**2. Materials and Methods**

Analyses were completed within the Secure Anonymised Information Linkage (SAIL) Databank held at Swansea University [21]. A cohort of individuals aged 2 to 25 years of age, alive and resident in Wales as of 31 August 2021, was created using the Welsh Demographic Service Dataset. Individuals who do not have a record in the NCCHD or were registered to a GP that does not submit data to SAIL were excluded. Approximately 80% of GP practices in Wales submit data to SAIL [22].

Measles vaccination status was assigned using an extract from the NCCHD, supplemented by Read coded vaccination status data from primary care GPs. MMR, measles and rubella (MR) and single antigen measles vaccination were all considered valid vaccinations in this analysis. In line with UK guidance, the first dose of vaccine had to have been given at 12 months of age or later with the second dose given at least one month after the first at 15 months of age or later [23].

A series of independent variables were identified to test for association with vaccination coverage. Gender, age as of 31 August 2021, month of birth, mothers' age at birth, health board of residence, and age first moved to Wales were taken from the Welsh Demographic Service Dataset. Urban/rural classification and deprivation quintile of residence were derived as described previously [24]. Broad ethnic group was derived from the Office for National Statistics 2011 census, with information taken from the Education Wales Schools and Pupils Dataset or primary care GP record, if census data were unavailable. Total number of primary care visits in the 1 September 2020 to 31 August 2021 year and age

first registered with a primary care GP in Wales were calculated using data from primary care GPs, and flags were derived for learning disability, diagnosed sight loss and hearing loss based on published primary care Read code sets [25,26]. Mothers' unique identifier, birth order, maternal smoker flag and premature status (born before 37 weeks' gestation) were taken from the NCCHD. A flag for ever being eligible for free school meals, attendance at a special school or ever being excluded from school was taken from the Education Wales Schools and Pupils Dataset. Information on mothers' highest qualification was taken from the 2011 census. Religion was as recorded in census data, otherwise as recorded in data from primary care GPs. Where there were contradicting values, the most recent record was kept. Mothers' religion was used as default; otherwise, where this was missing, the child's recorded religion was used. Country of birth (COB) was derived from the Office for National Statistics Annual District Birth Extract. If a child was born in Wales they appear in this data; otherwise, this information was taken from the 2011 census or data from primary care GPs. Where a child's COB was unknown, mother's COB was used. Mother and child's recorded language was taken from census data, and where this was unavailable, language data from primary care GPs was used. If this information was not recorded in either dataset but they were born in Wales it was assumed English/Welsh was a primary language. A Charlson Comorbidity Index score was created using data from primary care GPs based on published Read code sets [27]. Previous vaccinations (three doses of pertussis-containing vaccine, one of pneumococcal vaccine and two of rotavirus vaccine) as outlined in the UK schedule were derived using the same methods used for measles-containing vaccine.

The odds of being vaccinated with one and two doses of measles-containing vaccine were calculated, with independent variables considered significant at the 0.05 level. In a multivariable analysis of those aged 4 to 25 years, records with missing information were dropped. The maternal smoker flag was dropped due to a high proportion of missing data. Mothers' recorded language and age first moved to Wales were excluded from the multivariable model due to co-linearity with child's recorded language and age first registered with a Wales GP, respectively. The final model was constructed stepwise in order of strength of association as indicated by the univariable analysis; variables which did not improve the Akaike Information Criterion score were dropped.

Unvaccinated individuals with a vaccine refusal Read code (68NY., 68NB., 68NP., 68NR., 68Nb., 68Na., 8I3x., 68N6., 68NM.) on their GP record were excluded from the equality analysis and described separately.
