Example

​

“Analysis populations are defined as follows:

• Intention to treat (ITT): this population includes all randomised participants regardless of whether they were later found to be ineligible, did not adhere to the protocol or were never treated.

• Per protocol (PP): this population contains all randomised participants who received their allocated trial treatment without major protocol deviations.

• Safety population: this population contains all randomised participants who received at least one dose of trial IMP and will be classified according to the actual treatment received.

The analysis of the primary outcome will be in the ITT population with sensitivity analysis in the PP population. The safety population will be used to report side effects ” [419].

 

Explanation

​

To preserve the benefits of randomisation, all randomised participants need to be included in the analysis and retained in the group to which they were allocated. An “intention-to-treat” analysis population is defined by the inclusion of outcome data from all participants in their originally allocated group – this is widely recommended as the preferred analysis population [361, 420, 421]. Attrition bias can be avoided when outcome data are obtained from all participants and included in the analysis, regardless of protocol adherence. While imputation of missing outcomes would allow an intention-to-treat analysis, it does not guarantee avoidance of bias except under strong assumptions about the missing data.

 

Many analyses are described as “intention-to-treat” but apply variable definitions in terms of handling missing outcome data (Item 27c) or excluding participants who deviate from the intervention or follow-up protocols [422, 423]. Some trials will define a “per protocol” analysis population that includes participants completing the study with no major protocol deviations. Exclusion of data from protocol non-adherers can compromise the randomisation and introduce bias, particularly if the frequency of and the reasons for non-adherence vary between the trial groups (e.g., “healthy adherers” bias) [424]. Depending on the aim of the analyses (Box 1), other analysis populations may be planned, and their rationale and definition should be explained.

 

A meta-epidemiological study of 322 comparisons from 310 randomised trials found that analyses deviating from intention-to-treat produced larger intervention effect sizes than those applying the intention to treat principle [425]. Other analysis populations may be planned (e.g., a safety population) and their rationale and definition should be explained. Reviews of two samples of 108 and 292 trial protocols from 2016 reported that 69% to 74% specified the analysis population [9, 10]. Other reviews have found frequent discrepancies between the main analysis population described in protocols compared with reports of completed trials (or systematic reviews) [64].

 

The protocol should clearly define the primary analysis population (and any other populations, as applicable). This includes addressing whether all randomised participants (completely observed outcomes or imputed outcomes) or a subset of randomised participants with observed outcomes will be included in the analysis, and in which trial group. Ambiguous labels such as an “intention-to-treat” or “modified intention-to-treat” analysis should be avoided unless they are fully defined in the protocol.

 

Summary of key elements to address

• Who will be included in the primary and other analyses (e.g., all randomised participants with either observed or imputed outcome data)

  • Any exclusions due to missing data or other reasons

• Trial group in which participants will be analysed (e.g., as-randomised)