Placebo Run-In Data and Baseline Definition

Why Baseline Definition Matters

In a placebo-controlled study with a placebo run-in, participants may be measured at screening, during or at the end of the run-in, and again at randomization. These measurements are not interchangeable. A participant may improve during the run-in as a placebo response develops, worsen as the disease progresses, or fluctuate because of measurement error. The value chosen as baseline determines which of these changes are treated as pre-randomization history and which are modeled as change after treatment assignment.

A baseline definition should therefore state both which measurement is used and the time origin for the model. In a randomized parallel trial, randomization is commonly the origin for post-randomization treatment comparisons. Screening can still be useful for describing run-in change, evaluating eligibility, or adjusting for an earlier measure, but it does not automatically replace the randomization baseline.

Centering Placebo and Progression Components

Let \(Y(t)\) denote the clinical outcome at time \(t\), with larger values representing better status. Let \(t=0\) be screening. For clarity, suppose the expected outcome before randomization can be represented by a starting level plus a disease-progression component and a placebo component. The components are defined relative to screening:

Here \(B_S\) is the expected screening outcome, \(G(t)\) describes the natural-history or disease-progression contribution, and \(P(t)\) describes the placebo contribution. The changes in braces are relative to screening. This representation does not assert that every observed within-person change is caused by placebo or progression: observed data also contain measurement variability and other individual differences.

If the model is instead centered at randomization time \(t_R\), subtract the expected outcome at \(t_R\). For a participant assigned to placebo, the expected post-randomization change is then

The run-in change is already part of the randomization-level outcome. It is not counted again as a post-randomization placebo effect. For an active-treatment arm, the model may add a treatment contribution after randomization; the randomized treatment contrast is evaluated over the same post-randomization period and against the same baseline convention.

Worked Example: Decomposing Improvement During a Placebo Run-In

Worked Example: What Changes When the Baseline Is Randomization?

A Short Derivation: Re-anchoring a Model

Re-anchoring does not alter the underlying expected outcome; it changes which value is called zero change. Starting with the screening-centered model, subtract its value at randomization:

The screening level and all component values at screening cancel. This identity explains why run-in response is included in the randomization-level outcome but does not appear as a further increment after randomization. It also shows why analysts must use a consistent time origin when estimating placebo and progression parameters.

Worked Example: Comparing Candidate Baselines

Choosing a Baseline and Checking the Design

The randomization measurement is often the most direct reference for a post-randomization treatment contrast because it is measured just before assigned treatment begins. Screening remains important: it may be used to define eligibility, describe the run-in trajectory, or provide a covariate for analysis. An end-of-run-in value can also be informative when the protocol explicitly defines it as the reference. The choice should match the scientific question and the timing of treatment assignment, not simply whichever value produces a larger or smaller change.

Several design features affect interpretation:

• Run-in eligibility rules. If participants are excluded for lack of improvement, poor adherence, or unstable measurements during run-in, the randomized population is selected. Treatment results then describe that selected population and may not generalize to all screened patients.
• Measurement error and regression to the mean. When eligibility depends on an unusually high or low screening value, a later measurement may be closer to the person’s typical level even without a true systematic change. A simple screening-to-randomization difference can therefore mix true change and measurement fluctuation.
• Different timing or measurement conditions. A screening value and a randomization value may be collected under different visit conditions, procedures, or intervals. Baseline choice should account for these differences and the protocol’s prespecified measurement rules.
• Model and data checks. Plot the observed run-in measurements by time and compare them with the assumed placebo and progression components. Check whether the number and timing of observations support estimating both components, and inspect whether run-in exclusions or missing measurements are concentrated in particular response patterns.

A constant progression rate and an exponential placebo onset are useful illustrative assumptions, not universal truths. Sparse observations during run-in may not distinguish a changing placebo response from progression or measurement variation. If the components are weakly identified, the fitted decomposition can be uncertain even when the total predicted trajectory is reasonably estimated. Baseline choice cannot by itself resolve that limitation.

Common Mistakes and Practical Comparisons

A common mistake is to say, “The outcome rose during placebo run-in, so the rise is the placebo effect that should be added again after randomization.” This double-counts the run-in response if the randomization measurement already contains it. The correction is to define the time origin explicitly and calculate post-randomization placebo change as \(P(t)-P(t_R)\), not as \(P(t)-P(0)\). Another mistake is to interpret a change score as a causal placebo effect. A change score is a difference between measurements; causal attribution requires an appropriate comparison and assumptions about the design.