Skip to content
Learn Motion
ExploreHow it worksMembership
Log in
Learn Motion

How Doctors Judge Whether a Medical Study Can Be Trusted

1The Clinical Question and Why Study Design Follows From It2Randomization, Allocation, and the Logic of Comparison3Blinding, Follow-Up, and Who Actually Got Analyzed4Reading the Result: Effect Size, Uncertainty, and Significance5Applicability: Does This Result Fit My Patient?6Combining Studies and Forming a Verdict
Blinding, Follow-Up, and Who Actually Got Analyzed

How Differential Dropout Changes Who Is Left

2 / 3
Watch the two arms as participants leave. In the treatment arm, many leave because of side effects; in the control arm, few leave. The people who leave the treatment arm are not a random sample — they are the ones who could not tolerate the treatment, and tolerance is often linked to milder disease. So the treatment arm that remains is enriched with people who were doing well anyway. The control arm still looks like the original randomized population. The groups are no longer exchangeable, and the treatment looks better than it is. Now reverse it: if the people who leave are the ones deteriorating, and they leave mostly from the control arm, the control group is enriched with people doing well, and the treatment looks worse. The point is not how much attrition there was, but whether the attrition was differential and whether the reason for leaving is related to the outcome.
0:00 / 0:00

Attrition is the loss of participants after randomization, and its effect on the result depends entirely on whether the loss is differential — that is, whether the reasons for leaving differ between arms and are related to the outcome.

Consider a trial of a new treatment for a chronic condition, with 500 participants in each arm at the start. In the treatment arm, 100 participants stop the treatment because of side effects and leave the study; in the control arm, 20 participants leave, mostly for unrelated reasons. At the end, the analysis compares 400 treated participants with 480 controls. The treated group is now enriched with people who tolerated the treatment — and tolerance is often correlated with milder disease and better prognosis. The control group still resembles the original randomized population. The two groups are no longer exchangeable, and the treatment looks better than it is, not because the treatment worked, but because the people who remained in the treatment arm were the ones who were doing well anyway.

The direction of the distortion is not fixed. If the participants who leave are the ones who are deteriorating, and they leave disproportionately from the control arm, the control group is enriched with people who are doing well, and the treatment looks worse than it is. The appraisal question is not "how much attrition was there?" but "was the attrition differential, and is the reason for leaving related to the outcome?" A trial with 30% attrition in both arms for similar reasons may be less distorted than one with 10% attrition concentrated in one arm.

References

  1. [1]Cochrane Handbook for Systematic Reviews of Interventions, Chapter 8: Assessing risk of bias in a randomized trialtraining.cochrane.org
Previous2 / 3Next

Learn Motion

Generate a course. Learn it properly.

Operated by Wuhan Daoyin Technology Co., Ltd.

Contact: [email protected]
Privacy PolicyTerms of Service

© 2026 Learn Motion