Why the Exposure Metric Must Match the Question
A drug concentration profile is a series of values over time, not a single exposure number. A biomarker measured at one time, or summarized over a period, may relate to different features of that profile. A peak concentration, total exposure over a window, and concentration near the next dose answer different questions. Choosing among them is therefore part of defining the exposure-response question, not merely a data-processing choice.
The comparison should be made using each participant’s exposure and biomarker data. A dose-group average can conceal meaningful individual variation, while a pattern across dose-group averages can arise even when there is little exposure-response relationship among participants within a group. The individual-level comparison still needs careful interpretation: exposure is not randomly assigned independently of all other participant characteristics, and association alone does not establish that exposure caused the biomarker change.
What Common Exposure Metrics Represent
The three common choices below summarize different properties of an individual profile. Their units and time windows are part of their meaning.
| Metric | What it summarizes | Typical interpretation and caution |
|---|---|---|
| Cmax | The maximum concentration within a specified profile or interval | Peak exposure; depends on the sampling schedule or the PK method used to estimate the peak. |
| AUC over a stated interval | The area under the concentration-time curve during that interval | Cumulative exposure; its units are concentration multiplied by time, and its value depends on the interval. |
| Trough concentration | Concentration at, or immediately before, a specified dosing time | Concentration near the end of an interval; a predose value is not automatically a steady-state trough. |
For a short-lived biomarker measured near a plausible concentration peak, Cmax may be a relevant candidate. For a biomarker measured after exposure has accumulated over several hours, an AUC over a biologically relevant period may be more appropriate. If the question concerns exposure remaining before a subsequent dose, trough concentration may be relevant. These are starting hypotheses to evaluate, not universal rules: biomarker biology and the timing of measurement determine which feature is plausible.
The biomarker outcome also needs a clear definition. For example, it might be a change from baseline at a specified postdose time, or an area under the biomarker time course. The earlier material on baseline and time-course controls is relevant here: the observed change should be distinguished from baseline variation and time trends that could occur without drug exposure.
Timing, Windows, and Individual Exposure
A metric should be calculated from the same participant’s PK observations and over a window that precedes, or is otherwise biologically relevant to, the biomarker outcome. A concentration measured at the biomarker time is not necessarily the relevant exposure. If a response develops after a delay, the concentrations earlier in the profile may matter more than the simultaneous concentration. Conversely, an AUC that includes substantial exposure after the biomarker measurement cannot be interpreted as a cause of that earlier outcome.
For observed concentrations at times \(t_0,\ldots,t_m\), the linear trapezoidal rule estimates the area over consecutive observations by treating the concentration between two times as changing linearly. The total is the sum of those trapezoids:
Here \(C(t_j)\) is the concentration at time \(t_j\). The result has units such as mg·h/L when concentration is measured in mg/L and time in hours. This calculation estimates exposure over the observed window; it does not establish that the window is the biologically correct one. Sparse sampling, an unobserved peak, or a long interval between samples can make the estimate less reliable.
Worked Example: Relating a Peak Metric to an Early Biomarker Change
Suppose four hypothetical participants receive a single dose. Their estimated Cmax values and biomarker changes from baseline at 1 hour are shown below. The biomarker change is in percentage points; Cmax is in mg/L.
| Participant | Cmax (mg/L) | Biomarker change at 1 h (percentage points) |
|---|---|---|
| A | 4 | 8 |
| B | 6 | 13 |
| C | 8 | 17 |
| D | 10 | 20 |
A peak metric is a plausible candidate if the question is whether higher peak exposure accompanies a larger early biomarker change. The Cmax values are participant-level estimates; their reliability depends on whether the PK sampling or model adequately captures each peak. The biomarker is measured at a stated time, so this comparison is specifically about Cmax and the 1-hour change—not about a response throughout the whole day.
For a descriptive linear summary, let \(x\) be Cmax in mg/L and \(y\) be biomarker change in percentage points. The slope is \(\sum (x_i-\bar{x})(y_i-\bar{y})/\sum (x_i-\bar{x})^2\). Here \(\bar{x}=7\) mg/L and \(\bar{y}=14.5\) percentage points. Thus:
In these four hypothetical observations, the descriptive slope is 2.00 percentage points of biomarker change per 1 mg/L higher Cmax. The pattern is increasing across the four participants. This small example does not establish a stable population relationship, demonstrate statistical significance, or show that higher Cmax caused the changes. It also cannot rule out that the pattern reflects differences between dose cohorts or participant characteristics.
Worked Example: Calculating AUC for a Cumulative-Exposure Question
A hypothetical participant has measured concentrations of 0, 8, 4, and 1 mg/L at 0, 2, 6, and 12 hours after dosing. Suppose the biomarker outcome is summarized at 12 hours, and the planned candidate metric is AUC from 0 to 12 hours.
Apply the trapezoidal rule separately to each interval. The 0–2 hour area is \((0+8)/2 \times 2=8\) mg·h/L. The 2–6 hour area is \((8+4)/2 \times 4=24\) mg·h/L. The 6–12 hour area is \((4+1)/2 \times 6=15\) mg·h/L. Therefore:
The average concentration over these 12 hours is \(47.0\ \text{mg·h/L}/12\ \text{h}=3.92\ \text{mg/L}\). AUC and average concentration are related but not interchangeable without stating the interval: AUC retains the time dimension, whereas average concentration divides by the interval length. This AUC is a reasonable candidate if cumulative exposure over 0–12 hours is relevant to the biomarker summary. It would be a poor choice for interpreting a biomarker measured at 2 hours if much of the AUC comes from exposure after that measurement.
The arithmetic assumes linear change between each pair of measured concentrations. With sparse sampling, the true curve may rise or fall substantially between observations, so the trapezoidal estimate can miss the true area or peak. The calculation is exact for the specified trapezoidal approximation, not necessarily for the underlying concentration-time curve.
When the Biomarker Response May Be Delayed
A biomarker measured hours after dosing may reflect earlier exposure rather than the concentration at the measurement time. If a delayed response is plausible, a preceding exposure window can be more informative than a contemporaneous concentration. But “delayed” does not identify one universally correct window. The proposed delay and window should be supported by biological knowledge and the sampling schedule, and competing windows should be treated as exploratory unless specified in advance.
The next example illustrates how a peak and an AUC can rank participants differently. That difference is informative about what question each metric answers; it is not, by itself, evidence that either metric is the correct biological driver.
Worked Example: Choosing a Window for a Later Biomarker Measurement
Two hypothetical participants have concentrations at 0, 2, 4, and 8 hours. A biomarker is measured at 8 hours, and a delayed response to exposure earlier in the interval is considered plausible.
| Participant | Concentrations at 0, 2, 4, 8 h (mg/L) | Biomarker change at 8 h (percentage points) |
|---|---|---|
| A | 0, 10, 3, 1 | 9 |
| B | 0, 6, 5, 2 | 13 |
Participant A has the higher observed peak: Cmax is 10 mg/L versus 6 mg/L for participant B. For a candidate cumulative metric, calculate AUC from 0 to 8 hours. For A, the interval areas are \(10\), \(13\), and \(8\) mg·h/L, giving AUC \(=31.0\) mg·h/L. For B, the interval areas are \(6\), \(11\), and \(14\) mg·h/L, giving AUC \(=31.0\) mg·h/L. These calculations use widths of 2, 2, and 4 hours, respectively, in the trapezoidal rule.
The two participants have equal estimated AUC despite different peaks, while B has the larger observed biomarker change. In this pair, the outcome does not increase with Cmax, and equal AUC does not distinguish the two changes. The data are far too limited to determine which metric is useful. They do show why the metric must reflect the question: peak magnitude and cumulative exposure are not equivalent. Because the biomarker is measured at 8 hours, an AUC from 0 to 8 hours is temporally eligible as a candidate; exposure after 8 hours would not be appropriate for explaining this measurement.
Assumptions, Diagnostics, and Alternative Choices
An exposure-response comparison is most interpretable when several conditions are considered. The PK and biomarker records must be matched to the correct participant, dose, and time. The biomarker outcome should be defined consistently, including how baseline and time-course changes are handled. The exposure metric and window should be biologically plausible and temporally aligned. There should also be enough individual exposure variation to examine a relationship, rather than relying only on a few dose-group means.
Check the underlying data before interpreting a trend. Review individual concentration-time profiles for implausible values, missing samples, and peaks that may have been missed. Confirm the units, timestamps, dose times, and AUC interval. Compare individual exposure values with individual biomarker outcomes in a plot or table, while marking dose cohort where relevant. Look for whether an apparent trend is driven by one participant, whether biomarker changes vary at similar exposures, and whether exposure values barely overlap across cohorts. These checks can reveal problems, but they do not prove that a chosen metric is biologically correct.
Dose and exposure are related, so an apparent relationship across participants may be driven by dose cohort. If all low exposures occur in one cohort and all high exposures in another, dose, cohort, and exposure effects are difficult to disentangle. Likewise, a biomarker change might be associated with participant characteristics that also affect PK. A simple plot or slope should therefore be described as an association, not as a causal drug effect. Sparse Phase I data may not support a complex adjustment; acknowledging the limitation is more defensible than implying it has been resolved.
| Candidate | Useful when the question concerns | Key limitation |
|---|---|---|
| Cmax | Peak exposure and a response plausibly related to peak concentration | May be missed by sparse sampling; does not summarize duration of exposure. |
| AUC | Cumulative exposure over a prespecified, relevant interval | Depends on the interval and interpolation; can conceal distinct peak and duration profiles. |
| Trough concentration | Exposure remaining near a specified dose time | May not represent earlier exposure; sampling time and dosing context must be stated. |
A common mistake is to calculate every available metric, select whichever has the strongest apparent association, and then present it as the biologically established driver. That selection can overstate evidence, especially with few participants. The correction is to justify a primary candidate using the biomarker question, timing, and biological rationale; specify its window; and label additional metric comparisons as exploratory. Metrics can be compared to understand sensitivity to the choice of summary, but a better numerical fit in a small data set does not establish a mechanism.
These summaries are useful for organizing the exposure-response question, but they do not themselves describe how a biomarker responds as concentration changes or how a delay arises. The choice of metric should be made before interpreting a pattern, and any remaining uncertainty about response shape or delay should be stated rather than hidden in the metric definition.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A biomarker is measured at 4 hours and is hypothesized to respond to peak exposure shortly after dosing. Which candidate metric is most directly aligned with that question, and what sampling limitation should be checked?
- Concentrations are 2 and 6 mg/L at 1 and 3 hours. Use the trapezoidal rule to calculate AUC from 1 to 3 hours.
- A biomarker is measured at 6 hours. Why would AUC from 0 to 12 hours be problematic as an explanatory exposure metric for that measurement?
- All participants in a low-dose cohort have low exposure and small biomarker changes, while all participants in a high-dose cohort have high exposure and larger changes. What is one important limitation of interpreting this as an individual exposure-response relationship?
- Two participants have the same AUC but different Cmax values. Does this show that AUC is the biologically correct metric? Explain.
Solutions. (1) Cmax is a directly aligned candidate for a question about peak exposure, but sparse PK sampling may miss the true peak; check how the peak was observed or estimated. (2) AUC \(=(2+6)/2 \times (3-1)=8.00\) mg·h/L. (3) The 0–12 hour window includes exposure from 6 to 12 hours, which occurs after the biomarker measurement and cannot explain that earlier outcome. (4) Exposure is confounded with dose cohort, so a difference between cohorts may produce the apparent pattern; there may be too little within-cohort exposure variation to assess an individual-level relationship. (5) No. Equal AUC with different peaks shows that AUC and Cmax summarize different profile features. It does not establish which feature is biologically relevant; that choice requires a justified question, appropriate timing, and adequate evidence.