Randomised Controlled Trials (RCTs)

HSRU is funded by the Chief Scientist Office of the Scottish Government Health Directorates . The author accepts full

responsibility for this talk.

Health Services Research Unit University of Aberdeen

Randomised Controlled Trials (RCTs)

Graeme MacLennan

Health Services Research Unit

James Lind• Born Edinburgh 1716• On HMS Salisbury in 1747 he

allocated 12 men with scurvy– Cider– Seawater– Horseradish, mustard,

garlic– Nutmeg– Elixir Vitriol – Oranges and Limes


Think about… • Consider how you would go about evaluating the

following interventions– Surgical versus medical termination of

pregnancy– Referral guidelines for radiographic

examination– Paracetamol and/or ibuprofen for treating

children with fever– Nurse counsellors as an alternative to clinical

geneticists for genetic counselling– Single dose of chemotherapy versus

radiotherapy treating testicular cancer http://news.bbc.co.uk/1/hi/health/7647007.stm

– Cervical cancer vaccine http://news.bbc.co.uk/1/hi/health/6223000.stm

http://news.bbc.co.uk/1/hi/health/7647007.stm

http://news.bbc.co.uk/1/hi/health/6223000.stm


The need to evaluate health care

• Variations in health care• Unproven treatments• Inadequacies in care• Inaccurate medical models• Limitation of resources• New innovations• …

Crombie (1996)


Evaluation process• Define research question• What is already known?• Identify appropriate study design• Define population, intervention and criteria for

evaluation• How large a study?• Consider measurement of evaluation criteria

(“outcomes”)– How often?– Timing? Length of follow up?– To whom? Who collects the data? What

format?• Analysis of data• Dissemination and implementation


Define research question and what is already known

• Research question (PICOT)– Population– Intervention– Control/comparator – Outcome– Target

• Has the question already been answered?– Conduct review to assess what is know

about intervention


Definition of population, intervention and

“outcomes”• Population– Strict definition (explanatory) or flexible

(pragmatic)• Intervention

– Dose of drug, timing etc• “Outcomes”

– Health related Quality of Life– Biochemical outcomes– Symptoms– Physical assessment– Patient satisfaction– Acceptability– Cost-effectiveness


Measuring “outcome”• Questionnaires, interview, medical

notes etc• Timing of questionnaires?

– Baseline (prior to treatment)– Short term outcomes– Long term outcomes

• Who collects the data?


Sources of systematic errors

• Selection bias– can be introduced by the way in which

comparison groups are assembled • Attrition bias

– systematic differences in withdrawal/follow up• Performance bias

– Systematic differences in care provided• Observation/detection bias

– systematic differences in observation, measurement, assessment


What is a randomised controlled trial?

Simple Definition• A study in which people are

allocated at random to receive one of several interventions

(simple but powerful research tool)


Simple RCT model

Trial participants

RANDOMLY allocated to experimental or CONTROL group

CONTROLEXPERIMENT



• Random allocation to intervention groups

• all participants have equal chance of being allocated to each intervention group

why RCTs are referred to as randomised controlled trials


Terminology • Interventions are comparative regimes

within a trial • Prophylactic, diagnostic, therapeutic e.g.

– preventative strategies– screening programmes– diagnostic tests– drugs– surgical techniques



• One intervention is regarded as control treatment (the group of participants who receive this are the control group)

• NOTE: Contemporaneous (not historical controls)

why RCTs are referred to as randomised controlled trials


TerminologyControl Group• can be

– conventional practice

– no intervention (this may be conventional practice)

– placebo

Experimental Group

• receive new intervention

• (also called treatment group or intervention group interchangeably)



• RCTs are– Experiments: investigators can

influence number, type, regime of interventions

– Quantitative: measure events rather than try to interpret them in their natural settings

– Comparative: compare two or more interventions


What is a randomised controlled trial? More Complex Definition

• Quantitative, comparative, controlled experiments in which a group of investigators study two or more interventions in a series of participants who are allocated randomly to each intervention group


Inclusion/exclusion criteria

• Decision rules applied to potential trial participants to judge eligibility for inclusion in trial

• See CONSORT statementwww.consort-statement.org

• Important that they are applied identically to all groups in a trial!

http://www.consort-statement.org/


What is randomisation? • Randomisation is the process of random

allocation• Allocation is not determined by

investigators, clinicians or participants• Equal chance of being assigned to each

intervention group • Individual people

– patients– caregivers (physicians, nurses etc)

• Groups of people, ‘cluster randomisation’• (Covered in more depth in later lecture)


Pseudo-randomisation• Other allocation methods include

– according to date of birth – the number on hospital records– date of invitation etc.

• These are NOT regarded as random• These are called pseudo- or quasi-

random


Terminology• Controlled clinical trials (CCTs) are

not the same as randomised controlled trials

• Controlled clinical trials include non-randomised controlled trials

and randomised controlled trials


Why use randomisation?

• Characteristics similar across groups at baseline

• can isolate and quantify impact of interventions with effects from other factors minimised

• Risk of imbalance not abolished completely even if perfect randomisation

• To combat selection biasUnpredictability paradox: review of empirical comparisons of randomised and non-

randomised clinical trials, Kunz and Oxman 1998 BMJ http://www.bmj.com/cgi/content/abstract/317/7167/1185


Why do we need a control group?

• Don’t need a control group if completely predictable results– Parachutes when jumping from

plane – New drug cures a few rabies

cases• But

– No intervention has 100% efficacy – Many diseases recover

spontaneously


Regression to the mean• Occurs when an intervention aimed

at a group or characteristic that is very different from average

• For example selecting people because they have high blood pressure then measuring them in future will see the blood pressure measurements closer to the mean of the population

Morton and Torgerson BMJ 2003 326:1083-4Bland and Altman BMJ 1994 308:1499 and 309:780


DISTRIBUTION OF RESULTSthreshold

measurement 1measurement 2


Hawthorne Effect• Experimental effect in the direction

expected but not for the reason expected

• Essentially studying/measuring something can change what you studying/measuring


Placebo Effect• Effect (usually, but not always positive)

attributed to the expectation that a therapy will have an effect

• The effect is due to the power of suggestion

• A placebo is an inert medication or procedure

Waber et al 2008 JAMA Commercial Features of Placebo and Therapeutic Efficacy

http://jama.ama-assn.org/cgi/content/full/299/9/1016

http://jama.ama-assn.org/cgi/content/full/299/9/1016

Effect size

Experimental group Control group

Hawthorne E.

R. mean

Placebo E.

Therapeutic E.

Real difference

Noise

Signal

EFFECT OF AN INTERVENTION


Minimising bias in RCTs• Blinding

– Single blind – participants are unaware of treatment allocation

– Double blind – both participants and investigators are unaware of treatment allocation

– Requires use of placebos in drug trials

Schulz and Grimes (2002)


Concealment of random allocation list

• “Trials with inadequate allocation concealment have been associated with larger treatment effects compared with trials in which authors reported adequate allocation concealment”

Schulz KF (1995). Subverting randomisation in controlled trials. JAMA, 274, 1456-8


Blinding, placebos• RCTs should use the maximum degree

of blinding that is possible• Placebo is a ‘dummy’ treatment given

when there is no obvious standard treatment– needed as the act of taking a

treatment may have some effect -need to attribute

– double blind treatments must be indistinguishable to those affected


Empirical evidence of bias

Methodological Issue Increase in treatmenteffect (OR)

Inadequate concealment oftreatment allocation

41%

Unclear method of concealmentof treatment allocation

30%

Trial not blinded (when couldhave been)

17%

Schulz KF et al. JAMA 1995; 273: 408-412


‘Explanatory’ and ‘Pragmatic’ questions

Explanatory• Can it work in an

ideal setting …..?• Efficacy• Hypothesis testing

• Placebo controlled• Double blind

Pragmatic• Does it work in the

real world …..?• Effectiveness• Choice between

alternative approaches to health care

• Standard care• Open


Key differences between explanatory and pragmatic trials

(1)Explanatory Pragmatic

Question efficacy effectiveness

Setting ‘laboratory’ normal practice

Participants strictly defined broader, clinically

indicated (uncertainty)

Interventions strictly defined as clinical practice


Key differences between explanatory and pragmatic trials

(2)Explanatory Pragmatic

Outcomes short-term long-term, patient-surrogates centered and resource orientated

Size small (usually largersingle centre) (often multi-centre)

Analysis treatment received intention to treat

Relevance indirect directto practice


Example of selection bias for PP in an open trial

Worse prognosis

ExpCtrl

None

None

White(2005)

E

E


Terminology: explanatory versus pragmatic

• Explanatory trials– estimates efficacy - that is the benefit

the treatment produces under ideal conditions

• Pragmatic trials– estimates effectiveness - that is the

benefit the treatment produces under routine clinical practice

Roland M, Torgerson D. What are pragmatic trials? BMJ 1998;316:285


RCT as the Gold Standard

• The randomised controlled trial is widely regarded as the gold standard for evaluating health care technologies because it allows us to be confident that a difference in outcome can be directly attributed to a difference in the treatments and not due to some other factor


RCT strengths• Confounding variables minimised• Only research design which can in

principle yield causal relationships– can clarify the direction of cause

and effect• Accepted by EBM school• Don’t have to know everything

about the participants


RCT limitations• Contamination of intervention

groups • Comparable controls• Problems with blinding• What to do about attrition?• Are patients/professionals willing

to be in trial different from ‘refusers’?- external validity

• Cost!


Other issues in RCTs (1)• Ethics• Management issues• Interim analysis and ‘stopping

rules’– part of ethical concern– mechanisms to avoid patient

harm– Data Monitoring and Safety

Committee required for trialsClemens F et al Data monitoring in randomised controlled trials: surveys of recent

practice and policies. Clin Trials 2005;2(1):22-33.


Other issues in RCTs (2)• A power calculation is essential for

the validity of a trial and will always be necessary for grant applications and in publications of the trial (later lecture)

• The methods of randomisation should always be reported. It is not enough to say that the patients were randomly allocated to the treatments. (see CONSORT)

Parallel group (simple) RCT design in practice

Patient eligible for either treatment

Patient gives informed consent

Yes No

Randomise

Exclude from trial

Experimental

treatment

Standard treatment

Standard treatment


Summary• “Gold standard” of research designs• Individual patients are randomly allocated to

receive the experimental treatment (intervention group) or the standard treatment (control group)

• Maximises the potential for attribution• Randomisation guards against selection bias

between the two treatment groups• Standard statistical analysis • Good internal validity• May lack generalisability due to highly selected

participants• Can be costly to set up and conduct, ethical

issues


Good study designGeneral considerations• maximise attribution

– Ensure no factor other than the intervention differs between the intervention and control group

– Random allocation, if adequately carried out, will in the long run ensure comparable groups with respect to all factors

• minimise all sources of error– systematic error (bias)– random error (chance)

• be practical and ethical


Minimise sources of error

Systematic errors (bias)• “inaccuracy which is different in its size or

direction in one of the groups under study than the others ”

• Minimise bias by ensuring that the methods used are applied in the same manner to all subjects irrespective of which group they belong to.

Random errors (chance)• “Inaccuracy which is similar in the different

groups of subjects being compared” • Adequate sample size, accurate methods of

measurement

Elwood (1998)


Study designs• Experimental (Randomised controlled trial)

– A new intervention is deliberately introduced and compared with standard care

• Quasi-experimental (non-randomised, controlled before and after)– Researchers do not have full control over the

implementation of the intervention (“opportunistic research”)

• Observational (Cohort, case-control, cross-sectional)– describes current practice– observed differences cannot be attributed

solely to a “treatment” effect


Evaluation of health care interventions

• Randomised controlled trials are considered as the “gold standard”

• However, some debate over the advantages and disadvantages of different research designs for assessing the effectiveness of healthcare interventions

• Polarised views– “observational methods provide no useful

means of assessing the value of a therapy” (Doll, 1993)

– RCTs may be unnecessary, inappropriate, impossible or inadequate (Black, 1996)

• Approaches should be seen as complementary and not as alternatives (Black, 1996)– Interpretation of RCTs in terms of

generalisability


Useful/interesting links• www.jameslindlibrary.org (History)• www.consort-statement.org (CONSORT)• www-users.york.ac.uk/~mb55/pubs/pbstnote.htm

(All the stats notes from BMJ)• www.ctu.mrc.ac.uk (MRC CTU)• www.rcrt.ox.ac.uk (under construction)• Doll R. Clinical Trials the 1948 watershe BMJ 1998;317:1217-

1220• The unpredictability paradox: review of empirical

comparisons of randomised and non-randomised clinical trials Regina Kunz and Andrew D Oxman BMJ 1998 317: 1185-1190

http://www.jameslindlibrary.org/

http://www.consort-statement.org/

http://www-users.york.ac.uk/~mb55/pubs/pbstnote.htm

http://www.ctu.mrc.ac.uk/

http://www.rcrt.ox.ac.uk/


References• Black. Why we need observational studies to

evaluate the effectiveness of health care. BMJ 1996: 312;1215-8

• Crombie. Research in Health Care. 1996• Doll. Doing more good than harm: the evaluation

of health care interventions. Ann NY Acad Sci 1993:703;310-13

• Elwood M. Critical appraisal of epidemiological studies and clinical trials. 1998 OUP; Oxford.

• Greenhalgh T. How to read a paper. 2001 BMJ; London

• Schulz and Grimes. Lancet Epidemiology series. 2002

Documents

Randomised Controlled Trials (RCTs)