Circadian and Placebo Effects in QTc Exposure-Response

Why Temporal Effects Need a Control

QTc can change over a study day even when drug concentration is zero. Circadian rhythms, meals, posture, measurement conditions, visit procedures, and ordinary variation may all contribute to observed changes. A rise from a participant’s baseline is therefore not, by itself, evidence of a drug effect. The relevant comparison is the QTc change expected under the same study conditions without active drug exposure.

In concentration-QTc analysis, time-matched placebo data help estimate that control trajectory. Study-time covariates describe systematic changes associated with when an ECG is taken, while the concentration term describes the modeled association with drug exposure after accounting for those changes. The interpretation depends on having comparable control and active-treatment observations and on representing the timing effects adequately.

A Model for Concentration and Study-Time Effects

Let \(Y_{ij}\) denote the baseline-adjusted QTc change, in milliseconds, for participant \(i\) at ECG occasion \(j\). Let \(A_i=1\) for active treatment and \(A_i=0\) for placebo, and let \(C_{ij}\) be the concentration aligned with that ECG, in a stated concentration unit. In the placebo arm, \(C_{ij}=0\). A useful mean model is

Here, \(\phi_{ij}\) is clock time, \(m_{ij}\) is time since the relevant meal, \(v_{k(ij)}\) represents visit or scheduled-occasion effects, and \(t_{ij}\) is study time, such as hours since dose. The functions \(f\), \(g\), and \(p\) describe the modeled clock-time, meal-timing, and study-time patterns. Visit effects may be categorical. The coefficient \(\delta\) allows an active-arm offset not explained by concentration; \(\beta_C\) is the concentration slope. The participant-specific term \(b_i\) and residual \(\varepsilon_{ij}\) account for repeated observations using a suitable mixed-effects structure.

These terms are not a requirement to include every possible time variable in every analysis. The schedule determines which effects can be distinguished and which terms are supportable. For example, if every ECG at a particular study time is always taken at exactly the same clock time, clock time and study time are not independently varied. Their separate effects cannot be identified from those data merely by including both labels in a model.

A periodic clock-time effect can be represented by a 24-hour harmonic, for example \(f(\phi)=a\cos(2\pi\phi/24)+b\sin(2\pi\phi/24)\), where \(\phi\) is in hours and \(a\) and \(b\) are in milliseconds. Flexible functions or prespecified time categories may be more suitable when the shape is not sinusoidal. Meal timing can likewise be represented using prespecified categories or a smooth function if the design and sample size support it. These choices describe a mean pattern; they do not establish a biological cause for that pattern.

A shared study-time function \(p(t)\) represents a temporal pattern common to the treatment arms. If a treatment-specific time pattern is plausible and the design provides information to estimate it, an additional active-treatment-by-time term may be considered. Without such a term, the model assumes that any active-versus-placebo difference not captured by concentration is adequately represented by \(\delta\), rather than by a distinct time-varying effect. A treatment-by-time term must be considered carefully because it can be difficult to distinguish from a concentration effect when concentration changes systematically with time.

Why Time Matching Separates Shared Changes

The basic justification is cancellation of temporal effects in a matched comparison. Suppose the expected baseline-adjusted QTc change under placebo at a particular set of timing conditions \(z\) is \(\mu(z)\). Under a model with a concentration effect, an active observation at those same conditions has expected change \(\mu(z)+\delta+\beta_C C\). Subtracting the placebo expectation gives

The shared temporal component \(\mu(z)\) cancels. If the arm offset \(\delta\) is zero or otherwise accounted for, the remaining contrast is the modeled concentration-related increment. In practice, control and active observations may not fall at identical clock and meal times. The model then estimates the placebo expectation at the active observation’s conditions; the quality of that adjustment depends on the model and on temporal overlap in the data.

This reasoning also clarifies what “placebo change” means. It is the observed control-arm trajectory under the study schedule. It can include circadian change, study procedures, measurement conditions, expectations, and other factors shared with the active arm. It is not automatically a pure physiological circadian curve, nor does it identify the causal contribution of each component.

Worked Comparisons and Model-Based Estimation

Diagnostics, Alternatives, and Limitations

A temporal adjustment is credible only if the observations can support it. Inspect the active and placebo schedules by clock time, time since dose, meal interval, and visit. Plot baseline-adjusted QTc changes against these variables by arm, and plot concentration against each time variable. Look for gaps where the model must extrapolate a placebo trajectory, sparse cells, and near-perfect alignment between concentration and study time. A smooth-looking fitted curve does not solve a lack of overlap.

Residual patterns can reveal an inadequate mean structure: systematic residual trends over clock time or study time suggest missing or misspecified temporal effects. Large differences between observed and predicted placebo changes at particular visits, or arm-specific patterns left in residual plots, also warrant investigation. Repeated observations require attention to within-participant correlation; standard errors that ignore that correlation can be too small. Comparisons across reasonable, prespecified representations of timing effects can show whether the concentration estimate is sensitive to those choices.

The main alternatives answer related but different questions. A simple baseline-adjusted comparison without control data describes change from baseline but cannot separate drug-related change from shared temporal change. A time-matched active-placebo contrast is easy to interpret when timing is closely aligned, but can be noisy and does not use all observations as efficiently as a model. A regression or mixed-effects model can adjust for several timing variables and repeated measurements, but depends on its functional form and overlap assumptions. Time categories are transparent but may use parameters inefficiently; smooth functions can represent gradual patterns but may overfit sparse schedules.

The placebo trajectory is a reference, not a guarantee that active and control participants would have identical QTc changes absent exposure. Randomization, comparable procedures, consistent ECG handling, and adequate representation of schedule effects support that comparison. If meal timing, visit procedures, or ECG collection differ systematically by arm, adjustment may not remove the resulting bias. Nor can a model reliably separate two effects when the design provides no independent variation—for example, if each concentration is observed only at one distinct study time and no informative time-matched controls are available.