FR
Русский English Français
Se connecter Créer un compte
YouEDC Biostat

Analysis plan and estimands

The question of the study as an estimand of ICH E9(R1) and a target trial, written before the data, checked against the eCRF and kept as versions.

An analysis is honest when its question was written down before the data could answer it. The Analysis plan keeps that question in the study itself: the estimand of ICH E9(R1) and the elements of a target trial, each tied to the eCRF field that will carry it. The portal reads the plan against the eCRF and says what does not hold, and every save becomes a version that is never changed.

What the check looks at: time zero time zero randomisation, or the start of the treatment at or before it eligibility, treatment, confounders after it the outcome, intercurrent events treatment known only here immortal time: the check calls it a problem A field's timing: the days of the visits with its form in the eCRF in force; no data is read.
Everything that decides who is compared must be known when follow-up starts.

The estimand

An estimand says exactly what the study estimates. Its five parts are in the plan:

  • Population: who is eligible, in words, and the eligibility fields of the eCRF.
  • Treatment and comparator: in words, and the field that tells which one a patient is on. In a randomised trial it is the randomised arm (ARM).
  • Variable: the endpoint's field and what is taken from it, for example the change from baseline to week 24.
  • Intercurrent events: what happens after the start and changes the meaning of the variable (stopping the drug, rescue therapy, death), each with its strategy: treatment policy, hypothetical, composite, while on treatment or principal stratum.
  • Population-level summary: the difference in means, a risk ratio, a hazard ratio and so on.

A plan holds up to six estimands: one primary, the others secondary or supportive.

The target trial

An observational study is read as the trial it imitates. Three of its elements are in the plan: time zero, when eligibility is met, the treatment is assigned and follow-up begins (the enrolment or a date in the eCRF); the confounders, what affects both the choice of treatment and the outcome; a negative control outcome, one the treatment cannot cause, whose difference between the groups shows confounding left over. In a randomised trial the confounders list is the covariates of the adjustment.

The design check

The check runs on every page of the plan and on Check without saving. It looks at which visits hold each field and on which day of the schedule:

  • every named field exists in the eCRF in force;
  • eligibility, the treatment and the confounders are collected at or before time zero; a treatment known only later makes immortal time, a confounder measured later may already be affected by the treatment;
  • the outcome is collected after time zero;
  • an observational study keeps the start date of the treatment, so that new users are told from patients already on it, and names a negative control;
  • the summary suits the variable's type, and a hypothetical or while-on-treatment strategy has the date of its event.

Each finding is marked Problem, Look at it, Not filled or In order. The check advises; it never stops a save.

Versions and the audit trail

The manager, the data manager or the statistician edits the plan; whoever sees YouEDC Biostat reads it. Each save writes a new version with its SHA-256, the eCRF version and the number of patients enrolled at that moment, and the counts of its check. A version is never changed, even by the database. Once a patient is enrolled, a change needs a reason. JSON downloads the exact text of a version: its SHA-256 is the one in the list of versions and in the audit trail, so anyone can show that the plan came before the data.