Impact of vaccination timing and coverage on measles near elimination dynamics: a mathematical modelling analysis

Categories: Disease & Virus

September 30, 2025

Study design

We simulated outbreak risk under different vaccination strategies and coverage between 2010 and 2019. We chose this period to avoid the potential disruptions of routine immunisation programmes associated with the COVID-19 pandemic. We generated stochastic simulations using a mechanistic transmission model that had previously been fitted to the daily number of confirmed cases reported in England stratified by age, region, and vaccination status³¹. We used the parameter estimates obtained from the deterministic model fits to simulate stochastic outbreaks between 2010 and 2019. The modelled scenarios included vaccination schedules from other European countries, such as introducing MMR2 at an earlier age, or recommending a later MMR2 at school entry, improving coverage of MMR1 and MMR2 separately, and changes of schedule and coverage together.

Outbreak data

Data on all laboratory-confirmed measles cases in England between 2010 and 2019 were obtained from Public Health England (now UK Health Security Agency). This dataset included the date of onset, region of residence, age and vaccination status of the 7504 cases and was used to fit the transmission model (more details on the dataset in Supplementary Section S1).

Vaccine data

Two data sources of vaccine coverage were used: The CPRD Aurum to estimate vaccine coverage by region and 1-year age bands and the COVER to supplement missing data for age groups not included in CPRD Aurum.

CPRD Aurum is a primary care dataset from general practitioner (GP) surgeries using EMIS Web® software that contains patient-level information on symptoms and diagnoses, clinical tests and results, immunisations, prescriptions and referrals to other services⁴⁰. In 2022, CPRD Aurum contained data from around 25 million patients and was broadly representative of England by geographical spread, age, sex and ethnicity⁴¹. Using a validated algorithm to identify vaccination records⁴², vaccination coverage at the ages of 1, 2, 3, 4, and 5 years stratified by region was estimated from the electronic health records. The results were previously published⁴². The CPRD data were only available for children born between 2006 and 2015; thus, it covered all age bands between 0 and 5 in the years 2010 to 2016.

COVER is a dataset published by NHS Digital summarising UK vaccination coverage at the ages 2 and 5 for the MMR vaccine for England and by geographical region²⁶. The COVER national coverage was available for children born between 2000 and 2019. The region-stratified coverage was only available for children born in 2004 and after.

For children born before 2006 or after 2015, the CPRD vaccine data had to be supplemented with estimated values from COVER data. COVER uses aggregated GP information based on operational data, which may be incomplete, not fully representative and not quality assured⁴³. COVER data that has been corrected for underascertainment is closer to CPRD estimates. A comparison between COVER data and CPRD data can be found in the Supplementary Material (see Supplementary Section S2). Based on this, COVER estimates used to supplement missing CPRD data were adjusted using the assumption that 50% of the unvaccinated children were vaccinated but did not have their vaccine recorded¹⁵. These corrected values were consistent with estimates from previous studies²⁷. To estimate the vaccine coverage in the missing age bands in the COVER data, we applied the relative difference of the age bands for the last completely available years, i.e. 2006 and 2017, to the COVER estimates to supplement the values for ages three and four. All values of vaccine coverage were stratified by region (see Fig. 5).

**Fig. 5: MMR vaccination coverage by region.**

In a sensitivity analysis, we fitted the model to the uncorrected COVER data for which the missing age strata were supplemented by the proportional change in uptake between age groups as observed in the CPRD data.

As the recommended ages for MMR vaccination are 1 year (MMR1) and 3 years and 4 months (MMR2), most children receive a first dose between one and two, and a second dose between three and four. Therefore, the vaccination coverage for MMR1 at age 1 and MMR2 at age 3 is nearly zero in the data, but a large proportion of children aged one have actually received a dose of vaccine (same with MMR2 for children aged three). To adjust the coverage data in the 1-year age band structure of the model, we assumed the coverage of MMR1 at the age of 1 was 75% of the coverage for MM1 at age 2, and 50% of the coverage at age 3, of the coverage of MMR2 at age 4.

Transmission model

We used a compartmental transmission model with compartments for susceptible, exposed, infected and recovered individuals (SEIR model) and single and double vaccinated individuals to reproduce the measles dynamics observed in England by age group, region, and vaccine status. The transmission model was presented in detail in a previous publication³¹. A more detailed description of the model fitting process and model parameters can be found in the Supplementary Material (see S3). The parameters estimated by the model were then used to generate stochastic simulations, describing the case dynamics simulated under a range of alternative scenarios (see Fig. 6)³¹.

We ran 2500 simulations per vaccination strategy by drawing 100 parameter sets from the model fits and running 25 simulations per parameter set. Parameters drawn from the model fits included infection rate, duration of maternal immunity, parameters for seasonality of transmission and importation, report rate of imported cases, vaccine effectiveness, existing immunity in older generations and parameters of spatial spread (see Table S7). In the stochastic simulations, the number of transitions was computed using binomial draws, with a rate of transition derived from the distribution of the population between compartments at the previous time step and the parameter set. Counterfactual scenarios were matched by using the same seed for the stochastic simulations.

Alternative vaccination scenarios

We explored how changes in vaccination impacted the number of cases simulated by the transmission model. To do so, we generated sets of stochastic simulations using the same parameter sets as the reference simulation set (described in section S2 in the supplementary material), but with changes in vaccination schedule, vaccine coverage, or both. Changes in schedule were based on other vaccination schedules implemented in Europe. Table 1 summarises the different scenarios for which we simulated measles outbreaks in England.

Table 1 Overview of the different vaccination strategies

We used the median number of cases between 2010 and 2019 and the IQR in each simulation set to estimate the impact of changes in vaccination schedule and coverage. The number of cases is compared to the reference set of simulations by computing the median and IQR of the percentage of change (increase or decrease) between each simulation set and the median number of cases in the reference simulations.

Changing vaccine coverage

We created scenarios in which the timing of MMR2 was not changed from the original schedule, but the overall coverage of MMR1 or MMR2 was increased across all regions and years between 2010 and 2019. We implemented an absolute increase of 0.5 and 1% for either MMR1 or MMR2 in every region between 2010 and 2019. As coverage for MMR2 at the age of five is lower than for MMR1, we further included scenarios increasing or reducing MMR2 coverage by up to 3%.

Changing MMR2 schedule

We explored two alternative scenarios with MMR2 recommended at the age of two instead of 3 years and 4 months. The first scenario assumed that the uptake would follow the same pattern of timeliness as the current MMR2 delivery; therefore, 1 year and 4 months were subtracted from all the dates when MMR2 was received. New coverage estimates were calculated from these updated dates. In the second, we assumed that MMR2 recommended at the age of two would be taken up with the same speed as MMR1, which is usually faster than MMR2²⁷. In the new schedule recommended by the JCVI, MMR2 would be delivered at 18 months, but since our model is stratified by age groups (1-year age bands until 6), we used 24 months in the simulations.

We also explored the impact of moving the MMR2 recommendation to 5 years of age, assuming that the speed of uptake was similar to the reference MMR2 speed. New coverage estimates were calculated from the updated dates.

Finally, we looked at potential scenarios in which an earlier MMR2 would influence coverage for MMR2. For this, we explored an increase by 0.25, 0.5 and 1% and added a scenario in which the coverage for MMR1 and MMR2 were equal. Lastly, we added one scenario in which the coverage of MMR2 would be negatively impacted by bringing MMR2 forward in the schedule.

We assumed that all vaccinated children had received their second dose at 2 at the start of the simulations. This means that the effects of a transition period where part of the children would still be vaccinated on the original schedule are not considered in these scenarios.

Sensitivity analysis

We conducted several sensitivity analyses: (1) we recreated the scenarios with changes in the vaccination schedule using uncorrected COVER data to describe the annual vaccine coverage, with CPRD data informing the proportion of vaccinated children at 3 and 4 years of age (Section S5 in the Supplementary). (2) We included waning of vaccine-induced immunity from age 5, whereby vaccinated individuals who developed immunity upon vaccination can become infected, with infection risk increasing with age. In the model with waning, age is used as a proxy for time since vaccination, and the waning rate is based on estimates of our previous work³¹. (3) As moving MMR2 delivery to age 2 may influence when waning begins, we ran a sensitivity analysis with waning of immunity starting at age 3 when MMR2 was given early (Section S6 in the Supplementary)³¹.

Inclusion and ethics

We received data governance approval from CPRD (protocol number 22_001706) and ethical approval from the London School of Hygiene and Tropical Medicine’s research ethics committee (reference number 27651).