top of page
eduo
visual
Biostatistics & Epidemiology
Intention-to-treat vs per-protocol analysis
Core Principle of Intention-to-Treat Analysis
🧷
Intention-to-treat (ITT) analysis includes all randomized participants in their originally assigned groups, regardless of whether they completed the assigned treatment, switched groups, or dropped out.
🧷
This approach preserves the benefits of randomization by maintaining balanced baseline characteristics between groups and preventing selection bias.
🧷
ITT reflects real-world effectiveness — what happens when a treatment is prescribed, not just when it's perfectly adhered to.
🧷
Board pearl: ITT is the gold standard for superiority trials because it provides a conservative estimate of treatment effect and maintains internal validity.
Per-Protocol Analysis: The Explanatory Approach
📍
Per-protocol (PP) analysis includes only participants who completed the study according to protocol — those who received their assigned treatment and had outcome assessment.
📍
This approach answers the question: "What is the treatment effect in patients who actually take the medication as prescribed?"
📍
PP analysis excludes protocol violators, dropouts, crossovers, and those with poor adherence.
📍
While PP analysis may show larger treatment effects, it loses the protection of randomization and can introduce selection bias.
📍
Board pearl: PP analysis is preferred for non-inferiority trials because it provides a less conservative estimate.
The Problem of Non-Adherence and Crossover
🔹
In real-world trials, participants may not take their assigned treatment (non-adherence), switch to the other group's treatment (crossover), or drop out entirely.
🔹
ITT analysis keeps these participants in their original groups, diluting the observed treatment effect toward the null hypothesis.
🔹
This dilution is considered acceptable because it mirrors clinical reality — prescribed treatments aren't always taken as directed.
🔹
Board distinction: A participant randomized to drug A who actually takes drug B is analyzed in the drug A group under ITT, but excluded from PP analysis.
Preserving Randomization and Avoiding Bias
⭐
Randomization creates comparable groups at baseline by distributing known and unknown confounders equally.
⭐
When participants drop out or switch treatments, they often do so for reasons related to the outcome (side effects, lack of efficacy, disease progression).
⭐
Analyzing only those who complete treatment (PP) can introduce systematic differences between groups that weren't present at randomization.
⭐
ITT preserves the "randomization balance" even when post-randomization events occur.
⭐
Board pearl: If dropout rates differ between groups, PP analysis will likely show biased results.
The Conservative Nature of ITT
✅
ITT typically produces smaller treatment effects than PP analysis because non-adherent participants dilute the difference between groups.
✅
This conservative bias is toward the null hypothesis (no difference between treatments).
✅
For superiority trials, if ITT shows a significant benefit, the true effect in adherent patients is likely even larger.
✅
This conservatism protects against false-positive findings and overestimating treatment benefits.
✅
Board clue: When a question states "analysis was performed on all randomized patients," this indicates ITT analysis.
Missing Data and the ITT Principle
🧠
True ITT requires outcome data on all randomized participants, but missing data is inevitable in clinical trials.
🧠
Modified ITT (mITT) includes all randomized patients who received at least one dose of study medication and had at least one post-baseline assessment.
🧠
Last observation carried forward (LOCF) and multiple imputation are methods to handle missing data while maintaining the ITT principle.
🧠
Board pearl: Even with missing outcome data, participants should be analyzed in their originally assigned groups to preserve the ITT principle.
Non-Inferiority Trials: When PP Takes Priority
⚡
Non-inferiority trials aim to show a new treatment is "not worse" than standard therapy by more than a predefined margin.
⚡
ITT analysis in non-inferiority trials can falsely suggest non-inferiority by diluting differences between treatments.
⚡
PP analysis provides a more stringent test of non-inferiority by comparing treatments in those who actually received them.
⚡
Regulatory agencies typically require both ITT and PP analyses for non-inferiority trials, with concordant results strengthening conclusions.
⚡
Board pearl: If both ITT and PP show non-inferiority, the conclusion is robust.
The As-Treated Analysis Alternative
📌
As-treated analysis groups participants based on the treatment they actually received, regardless of randomization.
📌
This approach is useful for safety analyses where you want to know adverse effects in those actually exposed to treatment.
📌
However, as-treated analysis completely breaks randomization and is highly susceptible to confounding.
📌
Board distinction: ITT analyzes by randomized group, PP excludes protocol violators, and as-treated analyzes by actual treatment received.
Statistical Power and Sample Size Implications
📣
ITT analysis typically requires larger sample sizes than PP analysis because the dilution effect reduces the observed treatment difference.
📣
Trials must be powered assuming realistic rates of non-adherence and dropout to detect clinically meaningful differences under ITT.
📣
High dropout rates can severely compromise statistical power even when using ITT analysis.
📣
PP analysis appears more "efficient" but at the cost of potential bias.
📣
Board pearl: A trial powered for PP analysis may be underpowered when analyzed by ITT.
Interpreting Discordant ITT and PP Results
🔸
When ITT shows no effect but PP shows benefit, consider whether non-adherence or crossover diluted the ITT result.
🔸
When ITT shows benefit but PP shows no effect, the result may be spurious due to selective dropout of treatment failures.
🔸
Large discrepancies between ITT and PP suggest problems with trial conduct (poor adherence, differential dropout).
🔸
Board pearl: Concordant ITT and PP results strengthen confidence in trial findings; discordant results require careful interpretation.
The Role of Adherence in Treatment Effect
🧷
Poor adherence dilutes treatment effects in ITT analysis but reflects real-world effectiveness.
🧷
Trials with run-in periods (where non-adherent patients are excluded before randomization) produce ITT results closer to PP results.
🧷
Adherence itself may be influenced by treatment assignment (side effects, early efficacy) creating informative censoring.
🧷
Board distinction: Efficacy (PP analysis) answers "Can it work?" while effectiveness (ITT analysis) answers "Does it work in practice?
Regulatory Perspectives and Publication Standards
📍
FDA and other regulatory agencies require ITT as the primary analysis for superiority trials to ensure conservative efficacy estimates.
📍
CONSORT guidelines mandate reporting participant flow through the trial, enabling readers to understand the difference between ITT and PP populations.
📍
Journals increasingly require both ITT and PP results to be reported for transparency.
📍
Board pearl: A trial reporting only PP results should raise concerns about potential bias and selective reporting.
Subgroup Analyses Within ITT Framework
🔹
Subgroup analyses should maintain the ITT principle — analyze participants in randomized groups regardless of adherence within the subgroup.
🔹
Post-hoc identification of "responders" and analyzing only this subset violates ITT principles and introduces severe selection bias.
🔹
Pre-specified subgroup analyses within an ITT framework maintain some protection against bias.
🔹
Board clue: "Analysis of treatment responders only" indicates a biased post-hoc analysis, not true ITT.
Crossover Contamination in Control Groups
⭐
In trials of surgical procedures or devices, control group participants may seek the intervention outside the trial (contamination).
⭐
ITT analysis counts these crossovers as control participants, underestimating the true treatment effect.
⭐
This is particularly problematic in trials with long follow-up periods where contamination accumulates over time.
⭐
Board pearl: High crossover rates from control to intervention suggest the treatment is perceived as beneficial by participants.
Time-to-Event Analyses and ITT
✅
For survival analyses, ITT requires following all randomized patients until death or end of study, regardless of treatment received.
✅
Censoring participants when they stop treatment (PP approach in survival analysis) can introduce immortal time bias.
✅
Competing risks must be considered — participants who die from other causes are still included in ITT analysis.
✅
Board distinction: In Kaplan-Meier curves, ITT includes all randomized patients from time zero; PP censors at protocol violation.
Pragmatic Trials and the ITT Imperative
🧠
Pragmatic trials aim to inform real-world clinical decisions and almost exclusively use ITT analysis.
🧠
These trials have broad inclusion criteria and flexible protocols, making PP analysis less meaningful.
🧠
ITT results from pragmatic trials directly inform clinical practice guidelines and policy decisions.
🧠
Board pearl: Effectiveness (pragmatic trials with ITT) may differ substantially from efficacy (explanatory trials with PP).
Common Misconceptions in Board Questions
⚡
ITT does not mean "ignore dropouts" — it means analyze them in their assigned group.
⚡
PP is not always a subset of ITT — modified ITT may exclude some randomized participants that PP includes.
⚡
ITT is not always larger than PP population — early dropouts before any treatment may be excluded from mITT but included in ITT.
⚡
Board trap: Don't assume ITT always shows smaller treatment effects — if dropouts are treatment failures, ITT might show larger effects.
Sensitivity Analyses and Robustness
📌
Best practice involves conducting both ITT and PP analyses as sensitivity analyses to test robustness of findings.
📌
Additional sensitivity analyses might include different methods for handling missing data or varying definitions of protocol adherence.
📌
If results are consistent across analytical approaches, confidence in findings increases.
📌
Board pearl: A treatment showing benefit in both ITT and PP analyses with different missing data methods has robust evidence of efficacy.
Board Question Stem Patterns
📣
"All randomized patients were analyzed in their assigned groups" → ITT analysis
📣
"Only patients who completed 80% of study medication were analyzed" → PP analysis
📣
"Analysis included all patients as randomized regardless of adherence" → ITT analysis
📣
"Patients who crossed over were excluded from analysis" → PP analysis
📣
"Primary analysis was conservative to avoid overestimating benefit" → ITT for superiority trial
📣
"Analysis aimed to show true biological effect of treatment" → PP analysis
📣
"Non-inferiority was tested in the adherent population" → PP for non-inferiority trial
One-Line Recap
🔸
Intention-to-treat analysis maintains randomization benefits by analyzing all participants as randomized regardless of adherence, providing conservative real-world effectiveness estimates for superiority trials, while per-protocol analysis examines only protocol-adherent participants, offering insight into treatment efficacy under ideal conditions and serving as the preferred approach for non-inferiority trials.
bottom of page