key: cord-0801464-5uht51lj authors: Dropkin, G. title: Variation in English Covid booster uptake date: 2022-02-01 journal: nan DOI: 10.1101/2022.02.01.22270236 sha: f52d32651cade6b1ba5802658bbf84e72b1e4edb doc_id: 801464 cord_uid: 5uht51lj Introduction: Variable and low takeup of the Covid booster is a recognised problem, associated with age, gender, ethnicity, and deprivation. Are there other relevant predictors? Methods: Data was downloaded from the UK Government Coronavirus Dashboard for Middle Super Output Areas in England, along with demographic, employment, and health data from public sources. Mixed models with a random factor for Upper Tier Local Authority were analysed as quasibinomial Generalized Additive Models, and the estimated random factors were then fitted with Bayesian linear mixed models using flu vaccination uptake, change in public health budgets, vaccination centres, and Region. Results: Models for the MSOA Covid 1st and 2nd vaccinations and the 3rd injection (including the booster), fit the data well. Index of Multiple Deprivation, proportion Aged 15-24 and 25-44, and ethnicity groupings Other White, Indian-Pakistani-Bangladeshi, and African-Caribbean-Other, are highly significant predictors of lower uptake. The estimated random factors vary widely amongst local authorities, and can be predicted by flu vaccine uptake, rise in public health budgets, and regional effects which are positive for London and South East, and negative for North West and North East. Vaccination centres did not reach 90% significance. Discussion: Covid vaccination rates at each stage are very well modeled if local authority random effects are included along with non-linear terms for demographic, employment and health data. Deprivation, younger age, and Other White, South Asian, and Afro-Caribbean ethnicities are associated with lower uptake. Modelling the local effects indicates that increasing public health budgets would improve vaccination uptake. The booster programme has been central to the UK government's strategy for containing in England during the autumn and winter of 2021-22. 1 However, takeup of the booster remains well below the levels achieved for the first and second vaccine doses. 2 Vaccination uptake is highly dependent on age, gender, ethnicity, and deprivation, as is widely recognised 3 4 5 6 7 8 9 Model C enlarges A with additional demographic and local health covariates for Communal establishments, multi-occupation housing, population weighted mean distance to GP, to A&E, to Pharmacy, proportion registered with GP, Health deprivation, and Education deprivation C = A + s(LCommun)+s(LHous)+s(LGPW)+s(LEDW)+s(LPHMW)+s(GPR)+s(IMDH)+s (IMDE) Model D enlarges A with most of the additional terms from B and C, but omits six terms whose contributions showed negligible or low significance: Density, Manufacturing, Accommodation and Food, Information and Communication, Real Estate, mean distance to GP. Models were fitted separately for each vaccination, using the quasibinomial family with smoothing parameter optimisation by marginal likelihood (method = "ML"). 27 The output includes the fitted model coefficients b and their Bayesian posterior covariance matrix V c which includes correction for smoothing parameter uncertainty (an option with the "ML" method). Model fit was evaluated with gam.check, and outliers detected by cooks.distance > 0.02, a somewhat arbitrary criterion. The actual and predicted proportions vaccinated in each MSOA were plotted, along with the smoother curves for particular covariates. Models were compared by ML value (lower value indicates better fit). For each fitted model, the first 149 coefficients c i are the estimated contributions of the levels of utla. Each c i has a standard error se i , obtained as the square root of the i th diagonal term of V c . "Caterpillar plots" were drawn using the c i and se i to show the variation in estimated random effects. These coefficients also enable comparison of the fitted value for a particular MSOA, with the hypothetical fitted value if the same population were located in a different UTLA, replacing c i with c j . If g denotes the (quasi) binomial link g(p) = log(p/(1-p)) and h is the inverse link h(t) = 1/(1+exp(-t)), the fitted value would be altered from fit 1 to h(g(fit 1 ) -c i + c j ). In a second stage of modelling, the c i were taken as observed, to be predicted using covariates available for Upper Tier Local Authorities. Region was treated as a random effect, with fixed effects for the uptake of Flu vaccination, the change in Public Health ring-fenced grant allocations from 2020-21 to 2021-22, and the number of Covid Vaccination Centres within the local authority. This stage used the Bayesian programme rstanarm 28 to obtain parameter estimates and credible intervals, and a plot of predicted and observed c i . Using the associated package "loo" 29 , a pointwise value of pareto_k > 0.7 was taken to indicate an outlier. 30 A Bayesian version of R 2 is used to describe overall model fit. 31 In fact the c i are not observed, but are an output of first stage modelling with associated standard errors. To estimate the impact of this uncertainty, simulated c* were drawn as multivariate normal with mean c (the vector with components c i ) and variance V c . Second stage modelling was repeated using the simulated c* to give fresh sampling output (4000 rows), and this process was repeated 100 times to produce a combined sampling matrix with 404,000 rows (including the original output from second stage modelling of the c i ). The mean, 5% and 95% quantiles of its columns were taken as corrected estimates and credible intervals of the second stage parameters, taking into account the uncertainty in the c i . . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; ML values and proportion of Deviance explained for the 4 models and 3 doses are shown in Table 1 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; expected. Newham 035 (Beckton Park), with the lowest uptake (0.223) in this local authority, nearly matches its predicted value (0.227). Overall, the model fits the data very well. Chart 2 shows selected smoothers. These plot the modelled impact of individual covariates on the overall fitted value (displayed on the scale of the linear predictor, before it is translated to fitted output by the inverse link function). The x-axis is limited by the 0.01 and 0.99 quantiles of the covariate. The first 6 smoothers shown have much larger impact on fitted values, and each of them has a negative effect: higher values of scaled IMD give lower fitted values, and likewise for the age bands and ethnicities shown. In the subsequent panels the y-axis is limited by the range of the smoother and is labelled "narrow scale". Thus an increasing male proportion also has a negative impact, but its effect is smaller than that of IMD, age, or ethnicity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; https://doi.org/10.1101/2022.02.01.22270236 doi: medRxiv preprint Chart 4 shows the parameter estimates and 90% credible intervals from second stage modelling of the scaled estimated random effects from model D for the 3 rd injection. Table 2 shows the parameter estimates and 90% credible intervals after simulating (x100) the random effects from all four models (applied to the 3 rd Injection) before second stage modelling. The second stage model fits increasingly well as the first stage model improves. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; https://doi.org/10.1101/2022.02.01.22270236 doi: medRxiv preprint Bayes R 2 improves when second stage modelling is applied to the random effects obtained from better first stage models (D is preferred to other models by ML value and leads to higher Bayes R 2 ). All four models lead to 2 nd stage models which show significant elevation (90% credible interval) for Flu vaccination and the rise in Public Health budgets as predictors of the simulated random effects, whilst the number of Vaccination Centres is only significant if the random effects are simulated from model C. North West Region is a significant negative predictor with all four models, and the North East is significant negative with model D. London is a significant positive predictor of random effects from models B and D. The South East is borderline positive with model D only. A comparable table for the 1 st dose shows that Bayes R 2 rises from 0.361 for modelling the estimated random effects from A, to 0.442 for B, 0.594 for C, and 0.593 for D. All four models lead to 2 nd stage models which show significant elevation (90% credible interval) for Flu vaccination and the rise in Public Health budgets as predictors of the simulated random effects, and likewise for London and South East regions, whilst North West is negative, and North East is negative for B C , C , and D. South West is positive for C and D. For the 2 nd dose, Bayes R 2 rises from 0.376 for modelling the estimated random effects from A, to 0.447 for B, 0.586 for C, and 0.587 for D. All four models lead to 2 nd stage models which show significant elevation (90% credible interval) for Flu vaccination and the rise in Public Health budgets as predictors of the simulated random effects, and likewise for London and South East regions, whilst North West is negative, and North East is negative for B C , C , and D. South West is positive for C and D, and borderline for A. Chart 5 shows the scaled estimated random effects from model D against the prediction from second stage modelling, for each dose. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint There is extensive literature on Covid-19 vaccine uptake. Before the vaccines were deployed, a telephone and web survey of attitudes, considered by the government's advisory group SAGE in December 2020 (see [4] ) found "marked differences existed by ethnicity, with Black ethnic groups the most likely to be COVID-19 vaccine hesitant followed by the Pakistani/Bangladeshi group. Other White ethnic groups (which includes Eastern European communities) also had higher levels of COVID-19 vaccine hesitancy than White UK/White Irish ethnicity". SAGE cited previous studies to conclude "Barriers to vaccine uptake include perception of risk, low confidence in the vaccine, distrust, access barriers, inconvenience, socio-demographic context and lack of endorsement, lack of vaccine offer or lack of communication from trusted providers and community leaders." Later, a meta-analysis of international studies by Qiang Wang (see [3] ) found that "Gender, educational level, influenza vaccination history, and trust in the government were strong predictors of COVID-19 vaccination willingness" and that healthworkers were less willing than the general public. A literature review of UK studies by Atiya Kamal (see [5] ) concluded "Ethnic minority status was associated with higher vaccine hesitancy and lower vaccine uptake compared with White British groups. Barriers included pre-existing mistrust of formal services, lack of information about the vaccine's safety, misinformation, inaccessible communications, and logistical issues". A strategy to overcome vaccine hesitancy was advocated by Mohammed S Razai (see [8] ), highlighting Confidence (importance, safety and efficacy of vaccines); Complacency (perception of low risk and low disease severity); Convenience (access issues dependent on the context, time and specific vaccine being offered); Communications (sources of information); and Context (sociodemographic characteristics). In an article entitled "What must be done to tackle vaccine hesitancy and barriers to COVID-19 vaccination in migrants?" (see [9] ) Alison Crawshaw highlighted "mistrust of the state and health system, stemming from historical events, data sharing policies and dissatisfaction with the initial handling of the pandemic" and advocated "engaging with communities to understand their concerns or barriers to vaccination and working together to co-develop tailored approaches to encourage uptake and rebuild trust". A cohort analysis by Helen Curtis et al. (see [7] ) of 57.9 million patients' primary care records, concerns the period from 8 December 2020 to 17 March 2021. Of patients aged ≥80 years not in a care home (JCVI group 2) 94.7% received a vaccine, but with substantial variation by ethnicity (White 96.2%, Black 68.3%) and deprivation (least deprived 96.6%, most deprived 90.7%). Patients with pre-existing medical conditions were more likely to be vaccinated with two exceptions: severe mental illness (89.5%) and learning disability (91.4%). January 2022) 32 shows survey data on the proportion receiving 3 vaccinations by occupation groupings, ranging from 80.4% (Health) to 48.0% (Elementary trades and related occupations); and for more specific occupations; and by ethnicity, ranging from 68.4% (White British) to 33.9% (Black Caribbean). An ONS technical article 33 explains the logistic regression models applied to vaccination status of individuals, controlling for sex, ethnicity, age, geographical region, urban or rural classification of their address, deprivation percentile, household size, whether the household was multigenerational. The model outputs 34 include regional variation, with negative impact of residence in the North West amongst persons aged 18 -34. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; An ONS study of ethnic contrasts in coronavirus death rates published on 26 January 2022 35 found that "Location, measures of disadvantage, occupation, living arrangements, pre-existing health conditions and vaccination status accounted for a large proportion of the excess rate of death involving COVID-19 in most ethnic minority groups; however, the Bangladeshi ethnic group and men from the Pakistani ethnic group remained at higher risk than White British people in the third wave, even after adjusting for vaccination status". A 2018 study of childhood immunisation rates in Italy by Veronica Toffolutti 36 , found that reductions in public health budgets were a significant predictor of falling immunisation rates. In the geographical analysis reported here, I initially sought to learn whether the current very low booster uptake in particular small areas was unexpected, or could be predicted from the national data on the UK Coronavirus Dashboard, at Middle Super Output level. In each MSOA, the data shows the number of people eligible for the vaccine, and the cumulative number actually vaccinated with the first and second doses, and the 3 rd Injection. The latter covers both people receiving the booster, and those with compromised immunity who were given a 3 rd primary dose. The ratio of those vaccinated to those eligible, is the uptake. But the actual numbers vaccinated and eligible give more information, and can be modelled as a binomial variable. The Generalized Additive Model, implemented in R with the mgcv package, is an established technique for non-linear modelling. The variance in this data exceeds what would be expected for a binomial variable, but can be estimated by relaxing the "binomial" assumption to "quasibinomial". Using demographic, employment and local health data as predictors, such models fit this data well. The models are improved by including a random effect assigned to the Upper Tier Local Authority within which the MSOA is found, allowing other aspects of the UTLA to influence the prediction. Four different models for each of the three vaccinations, all gave a good fit to the data, with the proportion of deviance explained ranging from 95.4% to 97.2%. Chart 1 shows how well the preferred model D performs for the 3 rd injection. This close fit means that the observed values in a particular MSOA are almost always near the prediction from the national data. The covariate structure was simplified to enable models to be computable on a PC. For example, whilst population data is available for each year of age, the models use broader categories such as aged 15-24 or 25-44. Even so, D has 33 smoothers, each with a smoothing parameter which must be estimated during fitting, along with 341 coefficients. The negative slope of the first smoother means that a higher IMD leads to a lower predicted uptake if all other variables are unchanged. Education Deprivation (a specific domain within the overall index IMD) also has a negative slope, although its effect is smaller (shown on Chart 2 with "narrow scale"). All six most powerful predictors have smoothers with negative slope. The dependence on age is to be expected, as the vaccine rollout targeted different age groups at different times. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; Ethnicity has significant impact on uptake, as many other studies have found. The smoother for the "Other White" group has the widest range, comparable to that for IMD, showing that the "Other White" proportion of the population has a strong impact on predicted uptake. Several other smoothers have small positive slope, such as those for employment in Public Adminstration, or Health and Social Care. The latter indicates that despite the well publicised vaccine hesitancy amongst a minority of NHS and care staff, the overall impact of employment in these sectors is to increase the uptake of the 3 rd Injection. This is consistent with the ONS data (see [32] ). For the 1 st and 2 nd doses, the increase is clear only when the proportion of Health and Care staff exceeds the 20 th percentile of this covariate. The smoother for average distance to A&E also has a positive slope, suggesting that people living further from an emergency department may be more concerned to take the vaccination. Likewise, increasing Health Deprivation is associated with increasing uptake. Conversely, increased GP registration is associated with decreasing uptake. However, there is some uncertainty in the GP registration data, and NHS Digital have commented on the excess of GP registrations over ONS population estimates. 37 Fitting any of these models gives estimates for the effect of each of the demographic, employment, and health variables included as predictors, and the estimated "random effect" of each of the 149 UTLA. For example "Knowsley" has a strong negative impact, and "Slough" has a strong positive impact. These effects are not to be confused with the actual uptake of vaccination in the local authority, which may be low or high due to deprivation or other fixed effects, and which may vary widely amongst MSOAs within the local authority. Uptake of the 3 rd Injection varied from 32.2% to 57.6% within Knowsley, from 17.3% to 65.5% in Liverpool, and from 36.9% to 75.8% in Sefton. The random effect alters the prediction from the fixed effects alone. If Beckton Park (in Newham) had been located in Slough with identical demography, employment, and health indicators, the predicted uptake of the 3 rd Injection would rise by 18%, whilst if it were in Knowsley the prediction would fall by 29%. The range in magnitude of the random effects is comparable to that of ethnicity groups. On the scale of the linear predictor, the smoothers for Other White, South Asian, and Afro-Caribbean groups range from 0.09 to -0.70, 0.05 to -0.42, and 0.07 to -0.49 respectively, each smoother descending as the ethnicity proportion rises. The random effects range from -0.15 (Knowsley) to 0.50 (Slough). The difference between the random effects for Knowsley and Slough is greater than the maximal difference between any two MSOA due to the population proportion of South Asian or Afro-Caribbean ethnicity, and only slightly less than the maximal difference due to the proportion of Other White. The second stage of modelling focused on the 149 random effects, seeking to explain their estimated values in terms of other information at UTLA level. This stage tested the impact of flu vaccination rates, public health budgets, vaccination centres, and Region, treated as a random factor. Ideally all of the UTLA and MSOA level variables would be incorporated in a single model. But that appeared prohibitively slow to compute, so the random effects estimates from the first stage mgcv modelling were considered as observations, to be modelled in their own right. Simulated random effects generated from the first stage model were used to refine the parameter estimates and credible intervals from the second stage, but this made little difference. For example the parameters and credible intervals for model D in Chart 4 are similar to those found for D after simulation in Table 2 . . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; The second stage model passes Bayesian checks and shows clear effects of Region, flu vaccination, and public health budget increase. Chart 5 shows the extent to which the chosen predictors actually explain the estimated random effects. None of the points are outliers (by pareto_k) and the model fits well for the 1 st and 2 nd doses with Bayes R 2 ~ 0.6. The lower value for the 3 rd injection, ~ 0.35, suggests that there may be other relevant covariates at UTLA level. It is striking that Knowsley, Liverpool and Sefton all appear at the lower left of Chart 5 and Knowsley is conspicuously low for all three vaccinations. The Merseyside local authorities are amongst the most deprived in England, but IMD already appears in the first stage model D as a fixed effect so was not expected to have any impact on the second stage model. Indeed, if IMD is averaged over the MSOA within each UTLA and then used as a predictor in second stage modelling, it has no significant effect. Simply being in the North West is the most powerful predictor of low uptake in this model, and in addition Knowsley had the 13 th lowest increase in public health funding (and the 3 rd lowest within the North West), rising by only 0.88%. Newham benefits from being in London and from the larger increase in its public health budget (rising by 2.09%), and possibly from the vaccination centre located in Olympic Park, an outlier with positive residual (see Chart 1). Flu vaccination rates are higher in Knowsley (49.2) than in Newham (45.5), so cannot explain the disparity in random effects. Slough benefits from being in the South East and possibly from a vaccination centre, whilst its flu vaccination rate and rise in public health budget are both close to their respective mean values. However the random effects for Slough also exceed prediction. The highest predicted value is in Lambeth, where the flu vaccination rate is below average, but which benefits from being in London with a 4.88% rise in public health budget and 3 vaccination centres. Its estimated random effect is close to prediction for all 3 doses. All such conclusions depend on the validity of the models. The first stage MSOA level modelling fits very well, and the resulting estimated random effects from all four models are highly correlated. That is, the random effects are not simply artefacts of the model, once ethnicity is included. The estimated random effects change dramatically if ethnicity is omitted from the model. The UTLA level effects could be described as a postcode lottery, as they are not explained by the population characteristics controlled for in the fixed effects, but are associated with other geographical factors. However, a "lottery" suggests pure chance whereas economic policy decisions affect Regional disparity and public health budgets, which in turn affect Covid vaccination rates. Much of the literature has focused on "vaccine hesitancy" of specific population subgroups. The modelling here confirms the impact of ethnicity along with deprivation and age, but also identifies additional factors, characteristic of the locality rather than the population living in it, consistent with evidence on childhood immunisations in Italy. These models indicate that whatever other barriers exist due to deprivation and within particular ethnicities, the annual change in local authority public health budgets is also a significant factor. Therefore, increasing local public health allocations would be one simple way to improve Covid vaccine uptake. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 1, 2022. ; https://doi.org/10.1101/2022.02.01.22270236 doi: medRxiv preprint NHS begins COVID-19 booster vaccination campaign. NHS England COVID-19) in the UK Vaccination against COVID-19: A systematic review and meta-analysis of acceptability and its predictors Factors influencing COVID-19 vaccine uptake among minority ethnic groups Ethnicity sub-group of the Scientific Advisory Group for Emergencies (SAGE) A Rapid Systematic Review of Factors Influencing COVID-19 Vaccination Uptake in Minority Ethnic Groups in the UK. Atiya Kamal, Ava Hodson Factors affecting COVID-19 vaccination acceptance and uptake among the general public: a living behavioural science evidence synthesis (v1.0 Trends and clinical characteristics of COVID-19 vaccine recipients: a federated analysis of 57.9 million patients' primary care records in situ using OpenSAFELY COVID-19 vaccine hesitancy: the five Cs to tackle behavioural and sociodemographic factors What must be done to tackle vaccine hesitancy and barriers to COVID-19 vaccination in migrants? COVID-19) in the UK. Download data English indices of deprivation 2019. Ministry of Housing, Communities & Local Government Lower layer Super Output Area population density (National Statistics) Middle Super Output Area population estimates (supporting information) Ethnic group by sex by age. DC2101EW. Census 2011 NOMIS official labour market statistics Industry by sex by age. DC6110EW. Census 2011 NOMIS official labour market statistics 17 DC1109EW -Household composition by age by sex NOMIS official labour market statistics Patients Registered at a GP Practice Seasonal influenza vaccine uptake amongst GP Patients in England. End of season data for 1 Vaccination sites as of 17 Public health ring-fenced grant 2021 to 2022: local authority circular mgcv: Mixed GAM Computation Vehicle with Automatic Smoothness Estimation Simon Wood Generalized Additive Models: An Introduction with R Smoothing parameter and model selection for general smooth models loo: Efficient leave-one-out cross-validation and WAIC for Bayesian models Aki Vehtari, Jonah Gabry, Mans Magnusson, Yuling Yao Compute a Bayesian version of R-squared or LOO-adjusted R-squared for regression models COVID-19) Infection Survey technical article: Analysis of characteristics associated with vaccination uptake. Office for National Statistics COVID-19) Infection Survey technical article: analysis of characteristics associated with vaccination uptake. 15 November edition of this dataset. Office for National Statistics Office for National Statistics Austerity, measles and mandatory vaccination: crossregional analysis of vaccination in Italy 2000-14 Thanks to David Taylor Robinson for helpful comments including the AHAH dataset and the suggestion that budgets could be a predictor, and to Isabelle Whelan for an unpublished essay "Austerity, NHS reform and vaccine uptake", written before the pandemic.