Intention-to-treat vs. per-protocol analysis in peptide trials

01

Four labels for analysis populations

Under the intention-to-treat principle, participants are analyzed according to the groups to which they were randomized, even when they do not fully follow the assigned treatment. ICH E9 describes a full analysis set as one kept as close as possible to that ideal. The purpose is to preserve the prognostic balance created by random assignment and estimate the effect of choosing or assigning the treatment strategy.

Modified ITT can exclude defined participants, such as those without any post-randomization information or, in some studies, those who never received a dose. Per-protocol analysis applies specified adherence or protocol criteria. As-treated analysis uses treatment actually received. These labels do not fully define the analysis; the report must state who was included, when data stopped, and how switching was handled.

02

Why randomization can be weakened after assignment

Randomization aims to balance measured and unmeasured prognostic factors between groups on average. Removing participants after randomization can disturb that balance when exclusion relates to treatment, prognosis, side effects, access, or early response. People who adhere may differ from people who stop in ways that also affect outcomes. Regrouping by treatment received can create similar confounding.

This is why a larger per-protocol estimate is not automatically a truer biological effect. It may partly reflect who could tolerate, access, or continue treatment. Modern causal methods can attempt to adjust for measured causes of adherence and switching, but they require assumptions. A plain complete-case comparison of adherers does not regain randomization simply because it sounds clinically focused.

03

Crossover, discontinuation, and missing data are separate

A participant can stop assigned treatment yet continue study visits and provide outcome data. Another can switch treatments, receive rescue therapy, miss a visit, or withdraw from all follow-up. Each event affects interpretation differently. ITT does not mean pretending missing outcomes exist; investigators still need a prespecified strategy to collect data and address missingness and post-randomization events.

Record reasons and timing by group. Discontinuation because of adverse effects carries different information from moving away, administrative loss, or recovery. If the analysis excludes everyone who stopped, a treatment that causes early discontinuation can be evaluated mainly among those who tolerated it. If a marketing page reports only completers, readers may never see the experience of the original randomized population.

04

Different analyses answer different questions

An ITT-style treatment-policy question may ask what happens when people are assigned to one strategy rather than another, including real trial patterns of adherence and switching. A per-protocol question may target outcomes if participants followed a specified regimen, but valid estimation requires clear protocol rules and appropriate handling of factors that influence adherence. As-treated comparisons focus on received treatment and face related confounding concerns.

The ICH E9(R1) estimand framework encourages trials to state the treatment effect of interest and how intercurrent events such as discontinuation, rescue medication, or death are handled. Readers should not assume an analysis-set label answers all of this. Check the population, variable, treatment condition, intercurrent-event strategy, and summary measure that define the planned question.

05

What to do when the estimates disagree

Compare ITT, per-protocol, as-treated, and sensitivity analyses without selecting the most favorable number. Note whether conclusions change in direction, size, or statistical uncertainty. Large disagreement can reveal substantial nonadherence, crossover, missing data, or strong modeling assumptions. It is a reason to investigate the trial flow and analysis plan, not permission to declare one estimate correct by preference.

In noninferiority trials, both ITT and per-protocol perspectives can be especially important because some forms of nonadherence may make groups look more similar. In superiority trials, post-randomization exclusions can exaggerate or reduce apparent differences. The design, estimand, and missing-data assumptions determine how each analysis should contribute; no universal slogan replaces that review.

06

A transparent analysis-population audit

Begin with the participant flow diagram: numbers screened, randomized, treated, followed, included in each analysis, and lost. Copy the exact ITT, modified-ITT, per-protocol, and safety-population definitions. Compare reasons for exclusion, discontinuation, rescue therapy, and crossover by arm. Then read the protocol and statistical analysis plan to confirm the definitions were not chosen after results appeared.

For a peptide claim, keep the exact drug product, formulation, route, comparator, population, and outcome attached to the estimate. An adherer-only result from a trial cannot establish efficacy for a compounded product, justify a dose, or predict an individual's response. Treatment questions require a licensed clinician; this guide is for evidence verification and does not recommend a peptide or protocol.

Check the safety population separately. Safety analyses often include participants who received at least one dose and may group them by treatment received, which can be reasonable for attributing exposure but answers a different question from randomized efficacy. Confirm exposure duration and follow-up, since a similar percentage with an adverse event can mask unequal time at risk. Do not merge efficacy and safety denominators into one simplified success rate.

• →Randomized count by arm
• →Definition of every analysis set
• →Treatment received and crossover
• →Discontinuation reasons
• →Outcome follow-up after stopping treatment
• →Missing-data assumptions
• →Sensitivity analyses and estimand