Dear Professor Mean, I’m confused by medical journal articles that talk about the use of “Intention to Treat” analysis. What does this term mean?
An intention to treat analysis specifies how to handle noncompliant patients in a randomized control trial. This analysis requires that patients be analyzed in the groups they were randomized into, regardless of whether they complied with the treatment they were given.
At first, this seems counter-intuitive. If you are trying to see how effective a new drug might be, why would you include patients who refused to take the drug? An analysis that only includes compliant patients has two drawbacks.
- Groups defined by compliance are no longer randomized and are thus subject to biases.
- Groups defined by compliance may not represent the practical impact of the treatment.
The intention to treat analysis has some controversy, however. When you are examining a randomized control trial, ask yourself whether it is logical to include noncompliant patients.
Intention to treat preserves randomization
The validity of a randomized control trial depends greatly on the process of randomization. Randomization insures that both measurable and unmeasurable factors will balance out on average. If a factor other than the treatment itself could possibly influence an outcome measure in your study, then randomization insures that patients with this factor are equally likely to receive either the treatment or the placebo. This prevents many types of bias that can occur in a non-randomized trial.
An analysis that excludes noncompliant patients is no longer randomized and might cause serious bias. Consider a hypothetical example where you are comparing a surgical treatment to a non-surgical control. Some patients might die prior to surgery. This is an extreme example of noncompliance. If we exclude these patients from the analysis, we are eliminating rapidly dying patients from the surgery group, but not from the control group.
Intention to treat analysis is more realistic
There are many factors that influence whether a patient complies or not with a treatment. Some of the factors that influence compliance might also influence the outcome measure. In particular, noncompliant patients tend to have worse outcomes than compliant patients, even in a placebo group. Perhaps patients who forget to follow a prescribed treatment will also forget to do other things important for their health. Thus an analysis that excludes non-compliant patients may produce a study population that is healthier than the patients that you see.
Intention to treat analysis is especially important for medications that are difficult to tolerate. If you exclude noncompliant patients, you are ignoring the influence of poor tolerability on the efficacy of a treatment.
Dropouts in an intention to treat analysis
Sometimes patients who don’t comply with a prescribed treatment will also not comply with the measurement of the outcome. For example, someone who stops taking a drug may also not show up for evaluation either. These people might more properly be labeled as dropouts rather than noncompliant patients. Dropouts may make it impossible to perform an intention to treat analysis, even if you wanted to.
There are a few situations where we can still handle this type of noncompliance. For example, if you are studying smoking cessation, you might conservatively label as a smoker anyone who stops participating in the smoking cessation program. In a study of weight loss, you might assume that a dropout has zero weight loss. Even so, you should always try to design a study that avoids or minimizes the possibility of dropouts.
Excluding noncompliant patients prior to randomization
Some studies evaluate follow patients for a period of time before they are randomized into treatment groups. These studies might allow you to exclude noncompliant patients prior to randomization. If it makes sense to exclude noncompliant patients, this approach allows you to preserve randomization. Be careful, though. Excluding noncompliant patients prior to randomization could make your research results more difficult to generalize to the real world.
A good example of the effect of compliance on an intervention is the MRFIT (Multiple Risk Factor Intervention Trial) study. This study looked at men who were at high risk of heart attacks. Half of the men were randomized into a Special Intervention (SI) group that received advice about dietary changes and were encouraged to quit smoking if they were current smokers. The other half recieved the usual care (UC) from their physicians.
The focus of the study was on a comprehensive intervention with smoking cessation as just one component. So please consider that the following discussion which focuses on the smoking aspects alone is an oversimplification of this study.
The MRFIT study had two separate compliance issues. First, some of the SI patients continued to smoke in spite of efforts to get them to quit. Second, some of the UC patients quit even thoug they didn’t get any special encouragement. The researchers were glad that some of the UC patients quit, I’m sure, and were not surprised that some of the SI patients continued to smoke. But the net effect of this was to dilute the effect of smoking cessation.
When you looked at those who smoked at baseline, more of them died than non smokers. And when you looked at those who quit during the study regardless of which group they were in, fewer of them died. But groups defined by their baseline smoking or by their change in smoking status during the study were not randomized groups.
An intention to treat analysis would look at the two randomized groups and would ignore the fact that some of the intervention group. Although there is plenty of data to suggest that smoking cessation reduces your risk of death, the MRFIT trials tell us is that this particular smoking cessation program doesn’t. In other words, getting someone to quit is effective; telling someone to quit smoking is ineffective.
There are some randomized trials that did show a difference in mortality, possibly because the patients were better at complying, possibly because of differences in the subject pool. And the issues are more complex than I can explain well in a simple message like this. Besides ignoring the dietary portion of the intervention, it’s worth noting that the were multiple outcomes in the MRFIT study and they were assessed at multiple time points. Also there were strong consistencies in the direction of the effects seen in the MRFIT trials and some attempts to adjust for compliance issues seemed to illustrate that the intervention would have been effective if it had convinced more people to quit smoking.
Still, the general explanation is that for this study, the ITT analysis, which ignores compliance, is a measure of the effectiveness of the smoking cessation program. An analysis that factored in compliance would no longer be a randomized comparison, but it would be a better measure of the effectiveness of smoking cessation.
Inquisitive Irene wants to know what the phrase “Intention to Treat” means in all the journal articles she is reading. Intention to Treat is a method that includes noncompliant patients in the groups to which they were originally randomized into. There are two reasons why you might use this approach. First, intention to treat preserves the effects of randomization. Second, intention to treat often provides an assessment of the practical impact of a treatment. There are some examples, however, where intention to treat may answer a different question than the one you are interested in.
You can find an earlier version of this page on my [original website][sim2].
[sim2]: http://www.pmean.com/original_site.html Intention to treat