The ‘#PEDroTacklesBarriers to evidence-based physiotherapy’ campaign will help you to tackle the four biggest barriers to evidence-based physiotherapy – lack of time, language, lack of access, and lack of statistical skills.
If you are new to the campaign, we suggest that you start at the beginning by looking at earlier posts on strategies to tackle the barriers of lack of time and language. These are available on the campaign webpage, blog, Twitter (@PEDro_database) or Facebook (@PhysiotherapyEvidenceDatabase.PEDro).
A lack of statistical skills is a common barrier to interpreting evidence and implementing evidence-based physiotherapy. Last month, the #PEDroTacklesBarriers campaign focused on interpreting comparative effects in trials. This month, we focus on understanding the importance of intention to treat analysis in trials with three clinician-researchers.
|Aidan Cashin, Exercise Physiologist and researcher, University of New South Wales, Australia
Area of practice: Comparative effectiveness of interventions for people with chronic pain
|Kate Scrivener, Physiotherapist, educator and researcher, Macquarie University, Australia
Area of practice: Post-stroke physiotherapy intervention and research.
|Mark Elkins, Scientific Editor of Journal of Physiotherapy
Area of practice: Physical and pharmacological therapies in respiratory disease and improving the understanding and application of published research by clinicians.
Intention to treat analysis: what is it?
Intention-to-treat is an approach to analysing results in randomised controlled trials. Intention to treat means that all participants who are randomized are included in the statistical analysis and analysed according to the group they were originally assigned, regardless of what treatment (if any) they received. Intention-to-treat is the recommended approach to analysing randomised controlled trial data.
In a hypothetical randomised trial, 100 participants with acute back pain were randomised to receive advice to stay active or bed rest. The primary outcome was back pain, assessed at baseline and at 4 weeks. Participants’ demographic and clinical characteristics (e.g. age, sex, pain scores, duration of pain, etc.) were similar in both groups at baseline.
At 4 weeks, 10 participants could not be contacted (7 in the bed rest group) and therefore they had no data at follow-up. An additional 10 participants did not adhere to the intervention they were initially allocated to – 3 patients randomised to advice to stay active group rested in bed and 7 participants in the bed rest group remained active.
There is a misconception that the best way to analyse data from this hypothetical trial would involve excluding participants who did not contribute to data at follow-up and those who did not adhere to the intervention. That approach is wrong as it introduces bias in results of the trial and does not represent what happens in everyday clinical practice.
Why is intention to treat important in a trial?
Both groups in the hypothetical trial were similar in relation to key demographic and clinical characteristics. Excluding participants who were lost to follow-up may create imbalance in these important characteristics, which in turn will bias the results of the trial. For example, perhaps the participants who were lost to follow-up had more severe pain and did not see any benefit with the recommended treatments and decided to ignore the researchers’ requests for data. Excluding them from the analysis would unbalance in a key clinical characteristic (pain intensity), as there were more participants with more severe pain who were lost to follow-up in the bed rest group. This is likely to generate biased treatment effects. Intention-to-treat analysis avoids this problem by preserving the original groups.
In clinical practice, it is common for patients not to do what clinicians recommend them to i.e. adherence is rarely perfect. Excluding trial participants who did not adhere to the assigned interventions (also known as ‘per protocol analysis’) creates an artificial scenario of perfect adherence that does not represent clinical practice and introduces bias to the results, which are typically overestimated. If adherence to treatments is poor, analyses by intention to treat may underestimate the magnitude of the treatment effect that will occur in patients who adhered to the treatment.