Modelling the population effectiveness of the national seasonal influenza vaccination programme in Scotland: The impact of targeting all individuals aged 65 years and over

Background For the last 17 years, the UK has employed a routine influenza vaccination programme with the aim of reducing the spread of seasonal influenza. In mid‐2000, the programme moved from a purely risk‐based approach to a risk and age group‐targeted approach with all those aged 65+ years being included. To date, there has been no assessment of the population effectiveness of this age‐targeted policy in Scotland. Objectives Statistical modelling techniques were used to determine what impact the routine vaccination of those aged 65+ years has had on influenza‐related morbidity and mortality in Scotland. Methods Two Poisson regression models were developed using weekly counts of all‐cause mortality, cause‐specific mortality and emergency hospitalisations for the period 1981‐2012, one using week‐in‐year and the other using temperature to capture the seasonal variability in mortality/hospitalisations. These models were used to determine the number of excess deaths/hospitalisations associated with the introduction of the local risk and age‐based vaccination programme in 2000. Results Routinely vaccinating those aged 65+ years is associated with a reduction in excess all‐cause mortality, cardiovascular and COPD‐related mortality and COPD‐related hospitalisations. Our analysis suggests that using the week‐in‐year model, on average, 732 (95% CI 66‐1398) deaths from all causes, 248 (95% CI 10‐486) cardiovascular‐related deaths, 123 (95% CI 28‐218) COPD‐related deaths and 425 (95% CI 258‐592) COPD‐related hospitalisations have been prevented each flu season among the those aged 65+. Similar results were found using the temperature model. There was no evidence to suggest that the change in policy was associated with reductions in influenza/pneumonia‐related mortality or influenza/cardiovascular‐related hospitalisations. Conclusions Routinely vaccinating those aged 65+ years appears to have reduced influenza‐related morbidity and mortality in Scotland. With the childhood vaccination programme well underway, these data provide an importance benchmark which can be used to accurately assess the impact of this new seasonal influenza vaccination programme.


Funding information Health Protection Scotland; Scottish Government
Background: For the last 17 years, the UK has employed a routine influenza vaccination programme with the aim of reducing the spread of seasonal influenza. In mid-2000, the programme moved from a purely risk-based approach to a risk and age group-targeted approach with all those aged 65+ years being included. To date, there has been no assessment of the population effectiveness of this age-targeted policy in Scotland.
Objectives: Statistical modelling techniques were used to determine what impact the routine vaccination of those aged 65+ years has had on influenza-related morbidity and mortality in Scotland.
Methods: Two Poisson regression models were developed using weekly counts of all-cause mortality, cause-specific mortality and emergency hospitalisations for the period 1981-2012, one using week-in-year and the other using temperature to capture the seasonal variability in mortality/hospitalisations. These models were used to determine the number of excess deaths/hospitalisations associated with the introduction of the local risk and age-based vaccination programme in 2000.

Results:
Routinely vaccinating those aged 65+ years is associated with a reduction in excess all-cause mortality, cardiovascular and COPD-related mortality and COPDrelated hospitalisations. Our analysis suggests that using the week-in-year model, on average, 732 (95% CI 66-1398) deaths from all causes, 248 (95% CI  cardiovascular-related deaths, 123 (95% CI 28-218) COPD-related deaths and 425 (95% CI 258-592) COPD-related hospitalisations have been prevented each flu season among the those aged 65+. Similar results were found using the temperature model. There was no evidence to suggest that the change in policy was associated with reductions in influenza/pneumonia-related mortality or influenza/cardiovascularrelated hospitalisations.
Conclusions: Routinely vaccinating those aged 65+ years appears to have reduced influenza-related morbidity and mortality in Scotland. With the childhood vaccination programme well underway, these data provide an importance benchmark which can be used to accurately assess the impact of this new seasonal influenza vaccination programme.

| INTRODUC TI ON
Every winter, the UK experiences a seasonal influenza epidemic that affects the morbidity and mortality of thousands of its citizens and puts increased pressure on NHS resources. 1,2 For healthy individuals, influenza is a self-limiting, though debilitating, illness from which full recovery is usually attained within 2-7 days. 2,3 There are some subsets of the population, however, who have been shown to have a higher risk of influenza associated morbidity and mortality. [4][5][6][7][8][9] Although the extent to which influenza increases morbidity and mortality varies from year to year, the virus continues to impose a considerable economic burden on society. 10,11 As a result, seasonal influenza epidemics are considered to be a significant annual public health threat.
Since the late 1960s, the UK has sought to limit the healthcare burden associated with influenza via a national vaccination programme. The programme initially targeted individuals who were at highest risk of influenza-related morbidity and mortality and was continually expanded to incorporate a wider range of risk groups. [2][3][4] In 2000, the programme was extended to include all persons aged 65+ years, moving the UK's vaccination policy from a risk based to a risk and age-based strategy. More recently, in 2012, the Joint Committee on Vaccination and Immunisation (JCVI) recommended the inclusion of those aged 2-17 years in the routine vaccination programme by offering intranasally administered live cold-adapted influenza vaccine (LAIV)-Fluenz. 12 Each of the constituent countries within the United Kingdom endorsed this JCVI recommendation which has since become government policy. The phased introduction of this new extension began in 2013 with the routine immunisation of children aged 2-3 years with extensions to incorporate older age groups in subsequent years. 13,14 Although healthy children are less likely to experience severe influenza-related morbidity and mortality, 15 they are two to three times more likely to be ill with influenza 14 and are a well-documented transmitter of the virus. 16,17 Previous transmission modelling and cost-effectiveness studies suggest the JCVI's recommendations will increase the overall efficiency of the influenza vaccination programme 18 and offer a highly cost-effective way of providing this risk group with direct protection against the impact of flu. 19 The reduction in flu circulation resulting from the vaccination of children should offer indirect protection to older adults and those with clinical risk factors, therefore reducing the number of severe flu cases and flu-related deaths in this subset of the population. [20][21][22][23] A clear understanding of how changes to the UK's vaccination programme has affected influenza-related morbidity and mortality is key to providing a benchmark from which policymakers can assess the success of the vaccination programme. While there is evidence to suggest that the influenza vaccine is effective at preventing influenza illness and its complications, 24-30 the population effectiveness of the UK vaccination programme is not fully understood. While previous modelling work in England suggests that the vaccination programme, which targets clinical risk groups as well as individuals aged 65+ years, is cost-effective 24 and associated with a reduction in the incidence of laboratory confirmed influenza illness 24 as well as reduced levels of pneumonia and influenza-related mortality, 31  The aim of this analysis was to use routine influenza surveillance data collected in Scotland to develop models that can be used to evaluate the population effectiveness of the national seasonal influenza vaccination programme. In particular, we are going to estimate the number of excess deaths/hospitalisations that may be been prevented by the mid-2000 policy recommendation to routinely vaccinate all persons aged 65+ years.

| Data sources
We use yearly mid-year estimates of population size along with data on mortality and emergency hospitalisations for this analysis.
Weekly counts of all-cause mortality, cause-specific mortality and Mid-year estimates of population size for the period of interest were obtained from the National Records of Scotland. 33 All of these data K E Y W O R D S effectiveness, influenza, modelling, population, statistics, vaccine were broken down by gender (male or female) and age category (0-1, 2-4, 5-12, 13-17, 18-44, 45-64, 65-74, 75-84 and 85+ years).
Weekly spatial temperature data for the geographical centre of Scotland are used to account for the direct effect on temperature on the seasonal variation in mortality and hospitalisations. We use data obtained from the British Atmospheric Data Centre 34 along with Shepard's inverse distance weighting methods 35 to estimate weekly minimum and average air temperatures for the period of interest.

| Modelling the number of deaths/ hospitalisations for a given week in the year
We use generalised additive models (GAMs) to derive Poisson regression models to estimate the expected number of deaths/hospitalisations for a given week in the year. 36,37 We use two different GAMs in this analysis. The first GAM as- week as a proportion of the number of weeks in that year, and minimum temperature is a continuous variable that denotes the weekly minimum air temperature for the geographical centre of Scotland.
Temperature severity is a categorical variable with five levels (0-4) as shown in Table 1 that records how mild or severe the weather was during a particular flu season. The cut-off values of −0.6°, −1.27° and −2.5° correspond to the 1st, 2nd and 3rd quartiles of the data set that contained the negative weekly minimum air temperatures.
We find that the best fitting WIY GAM is a function of year, week-in-year, age category and gender that models the number of deaths/hospitalisations in a given week t, for age group a and sex s (Count tas ) using the following equation: where f i denotes a cubic regression spline. Similarly, we find that the best fitting ST GAM is a function of year, minimum temperature, gender, age category and temperature severity that models the number of deaths/hospitalisations using the following equation: Both GAMs model the general trend of decreasing mortality over time using separate cubic regression splines for each gender, with 10-15 knot points (equivalent to one every 2-3 years), and the seasonal variability in mortality/hospitalisations using a cyclic cubic regression spline with 20 knot points (equivalent to one every 2.5 weeks).
The WIY (ST) model uses week-in-year (minimum temperature) to capture the seasonal variability in mortality. We follow the general purpose approach described in the documentation for the R package All the analysis is carried out using R version 3.10, 39 and overdispersion is tested for using the over-dispersion test provided by the qcc R package. 40 A 5% significance level is used throughout, and 95% confidence intervals are based upon a normal distribution.
TA B L E 1 Cut-off values used to define the temperature severity variables. The cut-off values of −0.6°, −1.27° and −2.5° correspond to the 1st, 2nd and 3rd quartiles of the dataset that contained the negative weekly minimum air temperatures

Severity value
Minimum air temperature (x)

| Sensitivity analysis
We conduct a sensitivity analysis using the ST model for those aged 65+ years to investigate how our model fits and estimates for the number deaths/hospitalisations prevented are affected when we (i) use minimum instead of average air temperature to define seasonality, and (ii) allow the flu season to vary from year to year corresponding to the national influenza surveillance data on consultation rates (reports per 100 000 population) from Health Protection Scotland.
Each flu season starts when the consultation rate exceeds 50 and ends when it falls below 50. From Figure 1, we can see that although both models are able to capture the long-term trends and the seasonal variation in the data, there are a number of instances where the models are not able to predict the number of deaths observed in moderate/severe influenza epidemics (e.g. 1989-1990 and 1999-2000). Excess all-cause mortality for each flu season is shown in Figure 2 where it is evident that, prior to the policy change in 2000, there are instances where the model over-predicts (negative excess all-cause mortality) and underpredicts (positive excess all-cause mortality). This is a consequence of using both winter and summer data to fit the models prior to 2000.

| All-cause mortality
After the policy change, however, there are few instances of underprediction and many more instances of over-prediction, suggesting that excess mortality has decreased in the period following the change in vaccination policy. It is worth noting that the model fits for the ST model were slightly better than those for the WIY model.
There is evidence to suggest that there has been a significant reduction in excess all-cause mortality among those aged 65+ years .020; .020, respectively). The greatest reductions have been among those aged 85 and over while the smallest reductions have been among those aged 65-74 (Table 2).

| Cause-specific mortality
Among those aged 65+ years, mortality attributable to influenza, pneumonia, cardiovascular disease, COPD and trauma were highly variable. Influenza-related mortality ranged from 0 to 139 deaths The model predictions for the 65+ grouping are presented in Table 3. As expected, there is no evidence of a reduction in trauma-related mortality following the change in vaccination policy. Furthermore, there is no evidence of a reduction in influenza and pneumonia-related mortality. Our analysis does, however, find evidence to suggest that there has been a significant reduc-

| Sensitivity analysis
Our sensitivity analyses results are presented in Table 5. For all cases, changing the temperature measure to minimum temperature results in a decrease in model fit. Improved model fits were observed for influenza and pneumonia-related mortalities as well as influenza and cardiovascular-related hospitalisations when we allowed the flu season to vary year on year. These improved models result in increases in the number of (i) influenza-related deaths prevented (from 15 to 24), (ii) pneumonia-related deaths prevented (from 85 to 96), and (iii) cardiovascular-related hospitalisations prevented (from 52 to 56). It is worth noting that the changes in model fits did not alter the statistical significance of any reductions in mortality/hospitalisations. These sensitivity analyses did not produce any substantial changes to the impact of the vaccination campaign.

| CON CLUS I ON S AND D ISCUSS I ON
For the last 17 years, the UK has employed a routine influenza vaccination programme with the aim of reducing the spread of seasonal influenza. 41 The programme has been extended to allow for the introduction of new risk groups and, in mid-2000, moved from a purely risk-based approach to a risk and age group-targeted approach with all those aged 65+ years being included. 42 The 2012 JCVI recommendations to include 2-to 17-year-olds 12 is a major undertaking that will require a substantial increase in NHS resources. To date, there has been no assessment of the population effectiveness of the age-targeted policy introduced in Scotland in 2000. Understanding how that policy extension has affected influenza-related morbidity and mortality is key to understanding the effectiveness of the current extension.
In this article, we have used routine influenza surveillance data It is believed that this study is the only one to analyse the population impact of the influenza vaccine on several causes of mortality/ hospitalisations. Since 2008, there have been a number of groups looking at the annual effectiveness of the seasonal influenza vaccine, [43][44][45][46] using mainly a test negative case-control study design, and these are now used to inform the WHO strain selection meetings each year.
That said, previous studies that have examined the impact on mortality of routinely vaccinating those aged 65+ years have produced contradictory results. Studies in the USA and Italy found no evidence of a reduction in all-cause mortality among the over 65s, 47,48 while a study in Holland did find evidence of a reduction. 49 The failure to find evidence in the USA and Italian studies may be due to a limited period with sufficiently high vaccine coverage. 31      Furthermore, the trends in our data may be associated with behavioural changes like reduced rates of smoking (smoking may predispose the individual to increased complication from influenza or other respiratory infections more evident in the winter). The inability to account for these other features in our model may lead to an overestimation of the amount of morbidity and mortality that has been prevented by the UK's influenza vaccination programme.
Previous analyses of excess winter deaths have use Serfling-type models based upon sine and cosine terms to model the cyclical pattern to mortality within a year. 31,50,51 With such models, where the data used to fit the model come from the summer months only, the excess deaths in winter are the difference between the observed and the deaths predicted assuming that the spring summer autumn seasonal trend carries on into winter. Such models are not appropriate for our work as (i) there is always likely to be an excess each season and (ii) they do not properly fit the winter peak. To achieve the latter, it is more appropriate to use either more harmonic terms in a Serfling-type model or to use GAMS which have greater flexibility to fit to the winter peak. We use GAMS because we assess the impact of the vaccination campaign by fitting the model to all the pre-vaccination data, both summer and winter, so that we have a model which, when projected into the future gives predictions of the number of deaths in winter, assuming that there had not been an intervention-the age-related vaccination campaign. We also fit to the summer data in the intervention period so that we can fit to general trends of decreasing mortality in the intervention period. Failure to do so would mean extrapolating the general decreasing trends from 1981 to 1999 10-13 years into the future. This is not optimal as the trends are unlikely to continue up to 10-13 years into the future.
In conclusion, our work suggests that influenza-related morbidity and mortality among those aged 65+ years has reduced following the decision to extend the UK's vaccination policy to incorporate all persons aged 65 and over. Given the recent recommendations to extend the programme to all those aged 2-17, these results provide healthcare professionals and policymakers with an important benchmark from which to assess the success of the most recent recommendation. Furthermore, the methods used in this analysis provide a framework that can be used to analyse the population effectiveness of the new influenza vaccination programme once the 2-to 17-year-old extension has been implemented.

ACK N OWLED G EM ENTS
We would like to thank Health Protection Scotland and the Scottish