradient within each cluster, plus a before-and-after comparison where a facility is commissioned during the observation window — the stronger identification strategy where available.

Why it would work

The dose-response relationships currently invoked come from road, rail, and aircraft noise, which differ from data centre noise in the two variables the mechanism is thought to depend on: the presence of discrete arousal events and the frequency content. Measuring the actual source removes the extrapolation.

The exposure measurement is already routine. The Lyver study demonstrated receptor-level measurement and buildout modelling at 25 locations for a county government (Lyver). What has never been done is joining that measurement to health endpoints in the same households.

Implementation path

Northern Virginia is the obvious primary site: roughly 300 facilities, about a third within 200 feet of residential zoning, and a decade of operating history next to stable populations — giving both a distance gradient and enough exposed households for statistical power. Dublin and the Netherlands offer clusters with linked health records and different regulatory baselines.

Plausible funders are national environmental health agencies, university public health departments, and state legislatures currently setting numeric limits without evidence. Deliberately not the operators: a study funded by the party whose permit depends on the result will not settle anything, which is exactly the dynamic that has left the groundwater question in Georgia unresolved.

Trade-offs and limitations

• Confounding. Facilities cluster near highways and industrial corridors, so road noise, air pollution, and socioeconomic gradients all correlate with proximity. Distance-gradient designs need careful covariate control, ideally with measured rather than modelled co-exposures.
• Statistical power. Effect sizes in transport noise are modest (relative risks near 1.08 per 10 dB), requiring large samples or long follow-up. A first study may establish sleep and annoyance endpoints while being underpowered for cardiovascular ones, and should be designed and reported honestly on that basis.
• Timescale. Cardiovascular endpoints accrue over years. Sleep and blood pressure outcomes are measurable within a season and are the sensible primary endpoints for a first study.
• The result may not favour the community case. A well-designed study could find that continuous low-level tonal exposure produces annoyance and sleep disruption without measurable cardiovascular effect. That is a useful finding, and anyone commissioning this work should commit to publication before seeing the result.