Common use of Moderation Clause in Contracts

Moderation. Did the process moderate the effect of the treatment on the outcome? • Moderation focused questions seek to establish whether the causal effect of the intervention varies amongst groups within the population (contextual variables). • These questions help to determine who is likely to benefit from the intervention and amongst those who benefit, which group of the population have the largest benefit. Processes of trial conduct and context could pose either questions related to mediation or moderation. Processes related to delivery of treatment could pose questions related to effect modification and processes related to mechanisms of action of the treatment could pose questions related to mediation. In essence, this grouping of processes not only maps out the different processes conceptually, but also provides guidance on the type of causal questions that could be asked and by extension the methods that could be deployed to study these processes. All of these would come into play if causal inference was an aim of conducting pe but the findings of the review seem to suggest that causal inference was of little interest. This was most evident in the heavy use of qualitative methods among the studies reviewed. Qualitative and quantitative methods differ starkly in their perspectives of reality where the former considers reality to be a social construct, shared by many individuals and the latter considers the existence of a single reality (Guba & Lincoln, 1981). This is an oversimplified characterisation of the perspective of reality of the two sets of methods. However, this was done to highlight how different each set of methods considers to be reality and by extension the means that each deploy to understand and find out about this reality. Qualitative studies typically following along one of the five approaches (narrative, phenomeno- logical, grounded theory, ethnographic and case study research) and each approach considers reality in different ways and holds different assumptions (▇▇▇▇▇▇▇▇, 2007). The findings from qualitative studies can be considered as detailed accounts of this reality. For example, in the phenomenological approach, it considers reality to be informed by the collective experiences of many individuals as they go through certain phenomena or events. The aim of a study employing a phenomenological approach would then be to give a detailed account of this collective experi- ence that is unique to this phenomenon. Such an account is a subjective evaluation of reality, a reality experienced by the individuals involved. These subjective assessments, in the context of pe, inform the various stakeholders of what it is like to going through specific processes and allows for a richer understanding of what happens at each part of the trial from different perspectives. An important consequence from this richer understanding is that it aids in forming new hy- potheses about causal relationships amongst processes and between processes and outcomes. These new hypotheses could indicate potential ways in which the treatment could be modified to bring about stronger treatment benefits. The shortcoming of a purely qualitative approach is that there is no objective measure of the degree in which the new hypotheses are to be trusted. The is where quantitative methods and specifically, methods of causal inference can aid in providing an objective assessment of this hypothesised causal relationship. Using a drug trial as an example, a qualitative finding could indicate that some patients did not adhere to the medication regime because of the bitter taste of the drug. A hypothesis could be that the bitter taste caused some patients to miss their doses and the lapse in dosing caused a drop in the treatment effectiveness. If such a drop in effectiveness was deemed important to remediate and a change in the composition of the drug was trivial, the drug could be changed to taste less bitter without any further action. However, if this change was not trivial, then the prudent thing to do would be to generate evidence in support of this hypothesis before any changes were made. Such would also be the case if there was a complex web of causation where the change in a single variable could have unexpected or unpredictable effects as in the case of complex interventions. The testing of this hypothesis could either be done using existing data if the necessary data was available or in a new trial designed specifically to evaluate the impact of the taste of the drug on its effectiveness. Having stated the role of qualitative and quantitative methods in the context of an rct, I would address the sparing use of quantitative methods in pe next. This sparing use could be explained either by a lack of need for questions of causality in pe or that there is such a need but barriers exist in conducting such an analysis. Addressing the former, it seems unlikely since the aim of an rct was to identify better options for treatment than existing ones. Furthermore, focusing on the processes that were of interest in the studies reviewed, many of them such as ‘adherence’ and ‘subject’s perception towards treatment’, had an unambiguous potential causal relationship with the outcome. We can therefore rule out a lack of need for causal inference in pe. Having ruled out the lack of motivation for the use of quantitative methods, the existence of barriers becomes the more likely explanation for the lack of use of quantitative methods for causal inference in pe. This is supported by the finding that only 10% of the studies reviewed sought to infer some relationship between the processes studied and the outcomes of interest. Given that many of the processes had potential causal relationships with the outcomes as indicated previously, the lack of studies seeking to test this hypothetical relationship suggests that this was a task with significant difficulties. If we also consider the type of causal questions that could be posed in a pe and the methods that could be employed to test the hypotheses posed by those questions, it becomes clear that for a subset of these questions, such as some questions of mediated causality, accessible tools are not available for use. Accessible in this context refers to tools that do not have prerequisite training or in-depth knowledge about statistical methods and programming in order to use. This lack of accessible tools is another indication supporting the notion that the observed sparing use of quantitative methods is secondary to the barriers encountered. The lack of accessible tools apply to a subset of questions posed in pe and the next chapter would focus on questions of mediated causality in pe. Having reasoned about the existence of barriers in the conduct causal inference within pe,the next few chapters would focus on elaborating and overcoming some of these barriers.