Prins, Ollendick, Maric, & MacKinnon (2015). Moderators and mediators of youth treatment outcomes.” – Article summary

Almost one-third of children being treated do not improve. There are two ways in which the efficacy of a treatment can be tested:

• Find predictors and moderators of treatment outcome
  This allows treatments to be given to specific subgroups of children under select treatment context to make sure that treatment has a maximum impact.
• Find mediators
  This allows treatments to focus on the mechanisms and change to improve treatment outcomes.

Moderators refer to variables for which treatment has differential effect at different values of the moderating variables. Mediators refer to variables that describe the process through which treatment achieves its effects. There is no causal relationship between the treatment and the outcome for moderators but there is a causal relationship for mediators.

Predictors refer to pre-treatment variables which have a main effect on treatment outcomes but no interactive effect. It can predict the response in different treatment groups but the effect size of the treatment is the same regardless of the value of the target measure.

A potential moderator variable needs to be measured before random assignment (i.e. prescriptive indicator). The interaction between the moderator variable an different treatment conditions establishes a moderation effect. Moderators may allow for personalized approaches as it tells us for which population a treatment may be effective. Predictors are also called prognostic indicators. They predict change across treatment conditions.

Common moderators or predictors are socio-demographic information (1), parent characteristics (2), family characteristics (3), child characteristics (4), school-related factors (5) and aspects of the disorder itself (6). There are a lot of variables that could be included as moderators and predictors. As there are too many, the choice of variables should be guided by theory, clinical experience and empirical studies.

Comorbidity does not predict or moderate immediate treatment outcomes. However, they may predict long-term outcomes. The effects of moderators and predictors need to be evaluated over time.

A test of moderation provides information about whether two treatments differ from one another due to characteristics of the sample or contextual differences but it also demonstrates whether treatments have similar effects across these dimensions (e.g. find evidence that something is not a moderator, then treatments can be applied across both groups of patients). The consistency of the treatments across subgroups and contexts provides important information about the generalizability of the effectiveness of a treatment.

Common mediating factors include parental characteristics (e.g. parenting practices) (1), child characteristics (e.g. emotion regulation) (2), familial characteristics (e.g. cohesion) (3) and school functioning (e.g. relationship with teacher) (4). Treatment theory typically guides the way in determining mediators that need to be investigated. Previous findings of therapeutic interventions may also inform the choice of mediators. Lastly, mediators may be conceptualized as risk factors and this can help guide the choice of mediators.

Both specific and non-specific factors should be tested as mediators. Specific factors refer to the processes aimed to be changed by an active treatment (e.g. avoidance behaviour). Non-specific processes refer to characteristics that are shared by most treatments (e.g. therapeutic alliance; therapists competence).

Some variables need to be assessed as both a mediator and a moderator. Iatrogenic effects refer to mechanisms that lead to worse treatment outcomes. Mediation analysis may indicate potential iatrogenic effects. Studying mediators my also test models of psychopathology (i.e. assess causal or maintaining factors).

Studying mediators typically involves complex and multiple pathways. Next, there are often multiple mediators. Combining mediators and moderators in a study may lead to more informative decisions regarding treatment effectiveness. Including multiple mediators allows the researcher to evaluate the specificity of mediational effects (i.e. variables only mediating for a specific treatment). The specificity of mediation effects can be related to types of outcome (e.g. symptom level; functional impairment; comorbidity). Action theory refers to the extent to which treatment actions lead to changes in mediators.

Moderated mediation refers to assessing whether the mediating process that account for intervention processes differ across a moderator variable value (e.g. developmental stage). In reciprocal mediation models, there may be a reciprocal causation between the mediator and the treatment condition. This means that a different order of causal relations may be true. In a sequential mediation model, two or more mediators intervene in a series between treatment condition and outcomes (e.g. treatment -> mediator 1 -> mediator 2 -> outcome).

There are several requirements for mediation to occur:

1. Treatment needs to affect the treatment outcome.
2. Treatment condition should predict changes in the mediator.
3. Change in the mediator should predict changes in treatment outcome while controlling for the treatment effect.

However, following this method, a large sample size is needed to have sufficient power and this is not always possible. Next, randomization of the mediator does not always occur even when there is randomization of treatment and this can lead to confounds.