How to read a paper on a randomised clinical trial (revisited).
Over the past few weeks, several people have asked me to republish some of my earlier posts on how to interpret papers. I will start by providing an update on how to read a randomised clinical trial. I hope that it is useful. .
None of the points that I make are my own, I have simply distilled several useful sources of information.
I know it is tempting to skip past the introduction to a paper, but I always read this so that I am clear on what the investigators are trying to discover. This may seem obvious but it is surprising how often the title or the abstract of the paper do not have a relationship to the study. The introduction also helps me update my knowledge on the subject of the study. At the end of the introduction I look very carefully at the hypothesis that they are trying to test. I then generally write this down and refer back to it as I read the paper.
This is a most important section and the authors should write this carefully and clearly. I look for the following main points;
The patients and setting
Have the authors obtained the sample of participants from a population that is relevant to my clinical practice? This is important because if the findings of the study are going to influence practice, then the patients and treatment setting should be similar to our practice setting. This is called generalisability. If we consider various orthodontic studies; the levels of generalisability from high to low, to clinical practice, could be;
- 10 to 16-year-olds in a “High Street” practice/office
- 10 to 16-year-olds treated in a University/Hospital setting
- adults treated in a “High Street” practice/office
- adults treated in a University setting by residents.
In short, you need to identify whether the group of patients are relevant to your clinical setting. If you feel that they are not, then you may feel that the findings of the study are not relevant to you.
What was the control group?
It is very important that the investigators match the control group to the intervention group. You can check this by looking at the baseline data which should be included in a table. It is also important if a study is evaluating a method of reducing pain that they have compared this against another pain control intervention or a placebo. They should not compare the new intervention against nothing. This is because it is unlikely that an operator would recommend “no intervention” to their patients. This will also take into account any placebo effect.
Did the authors carry out a sample size calculation?
This is very critical because this assures us that the study had sufficient power to detect a difference between the interventions. They study team should base this on previous literature and state the reference they used.
Did they clearly describe the intervention or treatment?
The authors should clearly state what their treatment involves and how it could potentially ‘ work’.
Issues of randomisation
It is very important that the investigators have designed the trial to reduce any bias attributed to the study team and operators. As a result, the following details of randomisation are an essential requirement of reporting a trial.
How did they do the randomisation?
For example, did they use computer-generated randomisation remotely from the site of the study (low risk of bias). Or did they draw lots for treatment out of a hat on the clinic (higher risk of bias)
How did they get good concealment?
Concealment ensures that when a person is being enrolled in a study the operator has no idea what the treatment allocation for the patient is going to be. This is important because if they are aware of a treatment allocation they may not enrol a patient into the study, because of their potential bias about a particular type of treatment. The authors should also state who generated the allocation sequence and who enrolled patients.
The ideal way to ensure adequate randomisation, allocation and concealment is to use a computer remote from the clinic. The person who is enrolling the patient into the trial then contacts the centre and provides details of the patient. Once this data is recorded the operator is then given details of the treatment allocation. There are many trials units that will carry this out for people running studies.
Blinding means that the participants, the operators and those recording the data do not know the treatment that the participant had received. This is very important because it ensures that any personal bias in providing the treatment, recording and interpreting data is minimised. Ideally, a study should be double-blind, but this is not possible for orthodontic studies because it is impossible to conceal the treatment allocation from the patient and operator. However, adequate blinding can be achieved by hiding the treatment allocation from the person who is recording the data.
The authors should provide a flow diagram of the flow of participants through the study. This means that we can identify features such as recruitment issues, the number of dropouts and any differences in dropouts between the groups.
The authors should present the data clearly. Preferably, they should publish the means and 95% confidence intervals should be presented for each variable. This allows a reader to interpret any uncertainty in the data. I have covered this in this post.
Finally, when I read the discussion I look closely to see if the authors have justified their results and related the conclusions to clinical practice. Finally, I look to see if the conclusions are supported by the data! This is not always the case…
The evaluation of a trial is made much easier if the journal adopts a set of reporting guidelines called CONSORT. These can be found here. These journals have endorsed the consort guidelines, American Journal of Orthodontics, Journal of Orthodontics, and the European Journal of Orthodontics.
I hope that these pointers are useful.