Theme 1: what is adaptation?

The ADAPT study was funded with a particular conceptualisation of adaptation (see definitions of key terms provided in table 1). This is the intentional modification of interventions to meet the needs of a new context, where there is an evidence-base of effectiveness in the original context. Despite this specific remit, emerging results from the ADAPT study’s findings indicate less certainty and clarity about definitions in general. To date, adaptation has largely focused on the modification of intervention components and delivery strategies. While emerging complex system perspectives have theorised ‘interventions’ as inseparable from the contexts in which they operate,^{6 7} limited attention has been paid to the need to modify aspects of the new context to accommodate the intervention. Rugged landscape theory offers interesting ways to think about why the new context may require modification, considering how the introduction of a solution to a problem cannot be understood in isolation, because its optimality depends on the solution of related problems already taking place in the landscape.⁸

The conclusion from the panel discussion was that conceptual thinking is required to differentiate adaptation from related but distinct terms. Previous research has distinguished adaptation from drift, with the latter defined as the process of unintentional or even undertheorised modifications.^{11 14 15} However, the panel recognised the difficulty in establishing the explicitness of intention or the clarity of purpose. For example, diverse stakeholders (eg, practitioners or policymakers) may modify an intervention to align with their own understanding of causal mechanisms, and so the idea of unintentional or undertheorised action is a misnomer.¹⁶ There is also a distinction to be made between adaptation and refinement. The panel discussed refinement as being the modification of an intervention to work in the same place or with the same population. It may be helpful to map this distinction onto the conceptual differences between scale-out and scale-up, with the former defining efforts to implement an evidence-based intervention within a new context and the latter referring to the expansion of delivery within a largely unchanged context.¹³ However, context is dynamic and further specification is needed for these definitions, in order to be more precise in how they are used.

Finally, in the ADAPT study findings to date, there has been limited consideration of whether there should be a minimal threshold of evidence in the original context for studies to be defined as cases of adaptation. Qualitative data from the ADAPT study’s interviews indicated that stakeholders variously draw upon evidence of intervention effectiveness, feasibility or acceptability to justify adaptation. Specifying criteria for the evidence-base may not be possible or helpful in practice, but it does raise important questions about the purpose of adaptation. For example, if the aim of intervention modification is to replicate effectiveness in the new context, then an outcome evaluation in both the original and new context will likely be required.

Theme 2: when do interventions need to be adapted?

The existing literature states that interventions require adaptation primarily when there are mismatches between contexts. This is evident from the ADAPT study’s systematic review,² where frameworks focus on the identification of dissimilarities as a starting point. Equally, the study’s qualitative findings indicate that the perceived uniqueness of the new context is often the driver for adaptation. Discussion among the panel reflected that there are still few decision-making approaches to assess the degree of contextual incongruence, even if there has been important progress in frameworks to map contextual features.^{9 10} There are notable challenges in making such assessments given that context is multidimensional and dynamic, with potential differences in one domain (eg, socioeconomic characteristics) and congruence in others (eg, legal frameworks). A nuanced understanding of programme theory is needed to help understand this contextual complexity. Logic models have been used alongside programme theory to depict central aspects, but while useful, they have been limited in integrating context. Rather they tend to overly emphasise linear, component-driven approaches to presenting interventions.¹⁷ In future, logic models should take advantage of recent progress in developing complex systems-based models,^{18 19} although there is still work to be done in operationalising how to graphically present context while maintaining usefulness. Continued work may also be undertaken to present unintended and potentially negative intervention impacts through dark logic models.

The panel also recognised a risk in focusing too much on contextual dissimilarities. It might create a culture of adaptation hyperactivity, whereby each context is considered so unique that extensive adaptation is always deemed necessary. This culture further risks neglecting de novo intervention development with potentially larger effect sizes in favour of adapting approaches with smaller impacts. Focusing on the likeness between contexts may actually reveal significant commonalities. For example, some parenting interventions have demonstrated clear transportability across contexts,²⁰ perhaps because the parent–child dyad may be similar even if wider family structures are culturally distinct. As a result, frameworks to compare contexts need to be balanced, placing equal weight on the similarities and dissimilarities of relevant system characteristics. The recent TRANSFER Approach provides a useful direction, focusing on assessing the transferability of systematic review findings to the context.²¹

Theme 3: what aspects of an intervention can be adapted and to what extent?

Even where adaptation is justified, existing research does not provide a clear consensus on what can be adapted, and at what point such extensive adaptation qualifies as de novo intervention development. Emerging findings from the ADAPT study’s systematic review indicated that frameworks draw on the concept of ‘core components’ to define the aspects of an intervention that cannot be modified.^{12 22 23} This approach reflects a common perspective held within implementation science, with an historic focus on the systematic and structured replication of evidence-based components that encase the intervention’s active ingredients.²⁴ Fidelity, here is concerned with adherence to these central activities, while peripheral elements can be modified.

The panel found this dominant focus on ‘core components’ problematic. At its most simple level, discussion observed whether it was possible to disentangle core and non-core components. There were also questions about why an intervention would have activities that did not actively support the theory of change. More fundamentally, this position does little justice to the progress of complex system thinking and realist evaluation, which question the notion of interventions as a set of discrete and bounded components.^{6 25} From these perspectives, interventions are seen as system disruptions, with a clear dynamic interdependence between intervention theories, components and the system in which they operate. These interactions are notable with macro-level interventions, such as tobacco legislation. Such interventions are often simple in their components but highly complex in their processes, with impacts relying on extensive changes through the wider system (ie, society) and its subsystems (eg, tobacco industry actors).²⁶ For example, introduction of smoke-free legislation across the UK from 2006 to 2007, which had high levels of compliance, may be attributed to a prolonged period of advocacy to tip the system in favour of change by securing public and political support.

These perspectives also offer a different notion of fidelity, namely functional fidelity, which was absent from the ADAPT study’s systematic review.^{11 27} This suggests that as long as the same theory of change can be activated in the new context, activities can be substituted and components adapted. Interestingly, this view resonates with the emerging findings from the ADAPT study’s qualitative data, where most stakeholders maintained that the theory of change is the key intervention element that cannot be modified. However, there is uncertainty about how to translate this approach into practice. Indeed, the panel reflected that some ideas within complex systems thinking are still based on conceptual reasoning and are difficult to operationalise due to limited empirical evidence. Therefore, methods need to be continually developed to take full advantage of all aspects of complex system perspectives, and there should be consideration of how to standardise intervention functions and adherence to them. Some examples are emerging to support this, including the recent application of ‘functional fidelity’ to re-theorise and explain mixed evidence on patient-centred medical home care.²⁸

Theme 4: who decides upon and conducts intervention adaptation?

The ADAPT study’s systematic review identified the universal importance of engaging with relevant stakeholders at multiple stages to prioritise what should be adapted.² Included frameworks drew upon concepts and approaches associated with community empowerment, action research and transcreation, with the latter being defined as the development and delivery of interventions in a manner that resonates with the target population.^{29 30} More generally, there has been the application of approaches such as the Analytic Hierarchy Process to involve the target population in the cultural adaptation of interventions.³¹ The panel acknowledged the importance of this involvement, especially from a complex system perspective where the focus may be on modifying aspects of the new context in order to accommodate an intervention. However, discussion equally recognised that existing engagement processes risk being somewhat tokenistic, with the locus of power largely residing with the intervention developers. This is reflected in current nomenclature. Where modifications are undertaken by developers, often in collaboration with researchers, they are commonly defined in the literature as acts of ‘adaptation’. Even though these voices provide valuable insight into the inner workings of an intervention, they can risk allegiance bias due to a vested interest in achieving an internationally ‘branded’ product. In contrast, when other stakeholders, such as implementers or participants, modify an intervention they are seen as engaging in ‘tinkering’ or ‘drift’. Here modifications are often considered as unplanned, unintentional and misaligned with the intervention’s theory of change.

In future, the challenge within research is to enable diverse stakeholders to engage with adaptation processes without them coming from a position of relative disenfranchisement and disempowerment. It is important to achieve this while responding to the risk of intuition bias, and recognising there may be modifications that are incompatible with the intervention aims and hypothesised outcomes. To this end, there are areas where progress should be made. The language pertaining to adaptation needs to move beyond the value judgement that many stakeholders are simply tinkering. Indeed, ‘bottom-up’ adaption can make a significant contribution as it reveals how individuals respond to and act upon their immediate social system.³² To support this process, it is useful to draw upon learning from other research areas. For example, reverse innovation is an emerging area of population health interest, where interventions are moved from lower to higher resource settings.^{33 34} This body of work is important in troubling the dominant narrative that only certain groups, or countries, can be the driving force for innovation and change.

There is also a need to move beyond thinking of interventions as having a single and coherent theory of change. Logic models to date have compounded this thinking, and there needs to be innovative consideration of how to present multiple perspectives. In treating theory as singular, and often belonging to the intervention developer, it is easy to think that only they can adapt the intervention with intention and theoretical insight. Yet in acknowledging the multiplicity of theories in existence, it helps to understand that stakeholders may be working to enhance fidelity according to their own sense of how an intervention functions.¹⁶

Theme 5: how are decisions made about re-evaluation in a new context?

The most significant evidence gap that has emerged from the ADAPT study’s systematic review² and qualitative data is the paucity of decision-making approaches to determine the nature and extent of re-evaluation required in the new context. While a range of study designs are being deployed to answer questions on the replicability of intervention feasibility, acceptability and effectiveness, the rationale for their use is rarely considered. Only one framework included in the ADAPT study’s systematic review attempted to provide a conceptual frame for addressing this issue,¹³ maintaining that the degree of re-evaluation required will be contingent on the similarity of contexts and the extent of modification undertaken. Where differences are minimal, re-evaluation may focus on measures of implementation or proximal outcomes on the pathway to longer-term change. Where contexts are significantly dissimilar, and extensive modification has occurred, a replication of the outcome evaluation may be necessary.

While the panel understood this framework to be a useful departure point for considering re-evaluation, there are areas that can be further explored. First, as in the case of rationalising adaptation, there needs to be improved understanding of what it means for contexts to be similar or dissimilar. Second, it may be useful to draw more heavily upon existing frameworks for assessing the applicability and transferability of evidence.^{35 36} However, these can be limited by unrealistic criteria, requiring data not available in the new context.^{35 36} This is a pertinent issue in the field of global health, where interventions are often moved to lower resource contexts. Third, value of information (VOI) approaches can support decision-making about re-evaluation.³⁷ VOI tools weigh the cost of obtaining information about an intervention (eg, effectiveness) against the value of this information in reducing uncertainty in decision-making.