To randomise or not to randomise: methodological pitfalls of the RCT design in psychosocial intervention studies

To randomise or not to randomise: methodological pitfallsof the RCT design in psychosocial intervention studies

A. BOTTOMLEY, Cancer Support Centre, Duckworth Lane, Bradford, West Yorkshire

BOTTOMLEY A. (1997) European Journal of Cancer Care 6, 222±230

To randomise or not to randomise: methodological pitfalls of the RCT design in psychosocial

intervention studies

We are seeing evidence of more studies investigating the effectiveness of psychosocial interventions of

cancer patients, predominantly within groups. As roles within cancer and palliative care diversify,

specially trained nurses and other health care workers are taking a more active role within psychosocial

intervention studies. Frequently, these studies are randomised controlled trials (RCTs). Often, the results

of these psychosocial RCTs have been laid open to general criticisms of design, implementation and

reporting. The following paper focuses specifically on the general and experimenter problems in

conducting RCTs within psychosocial interventions. It highlights the limitations and inherent problems

seen with RCTs of psychosocial interventions so that health care workers are aware of these before

considering undertaking psychosocial RCTs with cancer patients.

Keywords: psychosocial interventions, methodological review, randomised controlled trials, design

limitations.

Methodology

INTRODUCTION

The design of the randomised controlled trial (RCT) has

been used for some 50 years (Medical Research Council,

1948) and is now widely regarded as the `gold standard' in

evaluating different treatments (Charlton, 1991). In the

randomised controlled trial, patients are randomly allo-

cated into groups, to remove any differences between

groups. With a large enough sample possible confounding

factors are expected to be distributed equally between

groups. Bradley (1988) and Pocock (1983) note that the

`gold standard' view is held so strongly by some that often

no thought is given to the consequences of using such

designs and of any possible benefit an alternative design

may hold. This unthinking response to a lack of

randomisation as a weakness or flaw in a study has led

to some treatments being used without evaluation and

many studies remaining unpublished because randomisa-

tion was not sensible or ethical.

Given the increasing cost of health care, researchers are

being asked not only to justify the product but also the

process and the cost to society in general terms. The very

process of the RCT is expensive (Elkin et al., 1989; Aveline

et al., 1995). Often they require extensive funding for posts

(data collectors, trial's coordinators, etc.), ongoing costs

and the most qualified and experienced staff to undertake

the work. Extensive effort is required by the researchers to

ensure appropriate recruitment and randomisation occur.

This very fact therefore focused the RCTs into being often

conducted within the `centres of excellence' and leaving

the less robust quasi-experimental studies (e.g. non-

randomised, case-controlled, retrospective and prospective

cohort studies and descriptive studies) to be undertaken by

hospitals with fewer resources.

However, even when RCTs are conducted at centres

of excellence, problems of patient recruitment, attrition

(through death) opposition to randomisation by patients

and referrals, ethical problems raised of randomi-

Correspondence address: Cancer Support Centre, Duckworth Lane,

Bradford, West Yorkshire, UK. e-mail: [email protected]

European Journal of Cancer Care, 1997, 6, 222±230

Paper 035 MS

# 1997 Blackwell Science Ltd.

Ahed

Bhed

Ched

Dhed

Ref marker

Fig marker

Table marker

Ref endRef start

European Journal of Cancer Care

# 1997 Blackwell Science Ltd, European Journal of Cancer Care, 6, 222±230 223

sing seriously sick patients and problems in collecting

data (from sick patients) can either contribute to the fail-

ure of the RCT or at least produce results that are difficult

to interpret (McWhinney et al., 1994; Fallow-field, 1995).

Considering these points, it is important that health

care workers investigating psychosocial interventions are

aware of the general and experimenter limitations of RTCs

before undertaking such research (Pringle & Churchill,

1995). The following text aims to highlight these limita-

tions, relating them to past studies of psychosocial

interventions, so that future researchers may consider

the need for an RCT, where other methods may be more

viable, practical and likely to produce valid results.

SOME COMMON EXPERIMENTER AND

GENERAL PROBLEMS OF RCTS WITH

PSYCHOSOCIAL INTERVENTIONS

Sampling, design and reporting problems

In the RCT with cancer interventions, frequently data are

not reported regarding the baseline characteristics of pa-

tients declining to take any part in the study (e.g. Telch &

Telch, 1986; Cain et al., 1986; Fawzy et al., 1990; Greer et

al., 1992). These patients often represent more than 30%

of the total sample (e.g. Fawzy et al., 1990; Berglund et al.,

1995). In failing to report such details the non-representa-

tiveness of the sample being randomised is not reported.

Failures to recruit consecutive patients in these studies

(through patients declining due to fears of the treatment,

problems of possible allocation to groups, etc) also

introduce potential for selection biases. Unfortunately,

often these are not reported (e.g. Speigel, 1981; Cain et al.,

1986; Houts et al., 1986; Telch & Telch, 1986; Fawzy et

al., 1990). Nevertheless, they can limit the representa-

tiveness of the population and thus the generalisability of

the study results (Silagy, 1994). Reporting these problems

can greatly help fellow researchers, ensuring a reader can

fully assess the validity of the RCT and generalise the

results to other groups.

Drop out rates within psychosocial RCTs pose difficul-

ties. For example, Fawzy et al. (1990) conducted a brief 8-

week psychoeducational intervention with cancer pa-

tients. They found that after randomisation 25% of those

patients randomised into the controlled group dropped

out. This left 28 in the control and 38 in the experimental

group. Although some basic data were collected on the

drop-outs (e.g. age, sex), no information was provided on

why they dropped out or many other details that could be

valuable for readers when attempting to generalise the

results. In addition, when patients drop out for treatment-

related reasons (e.g. did not like the therapist, the

discussions, other members of the groups, etc.) the content

of each group is again no longer determined at random and

this poses a serious threat to RCTs (Schwarts et al., 1986).

Many psychosocial intervention studies fail to report

drop-out rates (e.g. Capone et al., 1980; Cain et al., 1986;

Houts, 1986). A few psychosocial intervention studies do

detail the reasons for the drop-out, which can be revealing.

For example, Speigel et al. (1981) recruited 109 women with

metastatic breast cancer to have supportive discussion

groups focusing on concerns of the cancer, coping and

`living as well as possible'. Eighteen refused after randomi-

sation, five died before contact, further deaths and other

problems reduced the numbers to 34 (treatment) and 24

(control). Available at 1-year follow-up were 16 in the

treatment and 14 in the control group. This makes it

difficult to know exactly what the results of this RCT show.

This is a major and yet a common problem with cancer

interventions and little can be done to overcome drop-out

rates. However, providing compressive details of drop-outs

(e.g. reasons, assessments, periods, etc.) can at least

provide some indication of the validity and generalisa-

bility of the results within the RCT design.

Representativeness problems

For some time it has been known that the results obtained

under the strict conditions of the RCT may not be

generally applicable to those in the normal clinical setting

(Schwartz & Lellouch, 1967). Equally, those cancer

patients who participate in the RCT may be very different

from the general cancer population. Often, when a RCT is

conducted it is at a `centre of excellence' that has all the

facilities and highly qualified staff necessary to maximise

the patients chance of a favourable outcome (e.g. Greer et

al., 1992). In addition, the patient may have an alternative

motive for participating in the RCT, for example, the

increased attention and care observed in these settings

(Stiller, 1989). Given that the cancer patient often suffers

from substantial and sustained distress, participation in a

well-designed and resourced psychosocial RCT will be a

source of help and patients tend to comply with treatment.

Equally, patients who refuse may be those who feel that

participation could be too time-consuming or too much

trouble. Therefore, the artificial environment centred

around the RCTs does not mirror everyday reality,

meaning those patients reported in the RCT might be

very different from those in non-randomised smaller-

centred studies of everyday life.

Furthermore, in conducting psychosocial RCTs, gener-

ally researchers (e.g. Speigel et al., 1981; Fawzy et al.,

1990) have used very stringent eligibility criteria (e.g. age

limits, disease type, site, stage and co-morbidity factors),

making for unrepresentative populations (Druckrey, 1996).

For example, Fawzy et al. (1990) selected patients for

randomisation into a control or psychoeducational group if:

recently diagnosed; with stage one or two malignant

melanomas; medical intervention requiring only excision

of the primary tumour; had undergone no previous

psychiatric treatment; over 18 years old; speak and read

English; excluded if: having immunotherapy, chemother-

apy, radiotherapy or receiving medication (such as steroids

or large doses of aspirin). All these criteria may be valuable

in setting up a clinically meaningful randomised experi-

ment, but clearly they limit the generalisability of the

results into the `real world' setting of cancer patients.

Research workers need to address this issue. One way

may be to call for empirical tests of the generalisability of

findings established in RCT to be undertaken in everyday,

typical settings.

Evidence of equality

To justify undertaking a RCT there has to be evidence that

the interventions (e.g. control, social support, cognitive

behavioural therapy, counselling, psychoeducation, etc.)

do not differ in their effectiveness. However, this principle

relies upon an awareness of the research evidence by those

proposing the study. It is clear that there is poor

dissemination of research evidence in the medical and

nursing profession. This can and does result in poor care

(Antman et al., 1992; Hutton, 1996). This can also mean

that inappropriately designed RCTs are accepted by peer

review and approved by ethical committees after evi-

dence on treatments is already statistically convincing

(Hutton, 1995). When research workers consider under-

taking a psychosocial RCT, it is critical that they have a

clear understanding of the literature on effectiveness and

use only interventions with evidence of èquivalence' or

`no difference'.

However, even if one finds evidence of equivalence

McNeil et al. (1978) argue that the gross idea of `no

difference' between treatments is incorrect and that this is

ignoring the patient's perception of the treatment. Bryne

(1990) also notes this point and states that patients' views

of its effectiveness must carry weight with any clinician

evaluating an intervention. Although little is known about

how much the patients' views affect outcomes, Silverman

(1994) suggests that research to indicate preferences on

outcomes in RCTs designs is required. For example, this

could be done by developing assessment tools to measure

satisfaction of allocation to a group. This information

should reveal any bias in the design and thus help

understand the reliability of the results.

Randomisation problems in psychosocial interventions

Initially, cancer patients may accept random allocation

in psychosocial intervention studies (and yet may prefer

one form of intervention). They may, for example, agree

to participate because this is the only way that they

have any chance of receiving that intervention. When

they find they are randomised into a group they do not

want they can become disgruntled (Brewing & Bradley,

1989) and have low levels of motivation to make the

treatment successful. Patients may either drop out or

participate but be unhappy about their allocation to the

group. From this point on it is likely that differences

between the groups will emerge in factors such as

enthusiasm for the study, willingness to attend the

groups regularly, level of participation during the

sessions, and expectations of the outcome of the group,

which are all factors that can influence the outcome of

cancer patients studies (e.g. Cain, 1986; Fawzy et al.,

1990; Hosaka, 1996).

This is an important limitation of RCTs in psychosocial

interventions with cancer patients, which usually mea-

sure psychological well-being, as opposed to physical

outcomes upon which the design of RCT was initially

intended. For example, take a patient who may be

unhappy about allocation to a social supportive group

intervention; he may decide not to attend the group on a

regular basis. This effect is not simply upon that patient,

but can affect the whole group dynamic by, for example,

failing to allow cohesiveness to develop, which can be a

critical factor in the success of a cancer group (Cella &

Yellen,1994; Speigel, 1995). Equally, a randomised patient

who is unhappy about entering a cognitive therapy group,

where participation and motivation are critical, is unlikely

to undertake the substantial task-focused requirements of

the intervention (Brewin & Bradley, 1989).

Furthermore, there are reports of patients' preferences to

cancer groups and dissatisfaction when allocated to a

particular group. For example, Taylor et al.'s (1986) postal

surveys of patients found that males preferred psychoso-

cial intervention groups that discussed information and

education and were far less interested in sharing personal

concerns. Up to 30% of the 667 cancer patient sample

stated they had distressing experiences in groups. Gray et

al. (1996) reported similar findings, with male cancer

patients preferring information-based intervention groups,

whereas females preferred intervention with intimacy and

emotional support.

224 # 1997 Blackwell Science Ltd, European Journal of Cancer Care, 6, 222±230

BOTTOMLEY Randomised controlled trials

Such preference effects can reduce any potential effect

the interventions may have had if the patients had

selected the group themselves, as patients would do in

the `real world' setting. In particular, it is important to

recognise that often cancer patients report `loss of control'

(Broadhead & Kaplin, 1991; Cella & Yellen, 1994)

associated with their plight and so one could expect

allocation into an unwanted group would add to this

factor. Carlson and Schag (1996) suggested that the success

of a group psychosocial intervention is related to allowing

the cancer patient group the opportunity to comment on

the structure of each intervention so that the programme

may be tailored to the specific needs of that group.

The randomisation of psychosocial interventions should

be conducted by an independent party. Ideally this will be

someone independent of the researchers and clinicians

involved in the study. This ensures the researchers who

measure outcomes remain unaware of the treatments

received by the patient; this is particularly important

where outcomes are assessed by interviews. Although at

least one study has reported adopting this approach

(Forester et al., 1993), many psychosocial RCTs fail to

report whether randomisation was conducted in this

manner (e.g. Telch & Telch, 1986; Fawzy et al., 1990;

Hosaka, 1995; Berglund et al., 1995; Goodwin et al., 1996).

Indeed, if this approach was adopted, it could well create

problems for researchers. For example, the use of a third

party to randomise will increase costs and the complexity

of the study. Patients may also be more likely to remain in

the study if, during the initial assessment, the researcher

makes the randomised choice and advises them which

group they have been allocated. If randomisation is done

by an independent party the process is more complicated

and subject to delays which may influence patient

participation in the study.

Equalisation in psychosocial RCTs

To fulfil the requirements of equalisation (equal distribu-

tion of all confounding variables) in psychosocial RCTs

large numbers of patients are required. The smaller the

sample the more likely it will be that confounding

variables will be unequally distributed between the groups

(Hutton, 1996). This is a fundamental issue for all research

but particularly important in these circumstances. These

can bias the results and there is a limited possibility of

addressing these biases (Shapiro, 1989). This is a problem

with many psychosocial interventions, which tend to have

small samples (e.g. Capone et al., 1980; Houts et al., 1986;

Ali & Khalil, 1989; Burish et al., 1991; Pruitt et al., 1993).

Again, this is particularly so in group formats where

patient accrual can be difficult and sample sizes are

frequently small (see Fawzy et al., 1995). Many published

studies report group sizes of around 20±30 per arm (e.g.

Spiegel et al., 1981; Johnson, 1982; Heinrich & Schag,

1985; Cain et al., 1986; Telch & Telch, 1986; Fawzy et al.,

1990). However, several smaller randomised studies have

been reported which draw conclusions which may not be

robust. For example, Hosaka (1996) reported a RCT study

with 20 women with breast cancer, randomised into two

groups of 10, one group undergoing individual and the other

a group psychoeducational intervention. After five 1-hour

sessions (for both groups) it was concluded that both

interventions were equally effective in significantly redu-

cing psychological distress. Unfortunately no details of how

equalisation of all possible known and unknown variables

could have been achieved with such a small sample are

noted. Key confounding variables that can be measured (e.g.

social class, educational levels) were not reported.

Statistical issues in psychosocial interventions using

RCTs

Although studies report statistical significance, none have

detailed the power of these effects. Most psychosocial

interventions use psychometric measures to assess end-

points at various stages in the interventions. However,

with such small samples, effects are likely to be very small

(Brecker, 1995) and much larger populations are required

to provide larger power and effects. Shapiro et al. (1995)

stress that the detection of differences between psycho-

social interventions of the magnitude that is likely to be

found requires at least 60 patients per group. Shapiro et al.

(1995) emphasise that if resources do not permit samples

of this order then outcome RCT of psychosocial interven-

tions should not be undertaken. To achieve such criteria

large multicentred RCTs would be valuable and indeed are

feasible (Goodwin et al., 1996). However, given that in

more than 20 years of research with cancer groups we have

had only one ongoing multicentred study reported (Good-

win et al., 1996), this would suggest the difficulty of such

an approach (Fawzy et al., 1990). Hospitals can lack

motivation to participate in studies that will have

substantial workloads with limited immediate patient

benefit and yet considerable disruption to the delivery of

health care. Maintaining the motivation of hospitals and

clinicians can also be difficult over the long periods that

studies would need to be conducted with cancer patients.

Recently Berglund et al. (1994) reported one study that

overcomes some of these problems and recognised the

need for larger samples. They reported a successful

psychoeducational study in which recently diagnosed



cancer patients improved significantly more in the inter-

vention (n = 98) than the no-treatment control group

(n = 101). A few other studies have also achieved larger

sample sizes (e.g. Greer et al., 1992) and it is important

that researchers aim to have similar sample sizes if we are

able to detect differences between different interventions

of RCTs.

In addition, given that long-term effects of RCTs in

psychosocial interventions are important, yet often ne-

glected (Tilsbury et al., 1992), end-point selection should

include long-term follow-up (e.g. 1-year post-intervention).

This is important in psychosocial interventions, as

evidence exists suggesting that some interventions may

not demonstrate results immediately and have a `sleeper'

effect (Bell et al., 1989).

Ethical problems in psychosocial RCTs

Ethical problems are common in RCTs with cancer

patients undergoing psychosocial interventions. These

can be grouped as follows.

Good for the individual versus good for the future

population?

For those patients randomised into an intervention they

are likely to want this will hopefully be of benefit to them.

It may offer extra hope and optimism, increase personal

attention, social networks and social support at a time

when these are often at a low level (Bottomley & Jones,

1997). Patients have been reported as undergoing RCTs to

`help other people' or ìmprove treatment for the future'

and to ìncrease my chances of getting good care'

(Alderson, 1996). Indeed, patients can be assured that the

results of robust representative psychosocial RCTs may

help determine future psychosocial interventions for other

cancer patients.

However, questions are raised of the ethics such trials

place upon some of the individual patients. De Rave (1994)

argues that the experimental randomised trial is a value

laden process. That is, certain research questions are

clearly preferred to others, reflecting cultural preoccupa-

tions and prejudices. De Rave (1994) contends that to

randomise the cancer patient is to treat the individual as a

statistic, deliberately impersonal and ruthless. However, it

has been argued that informed consent, where the

individual takes full responsibility for the randomised

trial, addresses this moralist stance (Byrne, 1991).

This may, to an extent, be true, yet many individuals

refuse randomisation in psychosocial interventions, often

as high as 30% of total patients (Greer et al., 1992; Burton

et al., 1995; Berglund et al., 1994). Those patients refusing,

on the ground that they feel unhappy about the randomi-

sation process, are therefore excluded from the provision

of psychosocial support offered. The same can be said of

those patients who agree to participate in the RCT, yet

drop out after being randomised into say either a waiting

list control group or a group they had no interest in.

Furthermore, it is feasible that patients may also find

themselves participating in a group they were unhappy

about being randomised into. They may stay with the

group, yet being at a vulnerable stage of life, may feel

unable to leave and obliged to remain until the interven-

tion has ceased.

Uncertainties of outcomes

Uncertainties of life are a consequence of the cancer

diagnosis. It may be argued that we add to this problem by

using a randomising approach when evaluating psychoso-

cial interventions. Here patients are given more uncer-

tainties about the outcome of events in their lives. For

example, much is written that patients can readily access

which suggests certain types of psychosocial interventions

may increase life (e.g. Speigal et al., 1989) or help patients

cope better (e.g. Speigel et al., 1989; Fawzy et al., 1995). To

face additional worries of being randomised into particular

psychosocial group interventions which patients may

already perceive prior to randomisation to have (a) a lot

of value (e.g. increase duration of life, reduce psychological

distress, Speigal et al., 1989) or (b) no value (e.g. waiting

list), or (c) the potential to cause them harm and distress

(Taylor et al., 1986) must be questioned.

Corbett et al. (1996) provide some evidence that

randomisation and therefore the uncertainty of outcomes

is distressing. In Corbett et al.'s (1996) study of 100

participants (in a study on information preferences or

RCTs) significantly more participants preferred informa-

tion that was less explicit about the role of chance. Over

half thought that they would find randomisation upsetting

and a quarter thought the outcome of a randomised

controlled trial might adversely affect the outcomes.

Bradburn et al. (1995) found that in focus groups of cancer

patients the uncertainties of best treatments could have a

demoralising effect on patients. Alderson (1996) suggests

that the uncertainty of all treatments cancer patients have

to face in the RCT design is a significant emotional burden

that many patients may not wish to experience. Patients'

personal accounts of randomisation in cancer studies

provide evidence of the isolated feeling that the uncer-

tainty of the process leaves them with when they most

need support (Thornton, 1992). Cancer patients often



report the need for health professionals to make the

decisions on interventions in preference to leaving matters

to chance (Inglefinger, 1980).

The uncertainty of outcomes with RCTs with cancer

patients could have a significant impact on the outcome

itself and researchers need to be aware of this problem.

Alternative models of randomisation

To attempt to overcome some of the ethical difficulties of

RCTs alternative models have been proposed to that of the

traditional RCT. The Zelen model is one approach in

which randomisation to the standard or new intervention

is undertaken without consent. Thereafter only those who

have been allotted to the new intervention are taken

through the process of informed consent detailing the

alternatives and the fact that this is a trial. However, this

model is a compromise on the RCT and leaves half the

patients participating without consent. No psychosocial

interventions have been reported using this approach.

Ellenberg (1984) proposed the pre-randomised plus

double consent trial, with each patient group being given

the chance to withdraw from the intervention allotted to

them. However, consent is still only partial and only those

patients who want to undertake that particular interven-

tion will receive it. Refusal rates will be high and

compensation with increased accrual is needed. Again no

psychosocial intervention groups have been reported to

adopt this compromise on the traditional RCT.

Brewing and Bradley (1989) have proposed an approach

which is to assign patients to their preferred treatment

where they express a view and to randomly allocate only

those who have no preference. The non-randomised

patients may be regarded as individuals in a prospective

cohort study, the randomised patients being viewed as a

separate study. This approach clearly overcomes many

ethical and design problems, but may well lead to

considerably larger pools of patients being required for

the randomised part of the study. This increases costs and

will use significantly more resources (e.g. more interven-

tions, staff costs, etc.) than the traditional model. How-

ever, despite the ethical and other advantages of this

approach no randomised psychosocial interventions are

reported using this method.

Clearly the traditional RCT remains the predominate

method in both past and current use with cancer

intervention groups.

Using waiting lists or no treatment groups in RCTs

In the past many RCTs with psychosocial interventions

have randomised an intervention against a non-interven-

tion control or a waiting list group (i.e. Capone et al., 1980;

Speigel et al., 1981; Telch & Telch, 1986; Fawzy et al.,

1990; Greer et al., 1992; Berglund et al., 1994). Here one

has to consider very carefully the ethical position,

particularly if the patient sample is in a palliative state

(e.g. a patient with cancer, which no longer responds to

curative treatment and who is deemed to die within 6

months). There is now a consensus of belief that the

evidence suggests that psychotherapeutic interventions

are better than no treatment for helping patients with

emotional disorders (Shapiro et al., 1995). Three published

meta-analyses of psychosocial interventions with adult

cancer patients have also produced similar findings

(Devine & Westlake, 1995; Mayers & Mark, 1995; Sheard

& Maguire, 1996), and, although limited evidence exists of

different interventions being more effective than each

other (Devine & Westlake, 1995; Fallowfield, 1995;

Trowell et al., 1995), it now may be questionable to

randomise patients into a non-intervention group. Even

randomisation into a waiting list control group is ethically

questionable where a 1- or 2-month period of waiting for

patients who may be dying is unacceptable.

This still occurs in cancer interventions, where patients

have their levels of psychological distress and coping

identified before randomisation and then are random-

ised into the waiting list or the no treatment control

group (Fawzy et al., 1990; Berglund et al., 1994; Goodwin

et al., 1996).

Therefore, with this latest knowledge future research

workers who consider undertaking RCTs of psychosocial

intervention groups may need to employ comparison

designs (e.g. two different types of active interventions).

Such designs may show smaller effect sizes between

interventions, and require larger samples, but they will

overcome the ethical dilemma of randomising into

questionable control groups, often not welcomed by

patients (Parloffe et al., 1986).

PROBLEMS FOR HEALTH CARE

PROFESSIONALS

In designing and reporting psychosocial RCTs researchers

need to be aware that clinicians may be unwilling to

participate if they believe that the experimental study will

disrupt patient care and the clinicians' and patients'

interaction (Taylor et al., 1984; Taylor, 1992; Tobias &

Souhami, 1993). For example, explaining the process of

randomisation can be difficult, distressing and time

consuming for both the health professional and the

patient. Tobias and Souhami (1993) state that to inform



a patient about participation in a RCT is doing them

individual harm. They state many patients are anxious

and are waiting for reassurance and direction at a time of

need when providing full informed consent simply

distresses the patient and is ùnnecessarily cruel'.

This point could be illustrated by comparing Speigal et

al.'s study (1989) with Goodwin's (1996) ongoing study.

Speigal et al. (1989) found significant improvements in

both psychological well-being and survival with women

(who had metastatic breast cancer) after undergoing a

supportive group intervention compared to a group of

women receiving no intervention. Other studies have

found results either confirming these findings (e.g. Fawzy

et al., 1990) or contradicting them (e.g. Ilnyeki et al.,

1994). Goodwin et al. (1996) are undertaking a multi-

centred ongoing RCT to evaluate group support compared

to no treatment (for women with metastatic breast cancer)

as used in the Speigal et al. (1989) study.

The dilemma for health professionals is that they could

face the difficulty of describing the details of the

potentially valuable (but yet unproven) intervention (e.g.

possibly increase life span, reduce emotional distress,

improve coping, increase social support networks, etc.),

gaining the patients' agreement and approval and then

later having to inform them that they have been

randomised into the no treatment group.

No matter how carefully the advantages and disadvan-

tages of any such study are explained to patients, the

results of the randomisation must lead to some `back

peddling' for the health professionals. Here the advantages

of the intervention are `talked down' and perhaps the

negative effects (e.g. time consuming, etc.) are `talked up'.

The patient may become distressed, which can only be

counter-productive to the health carer±patient relation-

ship and the patient may refuse to participate in the study.

Tobiase and Souhami (1993) argue that the support and

reassurrance given by the health care professional and the

patient's trust and confidence in the advice given may be

irretrievably lost due to such events.

Evidence to support the problems health carers may face

is seen in, for example, the difficulty many patients find in

understanding randomising (Hutton, 1996). Simes et al.

(1986) found that more than half of the cancer patients

failed to understand the concept of randomisation. The

greater the details provided to patients the more anxiety

and less willingness to participate occurred. Patients'

confidence in the clinician can be weakened when they

are told that there is no way of knowing whether any

intervention is better unless randomisation is conducted

(Alderson, 1996; Corbett et al., 1996). Invariably patients

ask `well what would you recommend doctor?'.

Therefore, unlike other, non-randomised studies (where

clinicians have fewer demands and disruptions), this can

lead to a bias in that clinicians undertaking RCTs are

those with the most interest and motivation for the

success of the study (Tobias & Souhami, 1993).

SUMMARY

This article has focused on the problems and pitfalls of

applying a RCT with psychosocial interventions. One

cannot discount that the RCT provides a rigorous basis for

evaluating psychosocial intervention groups, yet it may do

so at a cost of methodological problems that can limit the

generalisability to the cancer population at large.

These limitations need to be given careful consideration

by workers before embarking on resource-intensive RCTs.

The illustrated studies show that many common general

and experimenter problems can occur with psychosocial

interventions using the RCT design. These include

problems of representativeness to cancer patients in

general, ethical difficulties in randomisation of patients

and the very practical problems that face researchers when

attempting to fulfil the stringent criteria for a robust RCT.

The economics of today's health care systems are under

scrutiny and when it is unlikely that an RCT can be

conducted without its problems being properly addressed

and taken into account in the results, it must be

questionable whether this approach is to be pursued.

There are many alternatives to the RCT that are valid

with psychosocial intervention groups, including quasi-

experimental and naturalist approaches. Therefore, we

encourage researchers to examine all design options before

embarking on RCTs.

Acknowledgements

To Dr Jeanette James and Helen Fox for their comments

and the support of the Cancer Support Centre during the

preparation of this manuscript.

References

Alderson P. (1996) Equipoise as a means of managing uncertainty:personal, communal and proxy. Journal of Medical Ethics 22,135±139.

Ali N.S. & Khalil H.Z. (1989) Effects of psychoeducationalinterventions on anxiety among Egyptian bladder cancerpatients. Cancer Nursing 12, 236±242.

Antman E.M., Lau J., Kupelnick B., Mosteller F. & Chalmers T.C.(1992) A comparison of results of meta-analysis of randomisedcontrolled trials and recommendations to clinical experts.Journal of the American Medical Association 268, 240±248.

Aveline M., Shapiro D., Parry G. & Freeman C. (1995) Building



research foundations for psychotherapy practice. In ResearchFoundation for Psychotherapy Practice (Aveline M. & ShapiroD. eds), Wiley, Chichester, pp. 301±322.

Bell V., Lynes S. & Kolvin I. (1989) Play group therapy: processes,patterns and delayed effects. In Needs and Prospects of Childand Adolescent Psychiatry (Schmidt M.H. & Remschmidt H.eds), Hogrefe and Huber, Stuttgart.

Berglund G., Bolund C., Gustafsson U.L. & Sjoden P.O. (1994)One-year follow-up of the `starting-again' group rehabilitationprogramme for cancer patients. European Journal of Cancer 12,1744±1751.

Bloom B.S. & Super K.A. (1980) Health and medical care for theelderly and aged population. Journal of the American GeriatricSociety 28, 451±455.

Bottomley A., Hunton S., Roberts G. et al. (1996) Social supportand cognitive behavioural therapy with cancer patient: a pilotstudy of group interventions. Journal of Psychosocial Oncology14, 65±83.

Bottomley A. & Jones L. (1997) Social support and the cancerpatient. A need for clarity. European Journal of Cancer Care 6,72±77.

Bradburn J., Mahar J., Dalton P. et al. (1995) Developing clinicaltrial protocols: the use of patient focus groups. Psych-Oncology4, 107±112.

Bradley W. (1988) Clinical trialsÐtime for a paradigm shift.Diabetic Medicine 5, 107±109.

Breckler S.J. (1995) Psychosocial resource variables in cancerresearch: statistical and analytical considerations. Journal ofPsychosocial Oncology 13, 161±171.

Brewing C.R. & Bradley C. (1989) Patient preferences andrandomised controlled trials. British Medical Journal 299,313±315.

Broadhead W.E. & Kaplin B.H. (1991) Social support and thecancer patient. Implications for future research and care.Cancer 3, 749±794.

Burish T.G., Snyder S.L. & Jenkins R.A. (1991) Preparing patientsfor cancer chemotherapy: effects of coping preparations andrelaxation interventions. Journal of Consulting and ClinicalPsychology 59, 518±525.

Byrne P. (1990) Cancer and the ethics of clinical research. CancerTreatment Reviews 4, 20±21.

Cain E.H., Kohorn E.I., Quninlan D.L., Latimer K. & Schwartz P.E.(1986) Psychosocial benefits of a cancer support group. Cancer57, 183±189.

Capone M.A., Good R.S., Westile K.S. & Jacobson A.F. (1980)Psychosocial rehabilitation of gynaecological oncology patients.Archives of Psychological Medical Rehabilitation 61, 128±232.

Carlsson M. & Schag P. (1996) Educational group support forpatients with gynaecological cancer and their families. Psycho-Oncology 4, 102±109.

Cella D.F. & Yellen S.B. (1994) Cancer support groups: the state ofthe art. Cancer Practice 1, 56±61.

Charlton B.G. (1991) Medical practice and the double blind,randomised controlled trial. British Journal of General Practice41, 357±359.

Colon Y. (1996) Telephone support groups. Cancer Practice 4(3),156±159.

Cunningham A.S. & Tocco E.K. (1989) A randomised trial of grouppsychoeducational therapy for cancer patients. Patient Educa-tion and Counselling 14, 101±114.

Devine C. & Westlakes S. (1995) The effects of psychoeducationalcare provides to adults with cancer: meta-analysis of 116studies. Oncology Nurisng Forum 22(9), 1369±1381.

Druckrey E. (1996) Clinical trials versus socio-economic trials.Cancer Treatment Reviews 4, 51±53.

Elkin I., Shea M.T., Watkins J.S. et al. (1989) National Institute of

Mental Health Treatment of Depression Collobrative ResearchProgramme: general effectiveness of treatments. Archives ofGeneral Psychiatry 46, 971±982.

Fallowfield R. (1995) Psychosocial interventions in cancer. BritishMedical Journal 311, 1316±1317.

Fawzy F.I., Fawzy N.W., Hyun C.S. et al. (1990) Malignantmelanoma. Effects of an early structured psychiatric interven-tion, coping and affective state on recurrence and survival sixyears later. Archives of General Psychiatry 50, 681±689.

Fawzy F.I., Fawzy N.W., Arndt L.A. & Pasnau R.O. (1995) Acritical review of psychosocial interventions in interventions.Archives of General Psychiatry 52, 100±113.

Fawzy F.I., Fawzy N.W. & Wheller J. (1996) A post-hoccomparison of the efficiency of a psychoeducational interven-tion for melanoma patients delivered in group versus individualformats. An analysis of data from two studies. Psycho-Oncology5, 81±89.

Forester B., Cornfield, D., Fless J. & Thompson S. (1993) Grouppsychotherapy during radiotherapy: effects on emotional andpsychosocial distress. American Journal of Psychiatry 150(11),1700±1706.

Goodwin P.J., Leszcz M., Koopmans J. et al. (1996) Randomisedtrial of group psychosocial support metastatic breast cancer: theBEST study. Cancer Treatment Reviews 22, Suppl. A, 91±96.

Gray R., Fitch M., Davis C. & Phillips C. (1996) Breast cancer andprostate cancer self-help groups. Reflection on differences.Psycho-Oncology 5, 137±142.

Heinrich R.L. & Schag C.C. (1985) Stress and activity manage-ment: group treatment for cancer patients and their spouses.Journal of Consulting and Clinical Psychology 33, 439±446.

Hill D.R., Kelleher K. & Schumaker S.A. (1992) Psychosocialintervention in adult patients with coronary heart disease andcancer. General Hospital Psychiatry 14S, 28S±42S.

Hosaka T. (1996) A pilot study of a structured psychiatricintervention for Japanese women with breast cancer. Psycho-Oncology 5, 59±64.

Houts P.S., Whitney C.W. & Mortel R. (1986) Former cancerpatients as counsellors of newly diagnosed cancer patients.Journal of the National Cancer Institute 76, 793±796.

Hutton J. (1995) The critical role of statistics in professionalethics. Journal of Applied Philosophy 12(5), 253±251.

Hutton J. (1996) The ethics of randomised controlled trials: amatter of statistical belief. Health Care Analysis 4, 95±102.

Inglefinger F.L. (1980) Arrogance. New England Journal ofMedicine 303, 1507±1511.

Ilnycki A., Farber J., Cheang M. & Weinerman B.H. (1994) Arandomised controlled trial of psychotherapeutic interventionsin cancer patients. Annals of the Royal College of Psychiatry inSurgery 27, 93±96.

Johnson J. (1982) The effects of a patient education course onpersons with chronic illness. Cancer Nursing 6, 117±123.

Krupnick J.L., Rowland J.H., Goldberg R.L. & Daniel U.V. (1993)Professionally-led support groups for cancer patients: an inter-vention in search of a model. International Journal ofPsychiatry in Medicine 23(3), 275±294.

McNeil B.J., Hird M.P. & Cooper M. (1978) Patients' preferencesfor treatments. New England Journal of Medicine 299, 1397±1401.

McWhinney R.I., Bass M.J. & Donner A. (1989) Evaluation ofpalliative care services: problems and pitfalls. British MedicalJournal 309, 1340±1342.

Medical Research Council (1948) Streptomycin in tuberculosistrials: streptomycin treatments in tuberculosis: a MedicalResearch Council investigation. British Medical Journal 2,769±782.

Meyers T.J. & Mark M.M. (1995) Effects of psychosocial



interventions with adult cancer patients: a meta analysis ofrandomised experiments. Health Psychology 14, 101±108.

Parloff M.., London P. & Wolfe B. (1986) Individual psychotherapyand behaviour change. In Annual Review of Psychology(Rosenzweig M.B. & Porter L. eds), Pergoman Palto Alto,California.

Pocock S.J. (1983) Clinical Trials: A Practical Approach. Wiley,Chichester.

Presberg B.A. & Levenson J.L. (1993) A survey of cancer supportgroups provided by the national cancer institute clinical andcomprehensive centres. Psycho-Oncology 2, 215±217.

Pringle C. & Churchhill B. (1995) The randomised controlledtrials in general practice: gold standard or fool's gold. BritishMedical Journal 311, 1381±1382.

Pruitt B., Waligora-Serafin B., McMahon T., Byrd G. & BesselmanL. (1993) An educational intervention for newly diagnosedcancer patients undergoing radiotherapy. Pyscho-Oncology 2,55±62.

Schwartz D. & Lellouch J. (1967) Explanatory and pragmaticattitudes in therapeutic trials. Journal of Chronic Diseases 20,637±648.

Schwartz D., Flamant R. & Lellouch J. (1980) Clinical Trials.Academic Press, London.

Shapiro D. (1989) Outcome research. In: Behavioural and MentalHealth Research: A Hand Book of Skills and Methods. (Parry G.& Watts F.N. eds), Lawance Erbaum Associates, London, pp.163±189.

Shapiro D. & Aveline M. (1995) The Basics of PsychotherapyResearch. London, Wiley.

Sheard T. & Maguire P. (1996) The effects of psychologicalinterventions on anxiety and depression in oncology: two metaanalyses. Psycho-Oncology 5 (abstracts of the Conference of theEuropean Society for Psychosocial Oncology, Amsterdam).

Silgy C. (1993) Developing a register of randomised controlledtrials in general practice. British Medical Journal 306, 897±900.

Silverman W.A. (1994) Patients preferences and randomised trials.British Medical Journal 343, 1585±1586.

Simes J., Tattersall M.H.N., Coates A.S. et al. (1986) Randomisedcomparison of procedures for obtaining informed consent inclinical trials of treatment of cancer. British Medical Journal293, 1065±1068.

Speigel D., Bloom D. & Yalom I. (1981) Group support for patientswith metastatic breast cancer. Archives of General Psychiatry38, 527±533.

Speigel D., Bloom J.R., Kraemer? & Gotthell E. (1989) Effects ofpsychosocial treatment on survival of patients with metastaticbreast cancer. Lancet 2, 888±891.

Stiller C.A. (1989) Survival of patients with cancer: those inclinical trials do better. British Journal of Medicine 299, 1058±1059.

Taylor K.M. (1992) Integrating conflicting professional roles:physician participation in randomised controlled trials. SocialScience and Medicine 35, 217±224.

Taylor S.E., Falke R., Shoptaw S. et al. (1986) Social support,support groups and the cancer patient. Journal of Consultingand Clinical Psychology 54(5), 608±615.

Telch C.F. & Telch M.J. (1985) Psychological approaches toenhancing coping among cancer patients: a review. ClinicalPsychology Review 54, 325±344.

Thornton H.M. (1992) Breast cancer trials: a patient's viewpoint.The Lancet 339, 44±45.

Tobias J. & Souhami S. (1993) Fully informed consent can beneedlessly cruel. British Medical Journal 307, 1199±1201.

Trowell J.A., Berelowizte M. & Kolvin I. (1995) Design andmethodological issues in setting up psychotherapy outcomestudy with girls who have been sexually abused. In: ResearchFoundation for Psychotherapy Practice (Aveline M. & ShapiroD. eds), Wiley, London, pp. 103±117.

Watson M. (1983) Psychosocial interventions with cancerpatients. a review. Psychological Medicine 13, 839±846.



Documents

To randomise or not to randomise: methodological pitfalls of the RCT design in psychosocial intervention studies