Attrition

Attrition refers to the loss of participants from a study over time due to withdrawal, dropout, loss to follow up, death, or other reasons. Also called participant dropout or study withdrawal, attrition reduces the effective sample size and, when the reasons for dropout are related to the treatment or outcome, can threaten the validity of the study's conclusions.

Attrition is one of the most common and consequential threats to the internal validity of longitudinal studies and clinical trials. When participants leave a study before it is completed, two problems arise. First, the reduced sample size decreases statistical power, making it harder to detect true effects. Second, if the reasons for dropout are related to the treatment (for example, if participants in the active drug arm withdraw because of side effects), the remaining participants may no longer be representative of the original randomized sample, introducing attrition bias. Regulatory agencies, systematic reviewers, and journal editors pay close attention to attrition rates and patterns when evaluating the credibility of trial results. In quality assessments, high or unbalanced attrition is a common reason for rating a study at high risk of bias, particularly when the reasons for dropout may be related to the outcome.

Attrition occurs for many reasons. Participants may move to a new location and become unreachable (loss to follow up). They may decide to stop participating because of inconvenience, dissatisfaction, or personal circumstances (voluntary withdrawal). They may experience adverse effects that make continued participation intolerable (treatment related dropout). They may die during the study period. In some studies, participants are excluded from the analysis because of protocol violations or missing data.

The overall attrition rate is calculated as the number of participants who did not complete the study divided by the number originally enrolled, expressed as a percentage. There is no universal threshold for acceptable attrition. A widely quoted rule of thumb treats less than 5% loss as unlikely to cause much bias and more than 20% as a serious threat to validity, but these cutoffs are only rough guides. However, the acceptability of attrition depends not only on the rate but also on whether attrition is differential (different between groups) or non differential (similar across groups) and whether the reasons for dropout are related to the treatment or outcome.

Researchers use several strategies to minimize attrition: providing incentives for continued participation, maintaining regular contact with participants, reducing the burden of study visits, collecting contact information for multiple family members or friends, and designing protocols that are as convenient as possible. When attrition does occur, analytic strategies such as intention to treat analysis (which includes all randomized participants regardless of completion), multiple imputation (which estimates missing values based on observed data patterns), and sensitivity analyses (which test whether different assumptions about missing data change the conclusions) can help mitigate its impact.

Attrition should be documented and reported in every longitudinal study or clinical trial, regardless of the attrition rate. Even low attrition rates should be reported transparently so that readers can assess their potential impact. Attrition analysis should examine whether dropout rates and reasons differ between groups and whether participants who dropped out differ systematically from those who completed the study on baseline characteristics.

A common error is reporting only the overall attrition rate without examining whether it differs between treatment groups. Differential attrition is generally more threatening to validity than non differential attrition because it can create systematic differences between the groups that bias the treatment comparison. Another mistake is assuming that attrition below a certain threshold (such as 20%) is automatically harmless; even a 10% attrition rate can introduce meaningful bias if the dropouts are systematically different from completers.

Researchers also sometimes use per protocol or completer analyses as their primary analysis, which excludes participants who dropped out and can produce biased estimates, particularly in superiority trials. The intention to treat approach, which analyzes all randomized participants in their assigned groups, is the usual primary analysis for superiority trials, but it still needs outcome data for everyone randomized or a sound method for handling the missing outcomes. Finally, a study that does not plan for attrition at the design stage, by inflating the initial sample size to allow for expected dropout, can be left underpowered at completion.

A 12 month randomized trial of a web based weight loss intervention enrolls 400 participants (200 per arm). At month 12, 152 participants remain in the intervention group (24% attrition) and 170 in the control group (15% attrition). Analysis reveals that participants who dropped out of the intervention group had higher baseline BMI (mean 38.2 vs. 34.8 kg/m2) and were more likely to report difficulty using the technology platform. The researchers conduct both an intention to treat analysis using multiple imputation for missing outcome data and a completer analysis, noting that the treatment effect is smaller in the intention to treat analysis (2.1 kg weight loss difference) than in the completer analysis (3.4 kg), suggesting that attrition bias inflates the apparent benefit when only completers are analyzed.