Summary. A treatment study can begin with a well-defined group and still produce a misleading result if many participants are missing at follow-up. In addiction research, loss to follow-up is especially important because people who are harder to recontact may differ systematically from people who remain in a study. The key question is not simply “how many completed?” but “what assumptions were made about everyone who did not?”

Attrition changes the population you are analysing

Imagine a study that enrols 200 people and reports outcomes for 120. The observed 120 are no longer automatically representative of the original 200. If people with poorer outcomes are more likely to be missing, a complete-case analysis can make treatment look more effective than it was. If people doing well are less likely to return because they no longer want study contact, the bias could move in the other direction.

The Cochrane Handbook treats bias from missing outcome data as a core risk-of-bias domain. The important issue is whether the reasons for missingness are related to the true outcome and whether the analysis adequately explores plausible alternatives.

A follow-up percentage is not a quality score by itself

Question Why it matters
How many people were enrolled? Defines the original denominator.
How many contributed to the reported endpoint? Shows the amount of missing outcome data.
Why were participants missing? Missingness related to relapse, disengagement, transfer or recovery can create different biases.
Were missing participants compared with completers? Baseline differences can signal selection in the observed follow-up sample.
Were sensitivity analyses performed? Tests whether conclusions survive plausible assumptions about missing outcomes.

Why “assume everyone lost to follow-up failed” is not automatically safe

A common instinct is to classify every missing participant as a treatment failure. That may be conservative for some binary outcomes, but it can also be unrealistic and can bias comparisons if missingness differs between study groups. The LOST-IT framework, developed from a review of randomized trials, showed that plausible assumptions about outcomes among people lost to follow-up can sometimes change the interpretation of trial results. The LOST-IT study is useful because it turns a vague concern about attrition into a structured sensitivity question.

Addiction trials often face substantial follow-up challenges

A 2025 scoping review of randomized trials in cocaine-use disorder identified 106 trials and found that attrition was common and that reporting and handling of missing data were often limited. The review is not evidence that all addiction studies are unreliable; it shows why attrition methods need to be part of evidence appraisal rather than a footnote. Read the review.

Earlier work in alcohol-intervention research reached a similar methodological warning. A review of brief alcohol-intervention studies reported substantial variability in follow-up and found that people lost to follow-up could differ from those retained. That review illustrates why the characteristics of non-completers matter alongside the raw completion percentage.

What stronger reporting looks like

More informative studies show the participant flow from enrolment through each follow-up point, report reasons for missing data where known, state exactly which population each analysis includes, and present sensitivity analyses when missingness could plausibly alter the conclusion. For longitudinal research, repeated attempts to follow participants and pre-specified statistical methods are preferable to quietly dropping missing observations.

Readers should also separate three concepts: treatment retention, research follow-up and clinical outcome. Leaving treatment is not the same as being lost to research follow-up, and being lost to research follow-up is not itself proof of relapse or recovery.

How to read an outcome claim

Before accepting a statement such as “70% improved,” find the denominator and the time point. Is 70% of everyone enrolled, everyone who started treatment, everyone who completed treatment, or everyone reached at follow-up? Then ask how many people are missing and whether the conclusion changes under plausible assumptions about their outcomes.

For more on interpreting addiction evidence, see our existing guide to rehab success rates and evidence.

Sources and limitations

Sources checked 5 October 2026. This article is a methodological appraisal, not a formal systematic review and not an estimate of any provider’s treatment outcomes.

About this page

Published by AddictionResearch with AI-assisted preparation. General education; no independent clinician review is claimed.

Editorial standards Report a correction