Intention-to-Treat Analysis

ITT analyses all randomised participants by assigned group — including non-adherers, crossovers, and (ideally) those lost to follow-up. Per-protocol, by contrast, analyses only those who completed the assigned treatment.

A CONSORT-style flow — enrolled → randomised → two arms → analysed AS RANDOMISED, with non-adherent and crossover patients still counted in their original arm.

Follow patients down from randomisation: under ITT nobody is moved or dropped based on what happened after randomisation. A patient allocated to treatment who never takes it is still analysed as treatment.

Keeping everyone in their randomised group preserves the prognostic balance randomisation created and reflects real-world effectiveness. ITT is conservative for superiority trials but anti-conservative for non-inferiority — so per-protocol is reported alongside there.

• ITT = analyse as randomised (the default)
• Preserves randomisation → unbiased superiority estimate
• Conservative for superiority, anti-conservative for non-inferiority

Coronary Drug Project (NEJM 1980) — in the placebo arm, 5-year mortality was 15.1% among good adherers (took ≥80% of tablets) versus 28.3% among poor adherers. Both groups took only placebo, so the gap cannot be a drug effect — adherers are simply healthier and lower-risk. This “healthy-adherer effect” is exactly why analysing by treatment-received (per-protocol) misleads: it compares self-selected groups. ITT — analysing as randomised — is the defence against it.

• Excluding dropouts / non-adherers and still calling it ITT — that is per-protocol.
• Assuming ITT is always conservative — it is NOT for non-inferiority trials.
• Accepting an undefined “modified ITT” without scrutinising which patients were excluded.