Per protocol analysisBMJ 2010; 340 doi: https://doi.org/10.1136/bmj.c1825 (Published 07 April 2010) Cite this as: BMJ 2010;340:c1825
- Philip Sedgwick, senior lecturer in medical statistics
A randomised controlled trial investigated the effects of a low energy liquid diet on moderate and severe obstructive sleep apnoea in obese men.1 Men in the control group adhered to their usual diet. A total of 63 men were recruited, of which 30 were allocated to the intervention group and 33 to the control group. Two of the men allocated to the control group were dissatisfied with their allocation, having wanted the intervention, and subsequently withdrew from the trial.
The primary end point was the apnoea-hypopnoea index, the major disease severity index for obstructive sleep apnoea. Outcome was compared between the two groups using a per protocol analysis.
Which one of the following statements best describes the principle of per protocol analysis?
a) Groups of patients were compared on the basis of the treatment the patient had wanted to be allocated
b) Groups of patients were compared on the basis of the treatment the recruiting clinician had originally intended before trial recruitment
c) The treatment groups compared included all patients and as originally allocated after randomisation
d) The treatment groups compared included only those patients who completed the treatment originally allocated
Answer d best describes per protocol analysis. Per protocol analysis compares treatment groups and includes only those participants who completed the treatment protocol originally allocated. In the above trial, two men allocated to the control treatment withdrew from the trial after randomisation. These men would not have been included in the analysis of the primary end point in per protocol analysis, thereby compromising internal validity.2
Two men had a preference for treatment after allocation. Per protocol analysis does not take into account patient preference after randomisation (answer a is false). Some trials are designed to take patient preference into account and these will be described in future questions.
The trial participants were randomised to either intervention or control. The recruiting physician did not influence which treatment the patient received, thereby preventing any allocation bias (answer b is false).
Answer c describes intention to treat analysis, which analyses all patients in the treatment groups to which they were randomly allocated. This is regardless of whether they actually started the allocated treatment, subsequently withdrew from treatment, deviated from the treatment protocol, or received a different treatment. This therefore maintains the comparability in potential confounding factors between treatment groups achieved after random allocation. Intention to treatment analysis promotes internal validity.
In the above study, per protocol analysis was undertaken in addition to intention to treat analysis. This acted as a sensitivity analysis, and enabled the effect of the potential loss in internal validity to be investigated.
Cite this as: BMJ 2010;340:c1825
Competing interests: None declared.