Dropout and Loss to Follow-Up

Dropout and loss to follow-up describe two problems that reporting often merges. A participant who stops taking the drug but keeps attending visits has discontinued treatment, and their outcome is still measured and can still be analysed in the group they were randomised to. A participant who disappears from the study contributes no outcome at all, leaving a hole no analysis can fill without an assumption about what would have happened.

Attrition in long metabolic trials is substantial. The 68-week STEP 1 trial of semaglutide 2.4 mg in obesity saw roughly 7 percent of participants discontinue treatment because of adverse events, against about 3 percent on placebo, with discontinuation from any cause considerably higher on both arms. The pattern is not incidental to the drug class, since gastrointestinal effects cluster early and those who leave are disproportionately the people tolerating it worst.

The size of the loss matters less than its balance and its reason. Attrition equal in both arms and unrelated to outcome mostly costs precision. Attrition that differs between arms, or that is caused by the treatment or the outcome itself, biases the estimate in a direction that cannot be recovered from the remaining data, and a common rule of thumb treats losses below 5 percent as unlikely to matter and above 20 percent as a serious threat to validity.

The error to look for is efficacy quoted from completers, which converts a randomised comparison into a comparison of two self-selected groups and flatters a drug whose intolerant users have already left. Carrying the last observation forward does the same more quietly, freezing a weight or laboratory value at the moment someone quit, which for a treatment producing progressive change assumes precisely the benefit never observed.