Attrition bias
- Bias
If the people who drop out of a study differ between groups, the result among those who stay can mislead.
What it is
Attrition bias (bias from loss to follow-up) arises when the outcome of some participants is missing and the losses differ between groups in number or reason. It is typical of cohorts and trials. What matters is not only how many are lost but whether those lost had a different prognosis. Worst-case and best-case analyses show whether the conclusion survives the losses.
How it arises
- Some participants drop out, move away or die of other causes.
- If people with side effects or a worse course drop out more from one group, that group looks better.
- Analysing only those who complete follow-up compares groups that are no longer equivalent.
Worked example
- Trial with 200 patients per group. Treated: 20 lost, 20 events among 180 followed. Control: 5 lost, 30 events among 195.
- Complete cases: 20/180 ≈ 11.1% versus 30/195 ≈ 15.4%; treatment looks better.
- Worst case for treatment: all 20 lost treated had the event and none of the 5 controls: 40/200 = 20% versus 30/200 = 15%.
- Best case: no lost treated patient had the event and all 5 controls did: 20/200 = 10% versus 35/200 = 17.5%.
- The RR runs from 20/15 ≈ 1.33 (harm) to 10/17.5 ≈ 0.57 (benefit) depending on what happened to those lost.
- Reading: with such unequal losses (10% versus 2.5%), the conclusion is not robust.
Markers
- Worst case
- Losses
How to prevent or control it
- Report how many were lost from each group and why, for example in a participant flow diagram.
- Use active follow-up: calls, visits, death registries and several ways of contact.
- Run sensitivity analyses (worst and best case, imputation) and see whether the conclusion changes.
Common pitfalls
- Looking only at the total percentage lost and not at whether it differs between groups.
- Believing intention-to-treat analysis solves losses: without the outcome, assumptions are still needed.
Test yourself
Think of the answer, then tap the question to see it.
When do losses to follow-up cause bias?
When they differ between groups in number or reason, or when those lost have a different prognosis from those who stay.
What is a worst-case analysis?
Assuming the least favourable outcome for those lost from the favoured group (and the most favourable in the other) to see if the conclusion holds.
Which designs are most vulnerable to this bias?
Those that follow people over time: cohorts and clinical trials.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Attrition bias | Catalog of Bias (Catalogue of Bias Collaboration, 2017); Chapter 8: Assessing risk of bias in a randomized trial (Cochrane, 2024).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.