Why Timing Is Part of the Exposure–Effect Pair
An exposure value does not describe an effect by itself. It must be paired with a pharmacodynamic (PD) observation whose timing gives the comparison a meaningful interpretation. A concentration measured at one instant, an average concentration over several hours, and an AUC over a dosing interval summarize different exposure histories. Even when those quantities come from the same concentration-time profile, they are not interchangeable.
For an instantaneous PD measurement, a proposed exposure summary should describe a period that is relevant to that measurement and should not silently include exposure occurring afterward. For a PD response summarized across an interval, the exposure window should be aligned with that interval unless a justified biological delay calls for a different pairing. Alignment is a matter of defining the question precisely; it does not, by itself, show that exposure caused the observed effect.
The time origin matters. “Two to four hours” could mean hours after a dose, after an infusion began, after the first dose, or after a study visit began. These are not equivalent unless the design makes them equivalent. Record actual times when available; a nominal visit label is not a substitute for a timestamp if the measurement occurred substantially earlier or later than planned.
Specify the Effect Time Before Summarizing Exposure
First identify what the PD datum represents. An effect measured at 8 hours is an observation at a time point, even if the assay takes several minutes. A mean symptom score recorded from 10 to 12 hours, by contrast, summarizes an interval. A sample collected at a particular time may also be analyzed later; the analysis time is not the biological observation time.
For a point measurement at time \(t_E\), write down \(t_E\) and its reference point, such as hours after dose. Then state the exposure window \([a,b]\) and metric. A window ending at \(t_E\) is temporally prior to or contemporaneous with that observation, but it is not automatically biologically appropriate. If the response is expected to lag exposure, a prior window may be more plausible; the lag and window still require a defensible rationale rather than being selected just because they produce a clear pattern.
For an effect summarized over \([u,v]\), state both interval endpoints. A natural starting point is to compare it with exposure summarized over \([u,v]\). If the scientific question instead concerns exposure before the effect interval, specify that distinct window and explain the intended lag. Do not label a pre-interval exposure summary as though it were exposure during the effect interval.
Calculating a Window Summary
AUC over a window is the area under the concentration-time curve within its stated boundaries. Average concentration over that window is the AUC divided by the window duration. When concentrations are observed at discrete times, the trapezoidal rule approximates the area between consecutive observations by treating the concentration change as linear. If a window endpoint falls between observations, the analyst must obtain an appropriate concentration estimate at that endpoint, commonly by linear interpolation when that is suitable for the data and analysis.
Here the times \(t_0,\ldots,t_k\) span the window from \(a\) to \(b\), and \(C(t_i)\) is concentration at time \(t_i\). If concentration is measured in mg/L and time in hours, AUC has units mg·h/L and average concentration has units mg/L. The calculation depends on the observed or estimated concentrations covering the window; it does not justify extrapolating beyond the data without stating and defending that step.
Worked Example: Pairing an Effect at 8 Hours With a Prior Window
Hypothetical data. A PD response is measured at 8 hours after a dose. Concentrations are recorded at 0, 2, 4, 6, and 8 hours as 0, 4, 8, 12, and 10 mg/L, respectively. Suppose the stated exposure summary is average concentration from 6 to 8 hours.
The PD timestamp is \(t_E=8\) hours after dose, and the exposure window is \([6,8]\) hours after dose. The relevant concentrations are 12 mg/L at 6 hours and 10 mg/L at 8 hours. Using the trapezoidal rule:
The window lasts \(8-6=2\) hours, so:
Thus the defined pair is the PD response at 8 hours and average concentration of 11 mg/L over the preceding 6-to-8-hour window. The time ordering is explicit and the units are consistent. This pairing is a defensible description of timing, not proof that the 6-to-8-hour exposure produced the response.
For comparison, the AUC from 0 to 8 hours is \(4+12+20+22=58\) mg·h/L, giving an average concentration of \(58/8=7.25\) mg/L. That is a different summary over a different window. Pairing it with the same PD datum changes the exposure question; the number is not an interchangeable calculation of the 6-to-8-hour exposure.
Check for Temporal and Interval Mismatches
A useful review is to draw or tabulate the PD time and exposure window on one time axis. This makes it easier to see whether the exposure summary precedes the effect, overlaps its interval, or extends beyond it. An overlap may be appropriate for an interval response, but overlap alone does not establish a biological mechanism. Similarly, a window entirely before a point effect may be appropriate if a delay is expected, but the lag should be stated.
State whether it is a point value or an interval summary, and record its actual time or interval relative to a defined origin.
Give its start, end, time origin, and summary metric. Do not rely on vague labels such as “recent exposure.”
Identify any exposure after the point effect, any mismatch between effect and exposure intervals, and any intended delay.
Confirm that concentration observations cover the window and that any interpolation or extrapolation is disclosed.
Worked Example: An Effect Interval and a Misleading Broad Window
Hypothetical data. A symptom score is averaged across 10 to 12 hours after dose. Concentrations are 0 mg/L at 0 hours, 4 mg/L at 4 hours, 6 mg/L at 8 hours, 9 mg/L at 10 hours, and 11 mg/L at 12 hours. Compare the exposure during the symptom interval with average concentration over the entire 0-to-12-hour period.
For the 10-to-12-hour exposure window, the trapezoidal AUC is:
The average concentration during the same two hours is \(20/2=10\) mg/L. This window directly matches the interval represented by the symptom average.
For the broader 0-to-12-hour window, the interval AUCs are \(8\) mg·h/L from 0 to 4 hours, \(20\) mg·h/L from 4 to 8 hours, \(15\) mg·h/L from 8 to 10 hours, and \(20\) mg·h/L from 10 to 12 hours. Therefore:
Both summaries are arithmetically valid, but they answer different timing questions. The 10-to-12-hour average summarizes exposure during the interval represented by the symptom score. The 0-to-12-hour average includes earlier exposure and should not be described as exposure specifically during the symptom interval. Which summary is scientifically useful depends on the intended exposure–effect question; the timing description must remain accurate.
Worked Example: Detecting a Window That Uses Future Exposure
Hypothetical data. A PD measurement is taken at 4 hours after dose. Concentrations at 0, 2, 4, 6, and 8 hours are 0, 8, 12, 8, and 4 mg/L. An analyst pairs the 4-hour PD value with average concentration from 4 to 8 hours.
The proposed window contains concentrations measured after the PD observation. Its AUC is:
The corresponding average is \(32/4=8\) mg/L. The arithmetic is correct, but this window includes four hours of exposure after the 4-hour effect measurement. It cannot be interpreted as exposure that had already occurred when that effect was measured.
For a contrast, the 2-to-4-hour AUC is \(\frac{8+12}{2}(4-2)=20\) mg·h/L, and the average is \(20/2=10\) mg/L. This window ends at the PD measurement and is temporally prior to it. Whether that prior window is the right biological summary is a separate question; the example establishes why the 4-to-8-hour summary is temporally mismatched for explaining an effect observed at 4 hours. Later measurements can sometimes be informative for other questions, but they must not be presented as antecedent exposure.
Diagnostics, Alternatives, and Limits
Timing problems are often visible before any model is fitted. Look for effect records with missing or inconsistent time origins, exposure summaries with unstated endpoints, windows that extend past point effects, and effect averages paired with exposure summaries from substantially different intervals. Also check whether concentration samples actually bracket the window boundaries. A calculated AUC may look precise even when the profile is sparsely sampled or requires unsupported extrapolation.
There are several common alternatives, each describing a different temporal question. An instantaneous concentration at the effect time describes concentration at that time; it does not summarize earlier exposure. AUC or average concentration over a prespecified interval summarizes cumulative or average exposure within that interval. A dosing-interval summary gives a common interval-level measure, but may be poorly aligned with a PD observation made partway through the interval. A lagged window summarizes exposure before an effect time, which can be useful when a delayed response is scientifically plausible. None is universally correct: the window and metric should be stated so the reader can tell what temporal exposure quantity is being associated with the effect.
A matching window cannot correct a poorly defined effect measure, inaccurate timestamps, inadequate concentration sampling, or a misspecified biological explanation. For example, a PD response may depend on exposure in a tissue that is not represented by measured plasma concentration. The effect may also reflect prior dosing or persistence beyond the stated window. These issues call for a more careful scientific model, not an undocumented shift of the window.
The practical aim is not to find a window that yields the strongest association. Selecting a window after inspecting the observed effect pattern can make the resulting comparison difficult to interpret and can overstate evidence. A defensible analysis states the window and metric from the scientific question or design, reports timing departures, and treats conclusions cautiously when the data do not support the intended pairing.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A PD measurement is recorded at 6 hours after dose. Can average concentration over 6 to 10 hours be described as exposure preceding that measurement? Explain.
- A PD response is averaged from 2 to 5 hours. What timing information should be stated before pairing it with an exposure summary?
- For a window from 3 to 7 hours, the calculated AUC is 36 mg·h/L. Calculate average concentration and give its units.
- Why can a 0-to-12-hour average concentration and a 10-to-12-hour average concentration both be correct while representing different exposure questions?
Solutions. (1) No. The 6-to-10-hour window includes exposure after the 6-hour measurement, so it cannot be described as exposure that preceded that observation. A window ending at or before 6 hours would be temporally prior, though its biological suitability would still need consideration. (2) State that the PD datum is an interval summary over 2 to 5 hours after dose, with those actual boundaries and the time origin. Then state the exposure-window boundaries and metric, and explain any offset or lag if the exposure window differs from 2 to 5 hours. (3) The duration is \(7-3=4\) hours, so average concentration is \(36\ \text{mg·h/L}/4\ \text{h}=9\) mg/L. (4) The first averages exposure across 12 hours, including the earlier period; the second summarizes only the final 2 hours. Their values may differ because they summarize different parts and durations of the concentration-time profile, so they cannot be described as the same exposure measure.