The Outbreak Hidden Inside a National Average

In the official tables for 2024 to 2025, 91.8 per cent of five-year-olds in England had received a first dose of the measles, mumps and rubella vaccine. For the second dose, coverage was 83.7 per cent [1]. These are troubling figures because both sit below the 95 per cent target used for measles elimination. Yet they still compress nearly every vaccinated child and every missed appointment in the country into two decimals. An outbreak does not occur inside that average. It occurs among people who meet.

The arithmetic can hide very different worlds. Imagine two populations with the same 90 per cent coverage. In one, the unvaccinated tenth is scattered and mostly surrounded by immune neighbours. In the other, the same number of susceptible people attend the same schools, share households and use the same community spaces. The average is identical. The route available to a virus is not. Connectivity turns a count into transmission.

This is why public-health epidemiologists care about heterogeneity, the variation concealed within a summary. National vaccination coverage is valuable for tracking programmes over time and comparing broad performance. It cannot show whether missed doses are clustered by neighbourhood, age, school, deprivation or access to primary care. When the pathogen is as contagious as measles, those details can determine whether one imported case stops or becomes hundreds.

England's data contain several levels of compression. In 2024 to 2025, 88.9 per cent of children had received their first MMR dose by 24 months. By age five, 91.8 per cent had received it, showing that some late vaccinations had been caught up. Only 83.7 per cent had received both doses by five [1]. Each denominator describes a different birth cohort at a different checkpoint. None says whether the remaining children are the same people across measures or why their doses were missing.

Geography begins to pull the average apart. The first-dose figure at age five was 95.4 per cent in North East England and 84.4 per cent in London. For two doses, the corresponding regional values were 90.2 and 69.6 per cent [2]. Even regions remain large containers. The official equity audit reported first-dose coverage at 24 months ranging from 96.3 per cent in North Tyneside to 65.3 per cent in Hackney and the City of London [2]. The London data carry a warning that recent information-system and methodology changes may underestimate some coverage, especially the second dose. Measurement quality is part of the map.

A low recorded percentage can reflect missed vaccination, incomplete transfer of records or both. That distinction changes the response. If children are truly unvaccinated, services need to reach them. If doses were given but not recorded, record linkage and data repair may be the priority. Usually, both problems coexist. Treating a dashboard as an unquestionable census can send effort to the wrong place; dismissing imperfect data can leave a genuine immunity gap untouched.

Birmingham provided a harder kind of evidence. Between 13 October 2023 and 12 April 2024, the UK Health Security Agency recorded 406 confirmed measles cases in the city. The median age was 5.5 years, and 89 per cent of cases were unvaccinated. Seventy-eight per cent occurred among people living in the most deprived fifth of local areas, while none occurred in the least deprived fifth [3]. The outbreak did not distribute itself like the city's population.

Those figures do not mean deprivation is a biological cause of measles. They describe the social organisation of susceptibility and exposure. Families in deprived areas may face unstable housing, inflexible work, transport costs, crowded services or difficulty obtaining appointments. Primary-care records can fragment when people move. Some communities encounter language barriers or services that do not feel trustworthy. The result can be delayed vaccination and denser opportunities for transmission. A label such as hesitant can obscure those practical mechanisms.

The Birmingham investigators calculated measles rates of 47.6 per 100,000 people in the most deprived fifth and 13.8 in the middle fifth. Rates also varied across recorded ethnic groups [3]. These are descriptive inequalities from one outbreak. Ethnicity can be entangled with geography, age, deprivation, household structure, access and the completeness of population estimates. The study does not justify treating identity as a causal risk factor. It justifies asking which services and conditions produced unequal protection.

Zooming back out changes the picture again. England confirmed 2,911 measles cases during 2024. London reported 1,305, or 44.8 per cent, while the West Midlands reported 562. Birmingham alone accounted for 364 cases with symptom onset that calendar year [4]. A national epidemic curve was therefore the sum of local outbreaks at different stages. If one region is rising while another is falling, the national line can appear flat. The average can hide movement as well as place.

The same problem affects incidence, the number of new cases in a population over a period. A national rate divides all cases by the whole population. That denominator can make a severe outbreak in a small community look negligible. Local rates restore some scale, but administrative boundaries may split a connected community or combine populations that rarely interact. The correct unit is the transmission network, which rarely matches a neat line on a map.

Researchers have tested how badly aggregation can mislead. Nina Masters and colleagues built simulated populations with the same overall measles vaccination coverage but different fine-scale patterns of non-vaccination. When they aggregated the data into larger areas, distinct patterns converged on the same summary even though their outbreak potential differed. At 95 per cent overall coverage, local clustering could still create substantial vulnerability [5]. The model did not predict a specific English outbreak. It demonstrated the mathematical information lost when fine detail is averaged away.

The mechanism is herd protection. When enough people are immune, an infected person is unlikely to encounter a susceptible person before recovering, so transmission chains tend to end. The threshold is sometimes presented as a clean national percentage. In practice it depends on vaccine effectiveness, contact patterns, age and the virus's contagiousness. A cluster can fall below its local threshold even when the surrounding country sits above it. Herd protection is a property of connected populations, not a certificate awarded to a nation.

Fine-scale data can reveal risk, but detail creates obligations. Small numbers are unstable: a handful of records can swing a percentage in a small school or practice. Population denominators can be out of date. Publishing granular maps may identify or stigmatise communities, especially if the public mistakes low coverage for moral failure. Health agencies need enough resolution to act, with privacy rules and communication that explain uncertainty. The most detailed number is not automatically the truest or most ethical.

Better mapping also requires better denominators. The numerator is the number of people recorded as vaccinated. The denominator is everyone eligible. Both can be wrong. Duplicate patient records inflate or distort counts. Children registered in one place may live in another; recent arrivals may be missing. Some vaccinations delivered abroad, privately or at school may not reach the primary-care record. Coverage can even exceed 100 per cent in a badly aligned local dataset. Data cleaning is therefore outbreak prevention work, not clerical decoration.

Case surveillance has parallel blind spots. A confirmed measles case depends on symptoms being recognised, care being sought, the possibility being reported and a specimen being tested. Communities with poor access may be undercounted until severe cases reach hospital. During an outbreak, increased awareness can raise detection, making the curve grow partly because surveillance improves. Genomic sequencing and contact tracing can connect cases, but neither recovers every mild infection. A precise case total still represents a process of observation.

The solution is to layer imperfect sources. Vaccination registers estimate susceptibility. Laboratory reports confirm disease. School and primary-care data locate gaps. Hospital admissions show severity, while community organisations explain barriers that administrative data cannot see. A map becomes useful when it changes delivery: extended clinic hours, translated invitations, mobile sessions or catch-up vaccination in a setting people already use. The goal is not to colour an area red. It is to remove the reason it became red.

A useful local indicator also needs a trigger agreed in advance. A falling coverage estimate might prompt record validation first, followed by outreach if the gap remains. One confirmed case may justify checking contacts and nearby coverage, while several linked cases can trigger incident structures and extra clinics. Predetermined triggers reduce the temptation to wait until hospitals confirm that a statistical warning was real. They also keep every small fluctuation from producing a disruptive emergency response.

Targeting must not become exclusion. A national vaccination programme remains necessary because risk changes and people move. If resources flow only to yesterday's outbreak, another missed cohort can accumulate elsewhere. Universal provision creates the floor; local intelligence identifies where the floor has cracked. This balance is harder than a single campaign because it depends on routine staffing, accurate records and sustained relationships after media attention fades.

National averages can also conceal inequality in vaccine effectiveness, diagnostic access and disease severity. Influenza vaccination may perform differently by age, and tuberculosis incidence can be concentrated in particular urban communities even when a country's rate is low. Wastewater surveillance represents only the population connected to sampled sewers. The broader lesson is about scale: every population indicator is an answer to a question framed by its denominator and geography. Change either, and a different public-health reality may emerge.

Time creates another hidden dimension. Annual coverage can average months in which a cohort was poorly protected with months after a catch-up campaign. Weekly case totals can combine separate local curves whose peaks occur at different moments. Analysts therefore use age-specific cohorts and shorter time windows alongside maps. The purpose is the same: preserve enough detail to see a transmission opportunity before aggregation smooths it away.

Averages are still indispensable. Without them, trends become anecdotal and national accountability weakens. The danger begins when a summary is mistaken for a distribution. England's MMR percentage correctly says that coverage is below target. It cannot say which child is unprotected, whether susceptible children share contacts or where the next introduction will land. Those questions require data closer to the communities where transmission happens.

Birmingham's outbreak became visible in confirmed cases, but its conditions had been accumulating beforehand in delayed doses and uneven access. By the time a national line rises, a local chain may have passed through many households. The earliest warning may therefore be a modest percentage in a small area, provided someone trusts it enough to investigate and doubts it enough to verify. Public health has to read both the average and the exception.

A nation is a useful administrative unit. A virus experiences it as millions of encounters. That is the scale at which an apparently safe average can contain an outbreak, and the scale at which prevention must eventually succeed. The work begins by refusing to ask only how many people are vaccinated. It asks who is still susceptible, where they connect and what would make the next dose easier to receive.