The Unwritten Category: Reading the Gaps in NHS Ethnicity Data

In a digitised health record, a blank field can look like an absence of information. In practice, it is often a piece of information in its own right. Ethnicity is one of the clearest examples. When NHS datasets are used to describe inequalities in access, experience or outcomes, the ethnicity field is frequently incomplete or inconsistently populated. Those gaps do not simply reduce precision. They can change the shape of the story that the data appears to tell.

Three kinds of empty

A blank ethnicity field is not a single phenomenon. The first distinction is between a patient who was asked and declined to answer, a patient whose ethnicity was not known to the service, and a record where the field was never completed at all. In NHS coding, labels such as 'not stated' and 'not known' may appear alongside the familiar census-style categories. But the meaning of those labels depends on the system, the setting and the data dictionary in use.

A 'not stated' code may reflect a patient choice. A 'not known' code may reflect a administrative gap. A null value may reflect neither: it may be a field that was skipped during a busy registration, a default setting in a clinical system, or a record that was migrated poorly between platforms. These differences matter because treating all blanks as one category can hide the very process that produced them.

Where the gaps gather

Completeness is rarely uniform across the health service. Recording tends to be better where staff have repeated contact with a patient and a stable record, such as in some primary care and long-term condition pathways. It can be weaker in urgent, temporary or high-throughput settings, where registration is compressed and the clinical task takes priority over the demographic field. Mental health, maternity and cancer datasets may each have their own recording cultures and their own completeness profiles.

Variation also exists between organisations and between conditions. A trust with a well-established data quality programme may achieve much higher completeness than a neighbouring service with older systems or less analytical capacity. None of this means that incomplete records are necessarily biased in a predictable direction. It means that the pattern of missingness has to be examined rather than assumed away.

Why missingness is not neutral

When analysts calculate an inequality statistic, they usually compare outcomes between ethnic groups. If the ethnicity field is incomplete, the comparison depends on how the missing records are handled. Excluding blanks, also known as complete-case analysis, assumes that the records left behind are representative of the whole population. That assumption is often untested.

Suppose one ethnic group is more likely to have a blank field and also more likely to experience a poor outcome. If those records are excluded, the group's measured outcome may look better than it is. If missingness is concentrated among people with less severe illness, the opposite can occur. The direction of bias is an empirical question, not a matter of intuition.

There is a further complication. Sometimes 'unknown' is retained as a category in its own right. That can be useful, because it makes missingness visible. But it can also create a misleading comparison if readers treat 'unknown' as if it were a homogeneous ethnic group. In some datasets, the 'unknown' category is large enough to rival the smaller ethnic categories, and its outcomes may reflect a mixture of unmeasured disadvantage, administrative neglect and clinical complexity.

The denominator problem

Inequality statistics are not only about numerators. They also depend on denominators. Ethnicity-specific rates often use population estimates from the census, while NHS service data use their own registered populations. These denominators do not always align, particularly in areas with high population turnover or large numbers of people not registered with a GP. A percentage can look precise while resting on a denominator that is itself uncertain.

This is why a headline such as 'ethnic group X is twice as likely to experience Y' deserves a closer look. The comparison may be based on different levels of data completeness, different age structures, different geographic footprints and different handling of missing records. None of these issues makes the statistic worthless. They do make it conditional.

A reader's checklist for 2026

When reading ethnicity-linked health statistics, the following questions are a useful starting point.

• Locate the code definitions. Find the data dictionary or data quality statement. Check whether 'not stated', 'not known' and null values are treated separately or combined.
• Ask about completeness by group. A single overall completeness figure is not enough. Ask whether completeness varies by ethnic group, age, sex, deprivation and setting.
• Check how missing data were handled. Were blanks excluded, imputed, grouped into an 'unknown' category, or analysed in a sensitivity test?
• Look at the time period. Recording practices change. A rise in recorded ethnicity can look like a change in inequality when it is partly a change in data capture.
• Check for suppression. Small numbers are often suppressed to protect confidentiality. Suppression can remove exactly the groups with the least reliable data.
• Compare sources. A pattern seen in hospital data may not appear in primary care or mortality data. Triangulation is a check on artefact as well as a check on truth.
• Treat 'unknown' as a finding. A large unknown category is a signal about the data production process, not just a nuisance to be discarded.
• Avoid causal language. Missingness can bias an association without explaining it. A gap in the data is not the same as a gap in care, although it may be a symptom of one.

What good reporting looks like

Transparent reporting does not require perfect data. It requires that the limits of the data are visible. A strong analysis will show completeness alongside the main result, explain how missing records were handled, and test whether the conclusion survives a different reasonable choice. It will also distinguish between a change in health and a change in recording.

In 2026, UK health data is more granular and more widely used than ever. That is a gain for public understanding, but it raises the stakes for careful reading. The ethnicity field is not a decorative demographicextra. It is part of the machinery that turns a clinical encounter into a statistic about equality. When the field is left blank, the statistic is not simply incomplete. It is answering a slightly different question from the one the headline implies.

The unwritten category, in other words, is still a category. Reading NHS ethnicity data well means asking not only who is counted, but who has been left out of the count, and why.