How Do Psychological Treatments Work Investigating Mediators of Change
How Do Psychological Treatments Work Investigating Mediators of Change
How Do Psychological Treatments Work Investigating Mediators of Change
Invited Essay
a r t i c l e i n f o a b s t r a c t
Article history: Little is known about how psychological treatments work. Research on treatment-induced mediators of
Received 12 September 2008 change may be of help in identifying potential causal mechanisms through which they operate.
Received in revised form 1 October 2008 Outcome-focused randomised controlled trials provide an excellent opportunity for such work. However,
Accepted 1 October 2008
certain conceptual and practical difficulties arise when studying psychological treatments, most espe-
cially deciding how best to conceptualise the treatment concerned and how to accommodate the fact
Keywords:
that most psychological treatments are implemented flexibly. In this paper, these difficulties are dis-
Treatment
cussed, and strategies and procedures for overcoming them are described.
Mediation
Mechanisms Ó 2008 Elsevier Ltd. Open access under CC BY license.
Interpersonal psychotherapy
Cognitive behaviour therapy
Eating disorders
* Corresponding author. Tel.: þ44 1865 226479; fax: þ44 1865 226244.
E-mail address: [email protected] (R. Murphy).
effects of treatment. They indicate in whom and under what Studying the mediators of action of psychological treatments
circumstances treatment has the most effect. In contrast, medi-
ators are a consequence of treatment and ‘‘explain’’ in a statis- Two particular challenges arise when studying the mediators of
tical sense some of the effects of treatment on outcome. Thus, action of psychological treatments. First, psychological treatments
mediators correlate with treatment, are modified during treat- differ in how they may be conceptualised. Some tend to be viewed
ment and this change precedes the effects of treatment on as a complete unit rather than as a collection of procedures,
outcome. whereas in others the opposite is the case. Interpersonal psycho-
therapy (IPT; Klerman, Weissman, Rounsaville, & Chevron, 1984;
Investigating mediators in the context of randomised Weissman, Markowitz, & Klerman, 2000), for example, is usually
controlled trials conceptualised as a complete treatment or entity, as will be dis-
cussed below. In contrast, many cognitive behavioural treatments
Randomised controlled trials provide an often-missed oppor- (CBT) comprise a variety of different procedural elements that may
tunity to investigate the mediators of treatment effects and be viewed either in isolation or as an integrated group of tech-
guidelines have been proposed for doing this (Kazdin, 2007; niques directed at a specific goal. For example, CBT for depression
Kraemer et al., 2002). The key points are as follows. First, the may be thought of either as a unit or as a treatment that comprises
decision to perform a mediator analysis needs to be taken in a number of distinct procedures including behavioural activation
advance as it will influence the choice of measures used and and cognitive restructuring (Jacobson et al., 1996). This distinction
when they are applied. Next, hypotheses need to be formulated is important since it provides the possibility of either investigating
concerning the likely mechanisms of action of the treatment the action of the treatment as a whole or that of its component
under study. The hypotheses are likely to be derived from the parts.
theory underpinning the treatment and the findings of prior The second matter is practical rather than conceptual. It is that
research. Then, the treatment, the putative mediators and the all but the simplest psychological treatments are implemented in
outcome need to be operationalised and a suitable assessment a flexible manner. In other words, their precise use is tailored to the
protocol devised. The ideal situation is when mediators are individual patient’s psychopathology, circumstances and progress
investigated in the context of trials that include a comparison at making change, and as a result the treatment is not identical in
treatment or control condition as this can help to rule out the every case. For example, in most forms of CBT the precise time
possibility that what appears to mediate change is simply when particular procedures are implemented differs from patient
a general effect of receiving treatment (e.g., an expectancy effect) to patient. Indeed, in some cases certain procedures may not be
or a naturally occurring change (e.g., regression to the mean) used at all. This flexibility, inherent to good clinical practice,
rather than the specific effect of the treatment under consider- substantially complicates research on mediation.
ation. The nature of the comparison treatment is also relevant.
Minimal comparison treatments that are less effective than the
index treatment can serve as controls for the processes just Investigating the mediators of action of CBT and IPT in the
described, whereas treatments that are equally effective but treatment of eating disorders
operate via different mechanisms – as illustrated below – may be
used to ensure that change in the putative mediator is not There follows a description of the strategies and procedures that
a consequence of change in the outcome variable (‘‘reverse we have developed to address certain of the difficulties specified
causality’’). Note that if more than one treatment is being above. The context is the treatment of patients with eating disor-
compared, it is possible to study the mediators of action of both ders but the principles apply across different disorders and
treatments simultaneously, again as illustrated below. different treatments. In this particular case, two treatments are
The timing of the measurements is of critical importance as being studied, interpersonal psychotherapy for eating disorders
change in the mediator needs to be shown to have occurred prior to (Fairburn, 1997; Murphy, Straebler, Cooper, & Fairburn, in press)
change in the outcome of interest as otherwise it could merely be and ‘‘enhanced’’ transdiagnostic CBT for eating disorders (CBT-E;
a secondary effect. Establishing temporal precedence is particularly Fairburn, 2008a; Fairburn et al., in press). Both treatments are
challenging when the effects of the mediator are likely to be rapid outpatient-based and involve 20 treatment sessions over 20 weeks.
as is often the case. One solution is to measure both the putative Initially the sessions are twice weekly, then after 4 weeks they are
mediator and the outcome variable at frequent intervals weekly for 10 weeks, and finally there are three sessions 2 weeks
throughout the period over which change is likely to occur. apart. Both treatments are the sole interventions that the patients
Simultaneous measurement of both the mediator and the outcome receive other than continuation antidepressant medication in some
variable also allows one to control for the level of the outcome cases.
when predicting its subsequent change. It is important to stress IPT and CBT-E differ markedly in their rationale, strategies and
that it is not sufficient to simply measure the level of the outcome procedures, and are very likely to have different modes of action.
variable at some later point in time, for example, at the end of IPT is designed to help patients overcome their eating disorder
treatment. This is because treatment may produce change in the indirectly by resolving current problems in their interpersonal life.
outcome prior to the point at which the mediator is assessed with The treatment is derived from IPT for depression (Klerman et al.,
the consequence that some portion of the change observed in the 1984; Weissman et al., 2000) and closely resembles it. It has three
mediator could be the result of a causal path from early change in phases. The first generally occupies 3–4 sessions. The aim is to
the outcome variable to subsequent change in the mediator (DeR- describe the rationale and nature of the treatment and to identify
ubeis et al., 1990). jointly one or more current interpersonal problems that will
With regard to statistical analysis, a variety of data analytic thereafter become the focus of treatment. In the second phase,
strategies and procedures have been developed to assess whether these problems are examined in detail with the therapist helping
a putative mediator meets statistical criteria for mediation. Readers the patient first to characterise them and then identify possible
are referred to the following sources (among others): Baron and means of addressing them. In the final phase, the focus shifts to the
Kenny (1986); MacKinnon, Lockwood, Hoffman, West, and Sheets future, the goals being to ensure that any interpersonal changes
(2002); Shrout and Bolger (2002) and MacKinnon, Fairchild, and made in treatment are maintained and to minimise the risk of
Fritz (2007). relapse in the longer-term.
R. Murphy et al. / Behaviour Research and Therapy 47 (2009) 1–5 3
CBT-E is the latest version of the leading empirically supported this. This prediction is consistent with the unexplained finding that
treatment for eating disorders (Fairburn, 2008b). Originally, it was patients with bulimia nervosa who receive IPT show significantly
a treatment for bulimia nervosa but it is now transdiagnostic in less change in their frequency of binge eating and vomiting than
nature; that is, it is designed to be suitable for all forms of eating those who receive CBT, a difference that disappears over follow-up
disorder whatever the DSM-IV diagnosis. Unlike IPT, CBT-E due to continuing improvement in the IPT patients and little or no
comprises a collection of strategies and procedures focused directly change among those who received CBT (Agras, Walsh, Fairburn,
on the characteristic psychopathology of eating disorders. While Wilson, & Kraemer, 2000; Fairburn et al., 1993).
individual procedures target specific elements of this psychopa-
thology (e.g., binge eating, body shape checking), the procedures Hypothesised mediators of change in CBT-E
are designed to be used in concert, the goal being to address all the
main maintaining mechanisms operating in the individual patient. In the case of CBT-E, we have chosen to focus on the mediators of
Thus, CBT-E can be conceptualised either as a collection of discrete action of four specific treatment procedures rather than that of the
procedures, each of which has its own mechanism of action, or as treatment as a whole. Each procedure addresses a specific
a complete unit. Complicating the investigation of CBT-E is the fact psychopathological process thought to be central to the mainte-
that it is designed to be used very flexibly with the choice of nance of most eating disorders, and each is described in detail in
procedures and the timing of their implementation being tailored the full treatment guide (Fairburn, 2008a).
to the needs of the individual patient.
There follows a description of how we are attempting to identify 1 The ‘‘weekly weighing’’ procedure – It is predicted that the
the mediators of action of these two very different treatments reduction in weight concern which occurs early on in CBT-E is
starting with an outline of our hypotheses concerning how they largely mediated by a reduction in the frequency of weight
might operate. checking (as a result of the ‘‘weekly weighing’’ procedure).
Patients with eating disorders are extremely concerned about
Hypothesised mediators of change in IPT their weight (i.e., the number on the scale). They often check
their weight very frequently and as a result become focused on
Little has been written about how IPT works. Drawing both on inconsequential weight changes (Fairburn, 2008b). This over-
the original theory underpinning IPT (Klerman et al., 1984) and on concern with weight, and the associated fear of weight gain,
our experience observing how patients change during treatment, maintains strict dietary restraint and is a barrier to patients
we have formulated four hypotheses concerning the mechanisms changing their way of eating. It is addressed early on in CBT-E
through which IPT might achieve its effects. It is important to note by the weekly weighing procedure in which the therapist and
that we do not view these as the sole mechanisms through which patient jointly check the patient’s weight once a week and then
IPT operates nor are the hypotheses mutually exclusive. For together interpret the resulting finding in the light of education
example, in bulimia nervosa there is a rapid initial change in the and prior readings. Patients are helped not to weigh them-
frequency of binge eating and vomiting that is much the same in IPT selves between these times.
and CBT (Fairburn, Agras, Walsh, Wilson, & Stice, 2004; Wilson, 2 The ‘‘regular eating’’ procedure – It is predicted that the reduc-
Fairburn, Agras, Walsh, & Kraemer, 2002). This is likely to be the tion in the frequency of binge eating which occurs early on in CBT-
result of common ‘‘non-specific’’ processes associated with starting E is largely mediated by the adoption of a pattern of regular eating
treatment and, interestingly, its magnitude is a powerful predictor (as a result of the ‘‘regular eating’’ procedure). Patients with
of outcome (Agras, Crow et al., 2000; Fairburn et al., 2004). eating disorders have a distinctive way of eating. Their
Nevertheless, we hypothesize that in addition to these early non- temporal pattern of eating tends to be characterised by delayed
specific processes (which operate in both IPT and CBT), IPT operates eating and the avoidance of meals or snacks, or by a highly
through four IPT-specific mechanisms. unstructured way of eating (Fairburn, 2008b). Regular eating,
like weekly weighing, is introduced early on in treatment. It
1 The reduction in eating disorder features is largely mediated by involves helping patients establish a daily pattern of eating
a decrease in the severity of the interpersonal problem(s) targeted characterised by three planned meals and two snacks with no
in treatment. In this case, the putative mediator is a reduction in eating in between. Regular eating of this type has a variety of
the severity of the specific targeted interpersonal problem(s). effects one of which is to displace binge eating.
2 The reduction in eating disorder features is largely mediated by an 3 The ‘‘dietary rules’’ procedure – The reduction in dietary restraint
increase in interpersonal self-efficacy with regard to the specific seen in the mid-to-late stages of CBT-E is largely mediated by
problem(s) targeted in treatment. Thus, in this case the putative erosion of the belief that adherence to strict dietary rules is
mediator is an increase in the strength of the patient’s belief necessary to prevent binge eating, weight gain or fatness (as
that he or she is capable of overcoming the targeted interper- a result of the ‘‘dietary rules’’ procedure). The great majority of
sonal problem(s). patients with eating disorders engage in a highly distinctive
3 & 4 The reduction in eating disorder features is largely mediated by form of dieting characterised by attempts to adhere to multiple
an increase in general interpersonal self-efficacy (hypothesis strict and demanding dietary rules (Fairburn, 2008b). Doing so
3) or self-esteem (hypothesis 4). In the case of hypothesis 3, is valued by the patients. The ‘‘dietary rules’’ procedure is
the proposed mediator is an increase in the strength of the designed to address this form of dieting. It involves identifying
patient’s belief that he or she is capable of overcoming patients’ dietary rules and highlighting the fact that they
interpersonal difficulties in general rather than just those substantially impair their day-to-day life and maintain their
which have been the focus of treatment. In hypothesis 4, the eating problem, and so need to be targeted in treatment.
putative mediator is an increase in self-esteem. Patients’ fears about breaking these rules are identified and
challenged using behavioural experiments in combination with
The nature of these four putative mediators is such that it is likely relevant education about the triggers of binge eating and the
that they take considerable time to change. Therefore, in the case of processes underlying weight gain and shape change. As a result,
all four hypotheses it is predicted that the critical change in the patients learn that the feared consequences of breaking these
mediator takes place during the later stages of treatment and that rules do not occur and consequently the rules become viewed
the change in the eating disorder largely takes place subsequent to as a problem rather than as a strength. This erosion of the belief
4 R. Murphy et al. / Behaviour Research and Therapy 47 (2009) 1–5
0w 20w 20 & 40
TIME (weeks in treatment) weeks post-
treatment
IPT HYPOTHESES*
iptM1 iptM2
CBT HYPOTHESES*
Weekly
Weighing (w)
wM1 wM2
wO1 (wO2) wO3
Regular
Eating (r)
rM1 rM2
dM1 dM2
Key
Procedure
Range of possible timing of implementation
Variables are measured weekly across this range of time and data sampling is based on the timing of implementation of procedure
M1 Mediator at baseline (time point 1)
M2 Mediator post-procedure (time point 2)
O1 Outcome at baseline (time point 1)
(O2) Outcome post-procedure (time point 2) to be used to control for change in the outcome which has already occurred
O3 Outcome post-procedure & post-mediator (time point 3)
* Variables for both sets of hypotheses are measured in both treatment groups. Equivalent data are taken from matched cases in the
comparison treatment to allow cross-treatment comparisons to be made.
Fig. 1. Schematic diagram illustrating the research design used to identify mediators of action of IPT and CBT-E in the treatment of patients with eating disorders.
that adherence to strict dietary rules is necessary to prevent their shape checking and to evaluate the significance of the
binge eating, weight gain or fatness leads patients to moderate resulting effects. This results in a decrease in their concerns
and eventually abandon their dietary restraint. The use and about body shape. In common with the dietary rules proce-
timing of this intervention is tailored to the needs of the indi- dure, the use and timing of this procedure is personalised to
vidual patient based on an evaluation of the importance of the individual patient based on the importance of shape
‘‘dietary rules’’ in maintaining whatever eating disorder concern in maintaining their remaining eating disorder
psychopathology remains after the first stage of treatment. psychopathology.
4 The ‘‘body shape checking’’ procedure – The reduction in
concern about body shape which occurs in the mid-to-late stages In contrast with IPT’s hypothesised mechanisms of action, the
of CBT-E is largely mediated by a decrease in body shape checking four CBT procedures are thought to operate rapidly with there
(as a result of the ‘‘body shape checking’’ procedure). Patients being a short time lag between change in the putative mediator and
with eating disorders typically engage in frequent shape change in the relevant outcome variable.
checking (using multiple methods including body pinching,
body measuring, and scrutinising themselves in mirrors) Research design
(Fairburn, 2008b). In the ‘‘body shape checking’’ procedure,
patients are first made aware of their body checking (much of In terms of research design, the investigation of the four puta-
which may occur outside their awareness) and its adverse tive mediators of action of IPT is relatively straightforward (and is
effects. Patients are then helped to make strategic changes to illustrated in Fig. 1). The main points are as follows:
R. Murphy et al. / Behaviour Research and Therapy 47 (2009) 1–5 5