intention-to-treat vs per-protocol
intention-to-treat analyzes everyone as randomized, whether or not they completed treatment; per-protocol analyzes only those who did. ITT preserves the fairness of randomization; per-protocol answers a different, easier question.
when a trial's press release says it worked and the fine print disagrees, this distinction is usually where the difference lives.
why ITT is the default
Randomization only guarantees fair groups at the start. People who drop out are not random — they are often the ones the treatment failed, or who could not tolerate it. Analyzing only completers (per-protocol) quietly deletes those people and flatters the treatment. Intention-to-treat keeps everyone in the denominator, which is why regulators and journals treat it as the primary answer.
a live example
The UPSIDES peer-support trial (five countries, six sites) is on this site and shows the pattern honestly: in the intention-to-treat analysis, two of three subscales moved but the overall social-inclusion score did not reach significance; among participants who actually attended (per-protocol), the total score did. Both facts are worth knowing. Reporting only the second would have been marketing.
somewhere to put it
free. anonymous. people who’ve been where you are 🤍
get Resolv Social — it’s freewant the deeper story? read prescribed to fail
questions
is per-protocol analysis illegitimate?
No — it estimates what treatment does for people who complete it, which is a real question. It becomes misleading only when presented as the trial's main result, because completers are a self-selected group.
keep reading
more from the glossary
resolv social is not a clinical product and does not diagnose, treat, or cure any condition. if you’re in crisis, call or text 988 (u.s.), 24/7, free.