Two Different Ways a Previous Period Can Matter
In a crossover study, a participant receives treatments in different periods. The intended comparison concerns the effect of the treatment assigned in the current period, but a previous treatment may still matter when that period begins. It can matter because drug remains in the body, or because a drug-induced response has not yet recovered. These are related possibilities, but they are not the same phenomenon.
The distinction is practical. Plasma concentration measurements provide evidence about exposure in the sampled compartment, at the sampled times, and for the analytes measured. A pharmacodynamic outcome measures a response, which can have its own delay and recovery. Therefore, finding little drug in plasma does not establish that the response has returned to its untreated state.
The earlier material on first-order carryover represents a prior treatment's contribution to a later-period outcome. The PK/PD distinction describes possible biological mechanisms for that contribution; it does not, on its own, tell us whether a particular carryover effect is statistically important or how large it is.
Residual Drug Exposure: A Simple PK Model
A useful illustration is a one-compartment concentration that declines by first-order elimination after input has ended. Let \(C(t)\) be the plasma concentration \(t\) hours after the end of the prior dosing period, and let \(C_0\) be its concentration at that time. Under this model, the elimination rate constant \(k\) is positive and constant, and there is no further drug input:
This equation assumes the post-dose decline is adequately described by a single exponential over the time considered. It does not account, for example, for ongoing absorption, multiple disposition phases, active metabolites, or variation in elimination across participants. Those features can make the actual concentration-time course differ from this illustration.
Worked Example: Residual Plasma Concentration After a Prior Dose
Suppose a hypothetical participant has a concentration of \(80\ \text{ng/mL}\) when the prior treatment period ends. Assume a terminal elimination half-life of 6 hours and first-order decline with no further input. What concentration does the model predict 24 hours later?
For first-order elimination, \(k=\ln(2)/t_{1/2}\). Thus,
Substituting \(C_0=80\ \text{ng/mL}\) and \(t=24\ \text{h}\),
Under the stated model, \(5.00\ \text{ng/mL}\) remains at 24 hours. Whether that residual concentration can affect the later-period outcome depends on the drug, analyte, target, and endpoint. The calculation establishes a model-based concentration prediction; it does not establish that the carryover is either relevant or irrelevant.
Detectability and relevance are separate questions. A concentration can be below an assay's quantification limit yet still have biological activity. Conversely, a measurable concentration need not produce a meaningful effect on the endpoint. Active metabolites, drug in tissues, and differences between plasma and the site of action can further complicate the connection between a plasma measurement and the response.
Why the Response Can Lag Behind Plasma
One possible explanation for a mismatch between plasma concentration and response is delayed equilibration between plasma and the site where drug acts. A simple effect-compartment model represents this lag by an effect-site concentration \(C_e(t)\):
where \(k_{e0}\) is a positive equilibration rate constant. This model assumes the effect-site concentration moves toward plasma concentration at a rate proportional to their difference. It is a simplified description: it does not by itself specify the response, and it may not capture all mechanisms of delayed or persistent action.
For a concrete derivation, suppose plasma declines as \(C(t)=C_0e^{-kt}\), and the effect-site concentration at \(t=0\) is \(C_e(0)\). Multiplying the differential equation by the integrating factor \(e^{k_{e0}t}\) gives
Integrating from 0 to \(t\), for \(k\ne k_{e0}\), yields
Dividing by \(e^{k_{e0}t}\) and simplifying gives
The expression shows why plasma concentration at one time is not necessarily a good proxy for effect-site concentration at that time: the latter depends on the prior concentration history and on how rapidly the two compartments equilibrate.
Worked Example: Plasma Declines Faster Than the Effect Site
Consider a hypothetical drug with plasma concentration \(C_0=100\ \text{ng/mL}\) at the start of a drug-free interval and plasma half-life 4 hours. Assume the effect-site concentration is initially \(100\ \text{ng/mL}\), and its equilibration half-life is 12 hours. Use the effect-compartment model above to compare the predicted plasma and effect-site concentrations after 24 hours.
The rate constants are
The plasma prediction is
For the effect site, substitute \(C_e(0)=C_0=100\ \text{ng/mL}\) into the derived solution:
The model predicts a low plasma concentration alongside a substantially higher effect-site concentration at 24 hours. This is a consequence of the stipulated slow equilibration, not a general rule for drugs. It also does not guarantee a particular response: the relation between \(C_e\) and the endpoint must be specified separately.
Effects Can Persist After Drug Concentration Falls
A different mechanism is a delayed recovery of the biological response itself. A drug may initiate a process that continues after plasma drug has declined, such as slow receptor recovery, persistent target modification, or a downstream physiological change. A simple illustration is an effect deviation \(D(t)\) that returns toward baseline by first-order recovery after the initiating drug effect has stopped:
This model assumes no continuing drug-driven input to the response during the interval and a constant recovery rate \(k_{\text{out}}>0\). It is a compact approximation, not a universal model of PD recovery. In particular, it does not say that every response recovers exponentially or that plasma concentration is exactly zero.
Worked Example: A Persistent Response With Negligible Plasma Drug
Suppose a hypothetical treatment changes a pharmacodynamic endpoint by \(40\) response units at the start of a drug-free interval. Assume the drug is no longer driving the response and that the response deviation has a 30-hour recovery half-life. What deviation does the model predict after 60 hours?
The recovery rate is \(k_{\text{out}}=\ln(2)/30=0.02310\ \text{h}^{-1}\). Therefore,
The modeled response remains 10.0 units from baseline after 60 hours, even though the assumptions allow the plasma drug to have become negligible. The example illustrates PD persistence; it does not infer a plasma concentration from the response or establish that the same recovery model applies to a real endpoint.
What Each Measurement Can Establish
The main alternatives are not competing definitions but different kinds of evidence. Plasma concentration data describe measured exposure; pharmacodynamic data describe the response; a mechanistic PK/PD analysis attempts to connect them. Each has a different scope.
| Evidence | What it can help assess | Important limitation |
|---|---|---|
| Plasma concentrations | Residual measured drug or metabolite in plasma at sampled times | May miss unmeasured analytes, unsampled concentration patterns, tissue exposure, or a response that persists after plasma decline |
| Pharmacodynamic endpoint | Whether a relevant response remains changed from its expected baseline | The endpoint may be noisy, delayed, insensitive, or influenced by factors other than the prior treatment |
| PK/PD model | Whether concentration history and response dynamics are consistent with a proposed mechanism | Conclusions depend on model structure, parameter information, measurement quality, and assumptions |
In a crossover study, evidence of residual exposure or response matters because the outcome in a later period may reflect both the current treatment and remnants of the preceding treatment. The standard two-period interpretation relies on assumptions about carryover; if those assumptions are not plausible, the within-subject contrast may not represent the intended direct treatment comparison. As established in the earlier material on identifiability of carryover in a two-period design, sequence, period, and carryover contributions cannot generally be disentangled just by adding a carryover term to that design's model.
Diagnostics, Failure Modes, and Limits
A useful assessment brings together the expected mechanism, concentration-time information, and the endpoint's behavior. Look for whether the analyte panel includes relevant active metabolites, whether sampling can represent the concentration history of interest, and whether the response is known to have delayed onset or recovery. A single low plasma result is especially limited: it is a measurement at one time, not proof that exposure was absent throughout the interval or that the PD state had reset.
- Assuming an assay limit is a biological threshold. The quantification limit describes the assay, not necessarily the concentration below which a drug has no effect.
- Measuring only parent drug. An active metabolite or drug at a site not represented by plasma may remain relevant when the parent plasma concentration is low.
- Equating concentration with response. A concentration-response relationship may involve delay, indirect response, or persistent target changes; contemporaneous plasma and response need not track closely.
- Treating an unchanged endpoint as proof of no carryover. A measure can be too variable or insensitive to detect a remaining effect. Lack of a visible difference is not, by itself, evidence of absence.
These checks do not provide a universal numerical cutoff for deciding that carryover is absent. Such a decision depends on the drug and metabolite, the assay, the target, the endpoint, and the study's inferential purpose. Nor does observing a persistent response identify its cause automatically: natural time trends and other influences can also change an endpoint.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A prior treatment's plasma concentration is below the assay's quantification limit at the beginning of a later period. What can and cannot be concluded about PD carryover?
- Under the first-order elimination model, a concentration is \(32\ \text{ng/mL}\) at the start of an interval and the half-life is 8 hours. What concentration is predicted 16 hours later?
- Give one reason an effect-site concentration might remain above plasma concentration while plasma concentration is declining.
- A response deviation is 24 units and, under a first-order recovery model, its recovery half-life is 10 hours. What deviation is predicted after 20 hours?
Solutions. 1. The result indicates that the measured plasma analyte is below that assay's quantification limit at the sampled time. It does not establish that the drug is biologically inactive, that other active analytes are absent, or that a prior drug-induced response has returned to baseline. 2. Sixteen hours is two half-lives, so \(C(16)=32(1/2)^2=8.00\ \text{ng/mL}\), under the stated first-order model. 3. With slow equilibration, the effect site responds to the history of plasma concentrations and can decline more slowly than plasma; the effect-compartment model represents this lag. 4. Twenty hours is two recovery half-lives, so \(D(20)=24(1/2)^2=6.00\) response units, assuming no continuing drug-driven input and first-order recovery.