Skip to content

    Intention-to-treat: why keep people in the analysis after they stop?

    Intention-to-treat: why keep people in the analysis after they stop?

    Suppose a person joins a clinical trial, is assigned to a medication group and later stops taking the medication. Should the researchers still include that person when comparing the groups?

    An approach called intention-to-treat generally says to analyze participants according to the groups they were originally assigned to, even if they don't follow the assigned treatment throughout. The idea is straightforward, but the details—especially missing follow-up information—matter.

    A fictional study

    Imagine 200 people are randomly assigned to two groups of 100. One group is offered a new treatment and the other a comparison treatment. During the study, some people stop their assigned treatment because of side effects, cost, a change in preference or another reason.

    In an intention-to-treat approach, the researchers aim to compare people in their original groups rather than moving everyone who stopped into a separate group. The question is closer to “What happened after people were assigned these approaches?” than “What happened only among people who took a medicine exactly as planned?”

    The numbers and study are made up for this explanation. Real analyses depend on the outcome, the study plan and the information researchers were able to collect.

    Why keep the original groups?

    Random assignment is intended to make groups comparable at the start of a trial. If researchers remove everyone who stops treatment, the remaining groups may differ in ways related to the outcome.

    For instance, people who tolerate a treatment well enough to continue may differ from those who stop because of side effects. Comparing only those who remain can leave out an important part of the treatment experience. Keeping the original assignments helps preserve the value of randomization for the question being studied.

    It also reflects the fact that in real life people don't always continue a treatment. A result that includes those experiences can be relevant when considering an offer of treatment, though the study's setting and support may still differ from everyday care.

    Stopping treatment isn't the same as leaving the study

    Someone can stop taking the assigned medicine and still attend follow-up visits or provide outcome information. That information can be valuable. It shows what happened to the person after the treatment changed rather than treating the stop as the end of their story.

    Leaving the study entirely or losing contact creates a different problem: the researchers may not know the outcome. They can't simply count an unknown result as a measured one. The study report should explain how much information was missing and how the analysis handled it.

    This distinction is easy to miss in a headline. Look for separate numbers for people who stopped the intervention and people whose outcome data were unavailable.

    What happens when information is missing?

    Researchers may use statistical methods that rely on assumptions to handle missing outcomes. Different reasonable assumptions can sometimes lead to different estimates. A sensitivity analysis tests how conclusions change under alternative assumptions or methods.

    An intention-to-treat label does not make missing data disappear. Ask how much was missing, why participants stopped providing information if that is known, and whether the conclusions were robust when researchers examined the uncertainty.

    You don't need to become a statistician to notice these questions. A clear study report should describe the flow of participants and the main choices made in the analysis.

    Is it the only way to analyze a trial?

    No. Researchers may also examine outcomes among participants who followed a protocol more closely or according to the treatment actually received. Those analyses can answer related questions, but the groups being compared may no longer have the same protection from the original random assignment.

    For example, people who complete every visit may differ from those who cannot because of work, illness, side effects or transportation. A comparison that excludes them may be useful for a particular purpose, but it needs careful interpretation.

    When a paper presents several analyses, check which one was planned as the main analysis and how the authors describe the others. Similar results can be reassuring; differences are a reason to understand the methods, not to choose whichever result you prefer.

    Applying it to a GLP-1 headline

    If a weight-loss study reports an average change, look for how it handled people who stopped treatment and whether follow-up continued. Side effects, access and treatment changes are part of the experience of care, not footnotes that can always be ignored.

    The result is still an average for the study population under its conditions. It doesn't predict how much weight one person will lose or what will happen if that person stops a medicine. Discuss treatment decisions with your clinician.

    Intention-to-treat is a way of respecting the original assignments when comparing groups. Its value depends on a clear research question, good follow-up and an honest account of what remains unknown.