Diagnosing Absorption Misspecification From Early-Time Patterns

Why Early-Time Patterns Matter

After an oral dose, absorption shapes the rise in concentration and the timing and sharpness of the peak. A model that describes absorption poorly can therefore show systematic disagreement soon after dosing even when its later predictions look reasonable. Early-time residuals and individual concentration overlays help locate that disagreement in time and suggest whether the model’s absorption is too delayed, too rapid, or insufficiently prolonged.

These plots are diagnostic evidence, not a stand-alone test of a particular absorption mechanism. A similar pattern can arise if dose or sample times were recorded inaccurately, if early samples were handled differently by an assay, or if the residual-error model gives the wrong scale to those observations. The useful question is not simply whether early residuals are large, but whether their direction and timing are coherent across individuals and remain after plausible data and measurement explanations have been checked.

Reading the Sign and Timing of Residuals

Suppose the fitted model predicts a concentration curve \(C(t)\), but the actual curve is shifted later by a small time \(\delta>0\). The delayed curve can be represented approximately as \(C(t-\delta)\). A first-order Taylor expansion gives:

On the rising part of the predicted curve, \(C'(t)>0\), so a delayed observed curve tends to lie below the prediction: residuals are negative. After the peak, while the curve is falling, \(C'(t)<0\), so the delayed curve tends to lie above the prediction: residuals are positive. A model that absorbs too rapidly relative to the data can show the reverse sequence: positive residuals during the rise and negative residuals after the model has peaked too soon.

This argument describes a time shift, not every possible difference in absorption. A model that absorbs too briefly may produce a profile that rises and peaks too narrowly, while data with more prolonged input can stay below that prediction early and above it later. The exact pattern depends on the dose, sampling times, structural model, and other parameters. It is therefore more reliable to describe what the residuals do than to label a mechanism from a single point or plot.

Individual Overlays and the Population Pattern

An individual overlay places that person’s observations and predicted concentration-time curve on the same time axis. For this diagnosis, the plot should make dose time, actual sample times, and the early rising and peak regions easy to see. Where useful, display the population prediction and individual prediction distinctly: they answer different questions. A population prediction shows the typical model profile, whereas an individual prediction incorporates the person-specific parameter estimates used by the fitted model.

Individual predictions can help reveal whether the model can describe the observed profile after accounting for estimated between-individual differences. But they are not independent confirmation: the observations helped estimate those individual parameters. A good-looking individual curve does not by itself rule out structural misspecification, particularly with sparse early samples or substantial parameter shrinkage. Compare overlays with residuals and inspect whether discrepancies recur in subjects with enough early observations to show the curve shape.

Distinguishing Absorption Misfit From Timing and Assay Effects

Before changing the structural model, verify the information that determines where each observation belongs on the time axis. Check that dose times, sample times, time zones or clock conventions, and dose-to-sample intervals are correct. If a sample was recorded later than it was actually collected, plotting it at the recorded time can create an apparent phase shift. A dose-time error can affect every post-dose time for that occasion. If the pattern disappears after correcting a documented timing error, it was not evidence that the absorption model needed to change.

Next, look for clustering by assay, analytical run, site, or study. An additive assay offset can move early observations in one direction without changing the underlying profile shape. A scale difference can make discrepancies increase with concentration. If early samples were disproportionately measured in one batch, an assay-related pattern may mimic time-dependent structural misfit. Earlier material on residuals by study, site, and assay and residual diagnostics under proportional and combined error provides relevant checks: examine whether the trend remains after accounting for measurement scale and grouping.

When Absorption Is Too Brief in the Model

A model may also fail because its input ends too quickly relative to the data. In that case, observed concentrations can remain lower than the fitted curve during the early rise and higher than it later, as the observed profile continues to receive input after the model’s predicted rise has largely passed. The transition may be gradual rather than a clean reversal at a single time. Individual overlays can show whether the observed rise or peak is broader than the prediction, while early and near-peak residuals indicate which parts of the curve are missed.

Assumptions, Failure Modes, and Alternatives

These interpretations assume that the dose and sample times are reliable, observations are assigned to the correct dose and occasion, concentrations are measured on a consistent basis, and residuals are interpreted using a scale appropriate to the fitted residual-error model. They also assume that the model predictions being compared refer to the same individual, dose, and covariate conditions as the observations. When these conditions are doubtful, a sign pattern may describe data recording or measurement rather than absorption.

Useful diagnostics complement rather than replace one another. A residual-versus-time plot summarizes direction and spread across observations; individual overlays show whether those residuals correspond to a coherent mismatch in profile shape. A plot of residuals against dose or regimen can reveal that a pattern is confined to a particular dosing condition. Review of assay and sampling records targets nonstructural explanations. A more complex absorption model is an alternative to the current model, but it should be considered only after these checks and judged with the same diagnostic scales and relevant data, not selected simply because it can follow a few points more closely.

• Too few early samples: the data may not distinguish a delayed rise from a different peak time. Do not infer a precise absorption process from an unobserved curve segment.
• Unequal sampling: a pooled residual plot can be dominated by individuals with many samples. Check individual patterns and how observations are distributed over time.
• Large individual differences: a population-level pattern may reflect a subgroup or a few influential profiles. Check whether it is broadly repeated, rather than relying on the pooled trend alone.
• Residual scale effects: a concentration-dependent spread can make raw residual magnitudes misleading. Interpret residual direction and use the appropriate standardized scale, as discussed in residual diagnostics under proportional and combined error.
• Confounding with other model features: early residuals do not uniquely identify absorption. The same observed concentration can be affected by other model components and by measurement error. Avoid using this early-time diagnosis to settle questions that require examining other portions of the profile.