Why Sampling Design Belongs in the Results
A concentration-time data set is shaped by when samples were collected, how many were collected from each participant, and where those times fall relative to dosing. Those features determine which parts of a concentration profile are observed. A report that gives only the total number of concentrations can conceal important differences: 120 observations might come from 12 participants sampled 10 times each, or from 60 participants sampled twice each. The same total does not imply the same information about individual profiles or population parameters.
Reporting the sampling design makes the evidential basis of a population pharmacokinetic analysis visible. The aim is not to declare a design “good” or “bad” in isolation. It is to describe what was observed, explain which features of the data support particular inferences, and identify where sparse or uneven coverage limits interpretation. These are design-related limitations, distinct from conclusions about model performance.
Describe the Schedule, Not Just the Sample Count
A useful description states the planned or observed sampling times in relation to dose administration and identifies how samples were distributed across participants. Depending on the study, relevant details may include the number of participants with evaluable concentrations, the number of samples per participant, the range of sampling times, and whether times were concentrated around particular parts of the dosing interval. Report units and clarify whether times are measured from a dose or from another reference point.
Sampling can be informative for different features of a profile. Early post-dose samples may help characterize the rising phase and absorption; samples near an expected peak can help describe peak concentrations; later samples can inform the declining or terminal phase. Pre-dose samples can describe concentrations at the end of a dosing interval and, during repeated dosing, may help characterize trough concentrations. These interpretations depend on the drug, route, dose schedule, and model assumptions. A time label such as “peak” should not be used as if the peak concentration had necessarily been observed: a sample near an expected peak is not the same as a measured maximum.
The distribution across participants matters too. If nearly all participants contribute trough samples but only a few contribute early post-dose samples, the data may support population-level description of trough concentrations more strongly than absorption-related conclusions. A report should not imply that all participants had the same schedule if schedules differed. Summarize the actual pattern, including ranges or groups of schedules where necessary.
Worked Example: Describing a Rich Sampling Schedule
In a hypothetical single-dose study, 18 participants each provide concentrations at 0.25, 0.5, 1, 2, 4, 8, 12, 24, and 36 hours after dosing. The following description reports both the number of observations and their timing.
Each participant contributes 9 post-dose samples, giving \(18 \times 9 = 162\) planned concentration observations. Sampling spans 0.25 to 36 hours after dose, with five samples in the first 4 hours (0.25, 0.5, 1, 2, and 4 hours) and four later samples (8, 12, 24, and 36 hours). This schedule provides relatively dense early coverage as well as later observations. It does not, by itself, establish that every participant’s observed maximum or terminal phase is adequately characterized; that depends on the actual concentration profiles and the drug’s kinetics.
A report should not shorten this to “162 samples were collected” if the timing pattern is important to interpreting the analysis. Nor should it call the design rich solely because 162 is a large total: the per-participant schedule and its coverage are the relevant details.
Relate Sampling Times to Dosing
For repeated dosing, a concentration’s meaning depends partly on its position within the dosing interval. For example, a sample taken shortly before a scheduled dose is different from one collected shortly after it. State the dose interval and the sampling-time reference clearly, and distinguish nominal times from actual collection times when both are available and relevant. A summary such as “samples were collected at 2 and 12 hours” is ambiguous if it does not say whether those times are after the first dose, after the most recent dose, or after a specified dose in a multiple-dose schedule.
Sampling near a dose boundary can be especially sensitive to timing. A sample labelled as pre-dose but collected after the dose cannot be interpreted as a pre-dose concentration without qualification. Likewise, if several samples are clustered close together in a short interval, they may add less information about the full profile than samples spread across the interval. The report should describe the actual time coverage rather than infer it from a planned schedule.
Worked Example: Timing Within a Dosing Interval
In a hypothetical study with dosing every 12 hours, a participant’s dose times are 08:00 and 20:00. Concentrations are scheduled at 07:50, 08:30, 19:50, and 20:30. Relative to the 08:00 dose, the first two samples occur 10 minutes before and 30 minutes after dosing. The 19:50 sample is 11 hours 50 minutes after the 08:00 dose and 10 minutes before the 20:00 dose; the 20:30 sample is 30 minutes after the 20:00 dose.
This schedule samples near the end of one dosing interval and shortly after each of two doses. It does not provide a measurement at the exact time of the post-dose maximum, and a sample 30 minutes after dosing should not automatically be described as the peak. A concise report could state: “Samples were scheduled 10 minutes before and 30 minutes after selected doses in a 12-hour regimen; the late-interval samples were collected approximately 11.8 hours after the preceding dose.” The description makes the dose reference explicit and avoids treating scheduled collection times as observed concentrations.
Connect Data Coverage to Estimation Limits
A parameter can be estimated only to the extent that the observed data contain information about it under the chosen model. Sampling design affects that information. For instance, a concentration profile observed only at late times may inform the rate of decline but provide little direct evidence about early absorption. If observations cover only a narrow time range, different combinations of model parameters may describe that range similarly. A numerical estimate can still be produced, but a report should not present the existence of an estimate as proof that the data independently identify the parameter well.
A short derivation illustrates why time spread matters for estimating an elimination rate. Suppose, as a simplified example, that after an intravenous bolus the concentration follows a one-compartment model with first-order elimination, constant clearance and volume, and no additional dose during the interval:
Taking natural logarithms gives a straight-line relationship:
Thus, when both observations lie on this same elimination phase, the slope between two log concentrations estimates \(-CL/V\). If the times are \(t_1\) and \(t_2\), the corresponding positive elimination-rate estimate is:
This calculation relies on the stated model and on both observations being in the relevant elimination phase. If the samples are too close in time, concentration measurement error can have a large effect on the estimated slope. If they span a phase in which the model’s assumptions do not hold, the slope may not represent the elimination rate. The equation therefore explains both the potential value and the limitation of time-separated samples; it does not establish that any two observations are sufficient for a reliable parameter estimate.
Worked Example: What Two Late Samples Can and Cannot Show
Suppose, hypothetically, that concentrations after an intravenous dose are 4.00 mg/L at 8 hours and 1.00 mg/L at 24 hours. Assume for this calculation that both observations are on the terminal elimination phase described by the one-compartment model above.
The time difference is \(24-8=16\) hours. The estimated rate is:
Under the same assumptions, the implied half-life is \(\ln(2)/\hat{k}=0.6931/0.08664=8.00\) hours. The calculation illustrates how separated samples can inform a terminal decline rate. Two late samples do not, however, establish the early concentration-time profile, identify absorption after an oral dose, or demonstrate that the one-compartment assumptions are appropriate. A report should therefore describe the rate as an estimate supported by those late observations under the stated assumptions, not as evidence that all pharmacokinetic features are well characterized.
Sparse Sampling and Population-Level Information
Sparse sampling can be suitable for population analysis when observations from participants with different sampling times collectively cover the concentration-time profile. For example, one group may contribute early samples and another may contribute later samples. This can provide population-level information even when no individual has a full profile. However, the population analysis relies on the model to combine those observations, and the design may not support precise individual parameter estimates from each participant’s limited data.
The distinction is important in reporting. State both the number of participants and the per-participant sampling pattern, then explain what the combined schedule covers. If a parameter is informed mainly by model assumptions, prior information, or a small subset of observations, say so where relevant. Do not imply that a population estimate means each participant’s parameter was directly measured or separately determined.
Worked Example: Same Total, Different Sampling Information
Consider two hypothetical designs with 60 participants and 120 concentration observations each. In Design A, every participant contributes two samples at 1 and 12 hours after dose. In Design B, participants are divided into six groups of 10, with each group sampled at a different time: 0.5, 1, 2, 4, 8, or 12 hours after dose. Each participant still contributes two samples, both at that group’s assigned time on two occasions.
Both designs have \(60 \times 2=120\) observations. Yet Design A concentrates all observations at two times, whereas Design B spreads observations across six scheduled times. Design B may offer broader population-level time coverage if dose histories and the occasions are comparable and the model is suitable. Neither design gives an individual a rich profile, and the comparison alone does not prove that Design B yields better estimates: the distribution of participants, the exact timing, variability, and model assumptions also matter.
A report should make this distinction explicit rather than treating the observation total as a sufficient description. For example: “Each participant contributed two samples; scheduled times were distributed across six post-dose times, with 10 participants assigned to each time group.” That statement supports a more informed assessment of what the pooled data can represent.
Diagnostics, Limitations, and Common Pitfalls
Before interpreting a sampling summary, check whether it accurately reflects the data used in the analysis. Useful checks include the range of observed times, the number of samples per participant, whether a time point is represented by only a few participants, and whether the reported dose reference is unambiguous. A schedule that appears balanced in aggregate can still leave important gaps for particular participants or phases. If the data have little coverage in a region, model-based predictions there should be recognized as more dependent on assumptions than observations in well-sampled regions.
The main reporting alternatives are not interchangeable. A schedule table can show exact nominal times and participant counts at each time; a prose summary can efficiently describe overall patterns; a plot of sampling times relative to dose can reveal clustering or gaps. Often a concise narrative plus a table or figure is clearer than any one format alone. Regardless of format, include enough detail to distinguish per-participant coverage from pooled coverage and planned times from observed timing.
- Pitfall: reporting only the total number of concentrations. This hides whether samples were spread across many participants or concentrated within a few. Report both participant counts and the per-participant pattern.
- Pitfall: calling a nominal time a measured peak or trough. A scheduled sample near an expected peak is not necessarily the maximum. Describe its timing relative to dose without overstating what it measures.
- Pitfall: treating sparse observations as direct individual characterization. A population model can combine sparse observations, but that does not mean each participant’s individual profile or parameters are independently well determined.
- Pitfall: interpreting an estimate without its design context. A value may be calculable while remaining weakly informed by the observed time range. State which parts of the profile were observed and which conclusions rely more heavily on model assumptions.
Sampling-related limitations should be specific and balanced. “Sparse sampling may limit individual characterization of early absorption because few participants had early post-dose observations” is more informative than “the data were limited.” Conversely, avoid declaring a parameter unestimable solely because sampling was sparse; the relevant question is what the data, combined with the model and other information, can support.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A study reports 90 concentrations from 45 participants, with two samples per participant. What additional information is needed to judge whether the pooled schedule covers the concentration-time profile?
- In a repeated-dose study, why should a report state whether a sample was collected before or after a dose?
- Under the stated one-compartment elimination assumptions, concentrations are 8 mg/L at 6 hours and 2 mg/L at 18 hours. Calculate \(\hat{k}\) and state one limitation of this calculation.
- Why does a population estimate from sparse data not necessarily imply precise individual parameter estimates?
Solutions. (1) Report the actual or scheduled times relative to dose, how participants are distributed across those times, and the range of time coverage; the total alone does not show whether samples cover early, peak, and later phases. (2) Because position within the dosing interval changes the interpretation: a pre-dose observation may represent a late-interval concentration, whereas a post-dose observation follows the new dose. (3) The interval is \(18-6=12\) hours, so \(\hat{k}=[\ln(8)-\ln(2)]/12=\ln(4)/12=1.3863/12=0.1155\ \text{h}^{-1}\). The calculation assumes both observations lie on the same terminal elimination phase and does not characterize early absorption or establish model adequacy. (4) The population estimate combines information across participants through a model; each participant may contribute too few observations to characterize their own concentration-time profile or parameter values precisely.