Efficacy or delivery? An online Discrete Choice Experiment to explore preferences for COVID-19 vaccines in the UK.

COVID-19 vaccines are widely regarded as an integral component in the UK's pandemic recovery, and a comprehensive distribution strategy will be required to maximise uptake. However, to date, there is a dearth of research into factors that could lead to UK residents' acceptance or rejection of COVID-19 vaccines. This study used a discrete choice experiment to investigate the importance of vaccine properties, delivery and media coverage in amplifying or attenuating vaccine uptake. Efficacy was found to be the factor that most influenced vaccine selection; further, the positive effect of high efficacy was more pronounced for those aged 55+. Insights from this DCE aim to assist policymakers and public health communicators in planning and refining their delivery strategy for COVID-19 vaccines.

Introduction

The COVID-19 pandemic has had an immensely deleterious effect on the United Kingdom (UK). At the time of writing, there have been more than 1,000,000 positive cases and over 80,000 people have died within 28 days of a positive COVID-19 test (Gov.uk, 2021).

Widespread uptake of an efficacious vaccine will be a key facilitator of the UK’s adaptation to, and ultimately recovery from, the pandemic (Peirus and Leung, 2020). Evidence of how UK residents will respond to different COVID-19 vaccine options is therefore now needed to optimise delivery.

Several recent studies have asked UK residents directly about their behavioural intentions regarding COVID-19 vaccination. Limitations of this approach have been well documented, including the susceptibility of survey responses to social desirability bias (Malik et al., 2020). More generally, Faries (2016) demonstrated that stated intention only predicts 30%–40% of the variance in health behaviour.

Discrete choice experiments (DCEs), however, have been shown to be a robust method for predicting behaviours and preferences across a range of economics fields. They have been widely used in health economics to analyse trade-offs between patient experience and outcomes, health professionals’ treatment preferences and other applications (Clark et al., 2014); further, they have accurately predicted positive vaccination behaviour at a rate exceeding 80% (Lambooij et al., 2015).

However, DCEs have rarely been used in a rapidly evolving environment where the impacts of the situation are universal and the need for action is urgent, such as the COVID-19 pandemic. To date, there has not been a DCE involving a representative sample of UK residents which explicitly aims to enhance the design and implementation of the UK’s COVID-19 vaccination programme.

This paper therefore not only provides robust evidence regarding population preferences for UK policy-makers; it demonstrates the practical value of applying an established tool for economic analysis in a live policy environment.

Methodology

Sample

This DCE was conducted online, involving n=1,501 participants from Kantar’s LifePoints panel. Fieldwork was conducted from 27 August - 3 September 2020, and the average completion time was 10.1 min.

Parallel quotas on key demographics – including age, gender and geographical region – were enforced (based on mid-year population estimates from the UK’s Office for National Statistics (2019)). As these quotas were precisely met in field (see Table 1), data were unweighted.

Table 1

Sample composition according to gender, age and region of the UK.

Gender	%	n=	Age	%	n=	Region	%	n=
Male	49%	736	16–34	31%	466	North (including Scotland, N.E. & N.W. England, Yorkshire/Humberside)	31%	466
Female	51%	765	35–54	32%	480	Midlands (including Wales, E. & W. Midlands, E. England)	30%	450
			55＋	37%	555	South (including London, S.E. & S.W. England)	36%	540
						Northern Ireland	3%	45

Experiment design

One consequence of operating in a live policy environment was the need for rapid, responsive development of the DCE — drawing on the best evidence available to ensure that results could inform distribution strategy. As such, rather than conducting primary qualitative research to determine the DCE’s attributes and levels, selection was based on a review of similar vaccine DCEs, including those informed by focus groups (Determann et al., 2014, Determann et al., 2016, Dong et al., 2020).

As in other studies where secondary data were used for design (see Hoogink et al., 2020), the most influential attributes from the reference DCEs were chosen for inclusion; an additional UK COVID-19 programme-specific attribute – administration location – was also incorporated (see Table 2).

Table 2

DCE attributes and levelsb.

Attribute	Levels
Level of protection offered	1	50%
	2	70%a
	3	90%
Recommender of the vaccine	1	GPa
	2	NHS
Number of doses needed for full protection	1	Onea
	2	Two
Location in which the vaccine is administered	1	Local GP surgery a
	2	Mobile vaccination unit
Coverage in the media	1	Positive coverage in newspapers, television and radio a
	2	Positive coverage on WhatsApp, blogs and social media

Reference category.

The full DCE introduction is appended, please see Appendix.

A full factorial design would have involved presenting 48 vaccine profiles (24 31), making implementation unfeasible in a short online survey. Therefore, a rotation design – developed from an orthogonal main-effects array – was generated using the Support.CE package in R statistical software (Aizaki, 2015). A design with a relatively small number of blocks (two) minimised the possibility of imbalance between blocks following administration (Hensher et al., 2015).

The final design comprised 12 paired scenarios, across two blocks of 6 pairs. In each pair, an opt-out (neither vaccine) option was presented to maximise external validity. As recommended in the literature, D–efficiency was compared against a competing design (Mangham et al., 2008).

An example paired choice scenario can be seen in Fig. 1.

Fig. 1

Example choice scenario.

Data analysis

A clustered conditional logit model – run via the Survival package in R statistical software (Therneau, 2015) – was used in analysis. Selection of a COVID-19 vaccine based on its attributes – the main effects model – is written as:

Where:

represents the th vaccine chosen by participant n;

represents the alternative specific constant (ASC), denoting likelihood of vaccine selection relative to the opt-out ‘status quo’ (specified using dummy coding);

represent the effects of the vaccine attributes upon selection (specified using dummy coding); and

represents the random error term (the non-observable component of choices).

In conditional regression models, one of the primary assumptions is the independence of irrelevant alternatives. The Hausman and McFadden (1984) test was used to verify that this assumption was met.

A hybrid logit model examining the interaction between older age (55+ years) and DCE attributes was also run. This model (henceforth referred to as Model 2) was used to determine if the preferences of older residents – who are highly vulnerable to the virus (Yanez et al., 2020) – differ from those of the general population.

Results

Logit models

Results for the models can be seen in Table 3.

Table 3

Results of conditional logit main effect and age interaction models.

Attribute	Levels	Model 1: Main effects			Model 2: Age interaction
		Coeff	Odds ratio	SE (coeff)	Coeff	Odds ratio	SE (coeff)
ASC		1.42⁎⁎⁎	4.12	0.05	1.19⁎⁎⁎	3.27	0.06
Level of protection offered	50%	−1.30⁎⁎⁎	0.27	0.04	−1.11⁎⁎⁎	0.33	0.05
	90%	1.03⁎⁎⁎	2.80	0.03	0.92⁎⁎⁎	2.50	0.04
Recommender of the vaccine	NHS	−0.07⁎⁎	0.94	0.03	−0.04⁎⁎⁎	0.96	0.03
Number of doses needed for full protection	Two	−0.15⁎⁎⁎	0.86	0.03	−0.17⁎⁎⁎	0.85	0.03
Location in which the vaccine is administered	Mobile vaccination unit	−0.21⁎⁎⁎	0.81	0.03	−0.17⁎⁎⁎	0.84	0.03
Coverage in the media	Positive coverage on WhatsApp, blogs and social media	−0.17⁎⁎⁎	0.83	0.03	−0.15⁎⁎⁎	0.86	0.03
ASC * 55＋			–		0.74⁎⁎⁎	2.09	0.10
Level of protection offered* 55＋	50% * 55＋				−0.61⁎⁎⁎	0.54	0.10
	90% * 55＋				0.36⁎⁎⁎	1.44	0.08
Recommender of the vaccine * 55＋	NHS * 55＋				−0.08⁎⁎⁎	0.92	0.06
Number of doses needed for full protection * 55＋	Two * 55＋				0.07	1.07	0.06
Location in which the vaccine is administered* 55＋	Mobile vaccination unit * 55＋				−0.12	0.89	0.06
Coverage in the media * 55＋	Positive coverage on WhatsApp, blogs and social media * 55＋				−0.14⁎	0.87	0.06

.

.

.

The significant ASC (odds ratio 4.12, p < 0.001) indicates a population-level preference for vaccine selection over ‘opting out’. Efficacy was by far the most influential characteristic in determining selection of COVID-19 vaccines, as evidenced by the high odds ratio of selection of a vaccine with 90% efficacy relative to a vaccine with 70% efficacy (odds ratio 2.80, p < 0.001). Location of vaccine administration was also important: the local GP surgery was preferred over mobile vaccination units (odds ratio = 0.81, p<0.001).

Similar but more pronounced results were observed for those aged 55+ years in Model 2. Specifically, highly significant interactions were observed with the ASC (odds ratio 2.09, p < 0.001), efficacy (90% odds ratio 1.44, p < 0.001) and recommender (NHS odds ratio 0.92, p < 0.001).

.

.

.

Opt-outs

Fig. 2 illustrates the proportion of people who opted out of any of the pairs by selecting the ‘neither vaccine’ option. More than three quarters (77%) of participants selected one of the two vaccines in all six of the pairs; 7% opted out of all six of the pairs; and the remainder opted out at least once.

Fig. 2

Rate of opt-out of DCE scenarios, by age.

As would be expected given the interactions in Model 2, opt-out behaviour differed significantly according to age. Eighty-six percent of those aged 55+ selected a vaccine in all six scenarios, a proportion significantly higher than that observed for those aged 16–34 (65%; , p < 0.001) and 35–54 (77% , p < 0.001).

Discussion and conclusions

This is the first DCE in which the preferences of UK residents regarding COVID-19 vaccines are explored. The results underscore the importance of characteristics of both the vaccines and their distribution programmes in amplifying or attenuating uptake among UK residents. Positively, results highlight a population preference for vaccination against COVID-19 over the non-vaccination ‘status quo’; this preference was more pronounced among residents aged 55+.

This study also clearly demonstrates that efficacy levels of COVID-19 vaccines are central to their appeal: the odds that a vaccine with 90% efficacy was chosen were 2.80 times the odds that a vaccine with 70% efficacy was chosen. In Model 2, older age (55+) interacted significantly with efficacy level, indicating that high efficacy acts as an even greater inducement to older, more vulnerable residents.

Viewed within the context of the UK distribution programme (Department of Health and Social Care, 2021), these results are encouraging. Two of the vaccines approved for use – the BioNTech/Pfizer (Polack et al., 2020) and Moderna vaccines (Baden et al., 2020) – possess efficacy rates exceeding 90%, suggesting that many residents will choose to receive them without further encouragement. However, given the programme also encompasses a vaccine with lower efficacy which will be more widely distributed – the Oxford/AstraZeneca vaccine (Voysey et al., 2020) – there may still be some reluctance. Consequently, distinct distributional, behavioural or communication strategies may be required to augment population-level uptake.

The results of this study are largely congruent with other recent vaccine DCEs. As observed elsewhere, in addition to efficacy, number of doses and source of recommendation are influential (see Dong et al., 2020, Kreps et al., 2020). However, in the present study, location of administration was included, with the expectation that the need for rapid and widespread distribution would likely result in a range of location types. This attribute was unique to this study, and proved to be the second most influential in determining vaccine selection, with local GP surgeries preferred over mobile centres. In practical terms, the result signifies that local vaccination services (such as in GP surgeries) will be favoured by many residents across the UK, rendering capacity a potential issue.

This study demonstrates that it is possible to expedite the design of a DCE to ensure results can inform strategy, by determining attributes prior to the detail of these features becoming clear. However, this approach limits the extent to which the effects of specific features of the vaccines and their delivery programmes can be analysed. Future DCEs and other research should therefore seek to build on this study’s results, examining take-up in greater depth. Follow-up work should also aim to increase understanding of the public’s knowledge and perceptions of vaccine and programme characteristics via primary quantitative and qualitative research.

Nonetheless, the present study provides clear evidence for the utility of DCEs in formulating vaccination programmes, and for their use in fast-changing environments. The results may assist UK policymakers and public health communicators in preparing and refining the COVID-19 vaccine delivery plan.