Contents
A user’s guide
Alejandro R Jadad
1 Randomised controlled trials: the basics- An RCT seeks to measure and compare the outcomes of two or more clinical interventions.
- One intervention is regarded as the standard of comparison or control.
- Participants receive the interventions in random order to ensure similarity of characteristics at the start of the comparison
- Randomisation can be achieved through a variety of procedures BRIndividuals, groups, and the order in which measurements are obtained can all be randomised
- RCTs cannot answer all clinical questions
The randomised controlled trial (RCT) is one of the simplest, most powerful and revolutionary tools of research.1,2 In essence, the RCT is a study in which people are allocated at random to receive one of several clinical interventions.
The people who take part in RCTs are called participants or study population (or, less politically correct, ‘subjects’). Participants do not necessarily have to be ill, because as the study can be conducted in healthy volunteers, in relatives of patients, or in members of the general public. The people who design the study, administer the interventions, assess the results, and analyse them are called the investigators. The interventions are also called clinical manoeuvres, and include actions of such varied natures as preventive strategies, diagnostic tests, screening programmes and treatments. For instance, in a study in which patients with rheumatoid arthritis are randomised to receive either ibuprofen or a new non-steroidal anti-inflammatory drug (let's call it ‘perfectafen’) for the relief of pain, you and your colleagues are the investigators, the participants are the patients with rheumatoid arthritis, and the interventions are ibuprofen and the new drug, perfectafen.
Typically, RCTs seek to measure and compare different events that are present or absent after the participants receive the interventions. These events are called outcomes. As the outcomes are quantified (or measured), RCTs are regarded as quantitative studies. In the RCT comparing ibuprofen and perfectafen, for instance, the investigators could select pain as the main outcome, measuring it in terms of the number of patients who achieve complete relief one week after starting treatment. Also, because RCTs are used to compare two or more interventions, they are considered to be comparative studies. You should be aware that there are other types of studies that may be quantitative but do not include comparisons among groups (that is, case series). These studies are also known as non-comparative studies (see Chapter 7).
Usually, one of the interventions is regarded as a standard of comparison or control, and the group of participants who receive it is called the control group. This is why RCTs are referred to as randomised controlled trials. The control can be conventional practice, a placebo, or no intervention at all. The other groups are called the experimental or the treatment groups. In the example, the experimental group is the group that receives perfectafen and the control group is the one that receives ibuprofen.
RCTs are experiments because the investigators can influence the number and the type of interventions, as well as the regimen (amount, route, and frequency) with which the interventions are applied to the participants. I mention this because there is another group of studies in which the events are measured but not influenced by the investigators (they are called observational), and others in which the researchers do not even measure events, but try to interpret them in their natural settings (these studies are called qualitative). Other types of studies are described in more detail in Chapter 7.
In summary, RCTs are quantitative, comparative, controlled experiments in which a group of investigators studies two or more interventions in a series of individuals who receive them in random order.
Are the elements of RCTs very different from other studies?Apart from random allocation to the comparison groups, the elements of an RCT are no different from the components of any other type of prospective, comparative, quantitative study. These components include the rationale and objectives of the study, the research question that the investigators hope to answer, the methodology used to answer it, the results of the study, and its conclusions. These components will be discussed in the following four chapters.
What does random allocation mean?Random allocation means that all participants have the same chance of being assigned to each of the study groups.3 Therefore, allocation is not determined by the investigators, the clinicians, or the study participants.
Despite its simplicity, the principle of randomisation is often misunderstood by clinicians, researchers, journal reviewers and journal editors. It is therefore important for you to be aware that methods for allocating participants according to date of birth (odd or even years), the number of their hospital records, the date at which they are invited to participate in the study (odd or even days), or alternatively into the different study groups, do not give each of the participants the same chance to be included in each of the study groups. Therefore, they should not be regarded as methods that generate random allocation sequences. These methods are often described as ‘pseudo-random’ or ‘quasi-random’. The main problem associated with these methods is that knowledge of the group to which a participant is assigned can affect the decision about whether to enter him or her into the trial and this can bias the results of the whole trial.3,4 If no one cheats, however, these studies could produce well balanced groups. As discussed in Chapter 3, sequences generated randomly can also be subverted easily.
The studies that use pseudo-random or quasi-random methods of allocation are also known as a non-randomised controlled trials. Together with RCTs, these trials form a group of studies called controlled clinical trials. In fact, the RCT is also known as a randomised clinical trial. In other words, all RCTs are controlled clinical trials, but not all controlled clinical trials are RCTs.
What is the purpose of random allocation?By allocating the participants randomly, the characteristics of the participants are likely to be similar across groups at the start of the comparison (also called the baseline). If this is the case, the groups are called balanced at baseline. By keeping the groups as similar as possible at the start of the study, the investigators will be more able to isolate and quantify the impact of the interventions that they are studying, with minimal effects from other factors that could influence the course of the study participants. The factors that could influence the outcomes of a study, which are not related directly to the interventions, could be known or unknown. For instance, randomisation can balance the proportion of patients taking antacids in the study comparing ibuprofen with perfectafen. Although keeping the groups balanced in terms of known factors is important, it can also be achieved without randomisation, as long as the factors have been measured. For example, if perfectafen is evaluated in a retrospective study, the investigators can select a group of patients who received ibuprofen and took antacids which would match the proportion of patients who took antacids and received perfectafen.
In summary, the real value of randomisation is that, if it is done properly, it reduces the risk of serious imbalance in unknown but important factors that could influence the clinical course of the participants. No other study design allows investigators to balance these unknown factors.
You must understand that the risk of imbalance among the groups is not abolished completely, even if the allocation is perfectly randomised. There are many types of bias that can influence the composition and characteristics of the study groups. These biases are discussed in Chapter 3.
How can randomisation be achieved?The generation of random sequences of allocation can be achieved using one of a variety of procedures. Regardless of the method used, investigators should follow two rules: first, they must define the rules that will govern allocation; and, second, they should follow those rules strictly throughout the whole study.
The simplest methods to generate random sequences of allocation are flipping a coin (for studies with two groups) or rolling a die (for studies with two or more groups). As mentioned, the first step is to define the allocation rules. For instance, before flipping a coin, investigators may decide that the tails would correspond to group A and the heads to group B. In addition, when rolling a die in a study with two groups, investigators can choose to allocate participants to group A with odd numbers and to group B with even numbers. With three groups, they can choose to allocate participants to group A if the die shows 1 or 2, to group B if 3 or 4, and to group C if 5 or 6. And so on. Instead of dice or coins, investigators can also use other simple methods such as drawing balls of different colours or ballots with the study group labels from a dark bag.
Investigators can also use random number tables or computers to generate the sequences. Random number tables are tables that contain a series of numbers which occur equally often and are arranged in a random (therefore unpredictable) fashion. The numbers usually have two or more digits. The use of a random number table forces investigators to make more choices than with a coin or a die. As with the coin or the die, they must first decide the correspondence between the numbers and the groups (that is, odd corresponding to A and even to B; or numbers from 00 to 33 to group A, from 34 to 66 to group B, and from 67 to 99 to group C). Then, they have to select the starting point in the table (that is, the beginning, the end, or any point in the middle of the table marked by a pencil dropped with the eyes closed) and the direction in which the table will be read (that is, upwards or downwards). If the numbers in the table contain more than two digits, the investigators have to select the position of the numbers that will determine allocation. For example, if the table contains numbers with four digits (that is, 2314), the investigators can choose the last two digits (14), the first two (23), the second (3), the last (4), the first (2) or the last three (314). The crucial issue is that, once the procedure is defined, it is not modified at any point during the study.
A similar set of numbers may be generated by a computer that is programmed to do so or by most scientific calculators. The procedures and rules that the investigators must follow are identical to those described for the random number tables.
Regardless of the method chosen by the investigators to generate random sequences of allocation, the number and characteristics of the participants allocated to each of the study groups will probably differ (although slightly) at any given point during the study.3 To minimise these differences, investigators can use some strategies known as restricted (or block) randomisation or stratified randomisation.
Restricted randomisation is used to keep the numbers of participants in all the study groups as close as possible. It is achieved by creating ‘blocks’ of sequences which will ensure that the same number of participants will be allocated to the study groups within each block. For example, in a study with three groups (A, B, and C), the investigators can create six blocks: ABC, ACB, BAC, BCA, CAB, CBA. If they use a die to generate the sequences, then they can decide how each of the six numbers of the die will correspond to each of the blocks. In this case, however, each of the sequences will determine the allocation of three participants at the same time, not only one at a time. For instance, if it is decided that 1 corresponds to the sequence ABC, then the three participants that enter the study after a die has shown 1 will be allocated in that order: the first participant to group A, the second to group B, and the third to group C. Equally, if 2 corresponds to ACB, then the first participant to be included in the study after the die shows 2 will be allocated to group A, the second to group C, and the third to group B. The blocks can be of any size, but ideally the size should correspond to a multiple of the number of groups in the study (that is, six blocks for a study with two or three groups).3
Stratified randomisation is used to keep the characteristics of the participants (that is, age, weight, or functional status) as similar as possible across the study groups. To achieve this, investigators must first identify factors (or strata) that are known to be related to the outcome of the study. Once these factors are identified, the next step is to produce a separate block randomisation scheme for each factor to ensure that the groups are balanced within each strata.
On occasion, investigators may not desire the same number of participants in each of the study groups and can decide to allocate unequal numbers to each group, while preserving the homogeneity of the distribution of the characteristics of the participants across the study groups. This is called weighted or unequal randomisation. This type of randomisation tends to be used by investigators who wish to expose fewer participants to the experimental group because of concerns about unexpected adverse events. In the example of ibuprofen versus perfectafen, the investigators may decide to allocate one patient to perfectafen for each four patients who receive ibuprofen.
Unfortunately, the methods of allocation in studies described as ‘randomised’ are poorly and infrequently reported, even when such studies are published in prominent journals.5,6 As a result of these poor descriptions, it is not possible to determine, on most occasions, whether the investigators used a proper method to generate random sequences of allocation.
On the other hand, when the reports of studies described as randomised provide details of the methods of allocation, it has been shown that 5-10% do not use methods that generate random sequences.7,8
The reporting of randomisation and other aspects of RCTs will be discussed in detail in Chapter 5.
What can be randomised in RCTs?The most frequent unit of allocation in RCTs is individual people, either patients (the most common) or caregivers (that is, treating physicians or nurses).
Sometimes, however, it is more appropriate to randomise groups of people rather than individuals. This is known as cluster randomisation. Examples of these groups or clusters are hospitals, families, and geographical areas. Investigators frequently use this approach when the RCTs are designed to evaluate interventions that may affect more than one individual within a particular group (that is, an RCT evaluating the effect of a videotape on smoking cessation on prison inmates or the effects on parents following early discharge from hospital after childbirth). It is also used when the way in which the participants in one study group are treated or assessed is likely to modify the treatment or assessment of participants in other groups. This phenomenon is known as contamination. For example, contamination is present in an RCT comparing a booklet with strategies to increase patient participation in treatment decisions versus conventional practice, if clinicians who have given the booklet to a group of participants start using the strategies described in the booklet during the treatment of participants who do not receive the booklet.
In other cases, investigators may decide to randomise not only individuals or groups of individuals, but also the order in which the measurements are obtained from each participant. For instance, in an RCT evaluating the effects of morphine on cancer pain, the investigators could randomise the order in which analgesia, adverse effects, and quality of life are assessed.
Can RCTs answer all clinical questions?Although RCTs are considered ‘the best of all research designs’9 or ‘the most powerful tool in modern clinical research’,10 they are not a panacea to answer all clinical questions. There are many situations in which they are not feasible, necessary, appropriate, or even sufficient to help solve important problems.
RCTs are the ideal study design to answer questions related to the effects of health care interventions which are small to moderate. The term ‘intervention’ is widely used in health care, but infrequently defined. On most occasions the term ‘intervention’ refers to treatment. As discussed at the beginning of this chapter, however, this term should be used in a much wider sense to include any clinical manoeuvre offered to the study participants that may have an effect on their health status (that is, preventive strategies, screening programmes, diagnostic tests, the setting in which health care is provided, or educational models).
Even though RCTs are appropriate to evaluate health care interventions, you must be aware that there are many important issues in health care that could be studied by RCTs, but for which there are no RCTs available. In addition, even when RCTs are available, they may be insufficient to provide all the answers required by clinicians, patients, or policy makers.11,12 In these cases, you will have to wait for more RCTs to be completed, do more RCTs yourself, or use other types of studies either as your only source of research information or as a complement to the information provided by RCTs. Other study designs and other types of information, with their advantages and disadvantages, are discussed in Chapter 7.
There are also questions for which RCTs are not appropriate. These are usually questions related to aspects of health care that cannot or should not be influenced by the investigators. These include issues related to the aetiology or natural history of diseases. It would be inappropriate, for instance, to design an RCT in which people would be randomised to smoke or not to smoke for decades to compare the frequency of lung cancer between smokers and non-smokers.
In other circumstances, RCTs may not be appropriate even to study some interventions. For example, it may be unfeasible (particularly because of financial constraints, low compliance rates, or high drop out rates) to design an RCT to evaluate the effects of interventions with very rare outcomes or with effects that take long periods of time to develop. In these cases, other study designs such as case-control studies or cohort studies are more appropriate.
The corollary is that, before you start reading an RCT, or even searching for one, you should take into account that there are other study designs that may be more appropriate than an RCT to answer your particular questions (a brief description of studies other than RCTs is provided in Chapter 7). Even if an RCT is available, you must be aware that one RCT in isolation, even when it is appropriate and perfectly designed, is unlikely to provide all the answers that you need. You should consider the information provided by a single RCT as an important piece in a puzzle with many empty spaces. This information will have to be assessed and used in conjunction with other types of information (for example, data from other RCTs or from other study designs, and your own clinical experience), and the values and preferences of the people involved in the decisions, depending on the circumstances in which the decisions are being made.
References
1. Silverman WA, Chalmers I. Sir Austin Bradford Hill: an appreciation. Controlled Clin Trials 1992;13:100-5.
2. Jadad AR, Rennie D. The randomized controlled trial gets a middle-aged checkup. JAMA 1998;279:319-20.
3. Altman DG. Practical statistics for medical research. London: Chapman & Hall, 1991.
4. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias: dimensions of methodological quality associated with estimates of treatment effect in controlled clinical trials. JAMA 1995;273:408-12.
5. Altman DG, Doré CJ. Randomisation and baseline comparisons in clinical trials. Lancet 1990;335:149-53.
6. Moher D, Fortin P, Jadad AR, Jüni P, Klassen T, Le Lorier J, Liberati A, Linde K, Penna A. Completeness of reporting of trials in languages other than English: implications for the conduct and reporting of systematic reviews. Lancet 1996;347:363-6.
7. Mosteller F, Gilbert JP, McPeek B. Reporting standards and research strategies for controlled trials: agenda for the editor. Controlled Clin Trials 1980;1:37-58.
8. Evans M, Pollock AV. Trials on trial: a review of trials of antibiotic prophylaxis. Arch Surg 1984;119:109-13.
9. Norman GR, Streiner DL. Biostatistics: the bare essentials. St Louis: CV Mosby, 1993.
10. Silverman WA. Gnosis and random allotment. Controlled Clin Trials 1981;2:161-4.
11. Naylor CD. Grey zones of clinical practice: some limits to evidence-based medicine. Lancet 1995;345:840-2.
12. Freemantle N. Dealing with uncertainty: will science solve the problem of resource allocation in the UK NHS? Soc Sci Med 1995;40:1365-70.
|
Home | Contents | Foreword | Introduction | Acknowledgments | How to order
© BMJ Books 1998. BMJ Books is an imprint of the BMJ Publishing Group. First published in 1998 by BMJ Books, BMA House, Tavistock Square, London WC1H 9JR. A catalogue record for this book is available from the British Library. ISBN 0-7279-1208-9