Original Research

Intention-to-treat analysis: Protecting the integrity of randomization

Author and Disclosure Information

 

References

Randomization is a crucial part of most clinical trials. The purpose of randomization in a trial comparing 2 groups is to ensure that the groups differ only with respect to the interventions being compared. Randomization determines not only which treatment subjects receive (eg, drug vs placebo), but also how the results are analyzed at the end of the trial.

Intention to treat prevents biased outcomes

The intention-to-treat principle states that all subjects must be analyzed with respect to the group to which they were randomized. Consider a recent study by Spector et al1 in which patients with dementia were randomized to receive an intervention known as cognitive stimulation therapy (CST) or control (equivalent to placebo in a drug trial). The investigators used an intention-to-treat analysis. The intervention was a complex 14-session training program. Many patients did not complete all the sessions. Whether they had completed all the sessions or not, patients initially randomized to the intervention were still considered to have received the intervention when the results were analyzed.

At first, intention to treat doesn’t seem logical. If we are testing an intervention, doesn’t it make sense to evaluate its effect among patients who complied with it fully, and then compare them with patients who were not assigned to the intervention or failed to comply? The problem is, patients who fail to comply with an intervention for whatever reason (not attending all training sessions in the example above) may differ in an important way from those who do.

In the Spector study, for example, it is possible that patients who attended very few of the sessions were more likely to have some subtle cognitive deficits that limited their participation. If placed in the control group, the 2 groups would these patients were excluded from the analysis or differ not only with respect to the intervention, but also with respect to these subtle factors. The value of randomization, therefore, would be compromised. The treatment arm may even appear to be more effective than it really is.

Correspondence
Kiame J. Mahaniah, MD, Greater Lawrence Family Health Center, 34 Haverhill Street, Lawrence, MA 01841. E-mail: k_mahaniah@hotmail.com .

Recommended Reading

The art of medicine after retirement
MDedge Family Medicine
The Problem with Ps
MDedge Family Medicine
Hospitalists: Scourge or salvation?
MDedge Family Medicine
FIRSTConsult: A useful point-of-care clinical reference
MDedge Family Medicine
UpToDate: A comprehensive clinical database
MDedge Family Medicine
Interpretation of confidence intervals
MDedge Family Medicine
Remembering the meanings of sensitivity, specificity, and predictive values
MDedge Family Medicine
Number needed to treat
MDedge Family Medicine
Negative antigen test misses
MDedge Family Medicine
Disclosure of errors preferred by patients
MDedge Family Medicine