Background
Innovation implementation (the process of integrating research findings into behaviours at the level of adopters) happens ‘in context’. Context in many healthcare systems includes scarce (or at least finite) resources, variability in adoption of existing innovations, and ways of changing behaviour that often incur their own costs but are rarely factored into the final estimate of the cost effectiveness of innovation adoption [
1,
2].
One key element of current implementation context is the growing volume of innovation that health system decision and policy makers are compelled to consider for implementation. By way of example, General Practitioners in the UK face up to 30 new pieces of guidance per month, far more than can feasibly be adopted by a multidisciplinary team, practice or clinic [
3]. Faced with scarce resources and increasing demand [
4], systems must prioritize and decide which innovations to implement. Guidance on how prioritization should be undertaken by potential adopters is scarce.
Policy makers have resorted to economic criteria, such as cost-effectiveness, to help decide which innovations should be available for services to consider. However, ranking innovations on the basis of economic attributes such as cost-effectiveness employed by NICE or program budgeting and marginal analysis [
5] misses the role that other factors play in the choice to adopt or not at service purchaser or provider levels [
4,
6,
7].
Influential theories of innovation diffusion generally [
7] and healthcare specifically [
4,
6] suggest multiple general determinants of adoption behaviour. Some theorists [
6] identify a large number of possible determinants: the characteristics of the innovation; system antecedents (structure of the organisation, absorptive capacity for new knowledge, and receptive context for change); system readiness; characteristics of adopters; communication and influence and a wider ‘outer’ context of politics and structures. Other models adopt a more parsimonious approach that focuses on the characteristics of innovations and prior conditions [
7]. The attitudes and values held by potential adopters exist as variables in almost all theoretical models of innovation adoption and are a focus within the field of implementation science [
8,
9].
Clearly, attitudes are influenced by many factors and are unlikely to be decisive in themselves in determining whether an innovation is adopted or behaviour changed. Indeed, some evidence indicates that, in specific healthcare contexts, compatibility of guideline attributes with clinicians’ values can be negatively related to desired behaviour change. Foy et al. tested the influence of attributes of clinical recommendations on compliance with recommendations using a pre- and post-intervention research design [
10]. Foy et al. found that while guideline-norm ‘fit’ and guideline compliance before and after a behaviour change intervention (audit and feedback) were positively related, guideline recommendations seen as incompatible with clinician norms showed greater change following the intervention [
10]. Foy and colleagues’ pre-post design was limited by the absence of data points within the intervention period, thus the effects of the intervention on compliance behaviour during the intervention period are unknown [
10]. Further, since Foy et al. used a qualitative approach to establishing innovation attributes (
i.
e., using focus groups and interviews), it is possible that different experts might have described different attributes.
We sought to use clinicians’ values in a slightly different way. Rather than evaluating their relationship with the effect of the behaviour change intervention, we built them into the selection of an innovation to be implemented. This study forms the initial phase in developing a targeted implementation strategy. In the subsequent phase, we examine six other determinants in order to tailor a multi-faceted implementation strategy to the barriers associated with each factor [
11]. Our rationale was a simple one: mindfulness of a ‘fit’ between a potential innovation’s characteristics and potential adopter values and norms would lead to focusing on an innovation for implementation that has (at least) a ‘fighting chance’ of adoption. Omitting to address the value clinicians assign to the attributes of an innovation theoretically lowers the chances of the innovation being adopted [
7].
One technique for understanding the value of a product’s attributes is Conjoint Analysis. CA is a stated preference method with its roots in mathematical psychology [
12,
13] and Lancaster’s theory of value [
14]. Respondents are asked to choose between hypothetical products (or innovations) with different levels of a limited number of attributes. The findings can then be used to rank competing innovations whose levels of these attributes have been scored in advance. CA makes two assumptions: that innovations can be described according to their attributes, and that the value of an innovation (to an individual) is a product of these collective attributes [
15]. It has the advantage of simultaneously estimating the value/utility placed on a product/service (or innovation) while also identifying the relative (to other attributes) importance of the attributes making up the innovation. CA also describes the extent to which individuals are willing to trade off one attribute to gain another (for example, cost vs. quality). It has been applied to many areas: market research [
16], private sector environmental and transport economics, public service redesign and planning [
17]. Specific health applications include in vitro fertilization [
18], orthodontic services [
14,
19], and liver transplantation [
20]. CA has not been widely used in implementation research and to the best of our knowledge has never been used to prioritize possible targets for implementation efforts.
To demonstrate how this technique can be successfully applied to the prioritization of healthcare innovations, we present an application of Conjoint Analysis to the implementation of innovations for women with postnatal depression in one UK NHS Primary Care Trust (PCT). The results are not intended to be generalizable to other settings; they are presented here to illustrate the application of CA to a key stage in the implementation process and the challenges involved.
Discussion
Using the results of a local survey conducted in a Primary Care Trust, this paper describes one solution to the challenge of incorporating clinician preferences into the prioritization of innovations in healthcare systems in which resources are finite and limited. In doing so, this study demonstrates the feasibility of the application of Conjoint Analysis to implementation. The analysis provided three things:
1.
The importance of the attributes of innovations generally;
2.
Their importance at various levels;
3.
And a ranked picture of innovations according to the preferences of the people involved in having to implement them.
There are other widely used methods for prioritizing innovations for adoption or investment. Criteria such as multi-criteria decision analysis [
31], budget impact analysis [
32], and cost-consequence analysis [
33] are all useful. However, the conjoint analytic approach enabled us to consider the innovation options most likely to ‘fit’ local preferences before we developed an implementation strategy that went on to measure and target other determinants of innovation adoption. Improving the ‘fit’ with local values as well as mapping and targeting the multitude of other determinants, should – in theory – increase the efficiency of the eventual implementation strategy [
6]. A more efficient implementation strategy will, all things being equal, reduce the costs of behaviour change approaches and thus increase ‘policy cost effectiveness’ –
i.
e., an estimate of costs vs. impact that takes into account the costs of changing behaviour as well as the cost effectiveness of the innovation itself [
1].
By applying Conjoint Analysis to healthcare innovation preferences for implementation, we were able to provide a visible rationale for the decision of which innovation to invest scarce time, money and human resources on. To the best of our knowledge, this is the first time CA has been used in implementation science in this way. Previous applications of CA to service improvement, design or planning have primarily been designed to inform single services (for example, [
18,
19]). While applying CA to single service design or improvement may be valuable, it may not be efficient. Because CA utilities relate to the attributes of products or services, CA results can also be used in the future for ‘different-but-similar’ products. Enabling organizations to reuse preferences that can be applied to a range of topics and innovations, CA could be a more efficient means of gathering stakeholder data than repeatedly surveying people.
There remain some challenges in applying CA. Perhaps the most significant is identifying effective sampling strategies and achieving high response rates. Despite adhering to evidence-based sampling strategies [
34], our exercise resulted in only an 11% response rate. Conjoint Analysis questionnaires in healthcare can be complex, a complexity that is exacerbated because decomposing and describing healthcare innovations on the basis of their compound attributes is difficult. Making descriptions informative, nuanced and yet meaningful and accessible (to non-technical audiences) is difficult. Healthcare technologies and the factors involved in decision-making in this sector may be more numerous than for products in which Conjoint Analysis has been traditionally used. Higher response rates have been achieved in other healthcare contexts, suggesting that this challenge can be overcome [
21,
35].
A second difficulty is that the stability of stakeholder preferences over time is less well known. Although some studies in non-healthcare contexts suggest that preferences are relatively stable [
36,
37], there is scope for future research into the effect of time on preferences. Theoretically, once derived, preference models could be applied to future innovation implementation choices. More work is required to establish the stability of preferences in this context.
Balancing conciseness of language in the questionnaire and meaningfulness to clinicians was challenging. A further challenge was identifying attributes and levels that could be applied to a diverse range of innovations; for example, diagnostic techniques and treatment modalities.
While ratings-based conjoint was preferred in our context [
34], rating questionnaires can be difficult to complete. More technically, individual-level utilities are not available in ratings approaches. A variant of CA, Discrete Choice Experimentation (DCE), offers an alternative built around random utility theory (RUT). DCE is widely used in health economics [
25]. In contrast to rating-based Conjoint Analysis, respondents are faced with direct choices between options (‘Of these two options which would you choose?’), an approach that mirrors real decision making.
Finally, the CA approach outlined in this paper applied only to the adoption of innovations into a healthcare system. The approach tells us nothing about the choices people make to stop using or dis-adopt an intervention. There is considerable scope for adapting the approach to examining the relationship between value and norm compatibility and behaviour that is not desired by those seeking to foster sustainable adoption.
Conclusions
Increasingly, healthcare systems are faced with the problem of which innovations to implement. Conventional methods of prioritization are often intuitive, opaque, and based on socio-political factors such as which stakeholder group voice carries the most weight. There are other determinants of course, but the probability of innovations being adopted is influenced – if only in part – by the values and preferences of professionals (potential adopters) in healthcare systems and the characteristics of innovations. However, preferences can be difficult to gather and analyze systematically, rigorously, reliably and efficiently. Conjoint Analysis, with its central premise that an innovation’s value is the sum of its components, holds considerable promise. This paper has shown that despite the practical hurdles to be overcome, the proof of principle exists: preferences can be mapped, matched to innovation characteristics, and used to shape the design and implementation of interventions to change behaviour and encourage adoption.
Ethical approval
Ethical approval was granted for this study by National Research Ethics Service (Reference 09/H1311/81).
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
The method was identified and developed by CT, AH, KF, and DC. Design, data acquisition and analysis was conducted by CT, KF and AH. All of the authors contributed to the development and completion of the manuscript. All authors read and approved the final manuscript.