Why Compare Concentration Profiles Across Visits?
A concentration-time profile can differ from one study visit to another. The difference may be visible as a change in concentration at similar postdose times, a change in the shape of the profile, or a change in how much data are available. Comparing visits can help describe these patterns and identify questions for further investigation. But a plot or summary alone does not explain why a difference occurred.
A valid comparison depends on putting observations on a meaningful common time scale and showing which records support each visit’s profile. The same nominal visit label does not guarantee comparable dose, regimen, sampling schedule, or participant composition. Those features should be checked and stated rather than silently treated as equivalent.
Align the Time Scale Before Comparing
For a profile comparison, the horizontal axis should represent a comparable time quantity at each visit. In a dose-related display, that is commonly elapsed time since the dose associated with the concentration measurement. If a sample was collected at 10:15 and the relevant dose was administered at 08:00, the elapsed time is 2 hours 15 minutes, or 2.25 hours. Clock time alone would not show that this sample was collected 2.25 hours after dosing if the dose time differed at another visit.
Use a common axis range and units when the sampling windows permit it. Aligned panels—one panel per visit—make it easier to inspect each visit without overplotting observations. An overlay can help compare shapes, but it becomes difficult to read when many individual profiles overlap. In either display, make the visit labels, time scale, concentration units, and any dose or regimen differences clear.
Before interpreting a visual difference, check whether the visits are comparable in the features that define the profile: dose timing, dose amount, dosing interval, sampling schedule, and relevant study conditions. If the question is about a change between visits, also note whether the same participants contribute observations at both visits. These checks do not explain a difference; they clarify what is being compared.
Worked Example: Aligning Samples From Two Visits
Hypothetical data. At visit 1, a dose is given at 08:00 and a sample is collected at 10:15. At visit 2, the corresponding dose is given at 07:45 and a sample is collected at 10:00. Each concentration is 12 mg/L. Determine where the two observations belong on a postdose-time plot.
At visit 1, the elapsed time is \(10{:}15-08{:}00=2\) hours 15 minutes, which is \(2.25\) hours. At visit 2, it is \(10{:}00-07{:}45=2\) hours 15 minutes, also \(2.25\) hours.
Both observations therefore appear at 2.25 hours on the postdose-time axis, with concentration 12 mg/L. Plotting them at 10:15 and 10:00 on a clock-time axis would instead imply different positions without representing their equal elapsed times after dose. This alignment supports a descriptive comparison at a common time; it does not establish that all visit conditions were otherwise equivalent.
Visit-Specific Summaries and Paired Comparisons
A plot can show the shape and spread of observed concentrations, while a visit-specific summary gives a compact description at selected times. At each visit and time point, report the number of available observations, \(n\), alongside an appropriate center and measure of spread. Medians and quantiles are useful when concentrations are skewed or a few high values dominate. Means and standard deviations can also be informative when the arithmetic average is the desired descriptive target. The choice should match the question, and the summary should not hide the distribution it represents.
If the same participants have measurements at both visits at comparable times, show or calculate the within-participant differences as well as visit-specific summaries. A difference between two visit means is not generally the same as a summary of individual changes when the participants differ or observations are missing. Even with the same participants, the mean difference is a description of these observed paired changes, not an explanation for them.
Worked Example: Visit Summaries and Paired Changes
Hypothetical data. At 2 hours after dose, five participants have concentrations in mg/L as follows. Assume the dose and sampling conditions are comparable at both visits.
| Participant | Visit 1 | Visit 2 | Visit 2 minus Visit 1 |
|---|---|---|---|
| A | 8 | 10 | 2 |
| B | 12 | 14 | 2 |
| C | 6 | 7 | 1 |
| D | 20 | 22 | 2 |
| E | 10 | 13 | 3 |
The visit 1 mean is \((8+12+6+20+10)/5=56/5=11.2\) mg/L. The visit 2 mean is \((10+14+7+22+13)/5=66/5=13.2\) mg/L. The difference between the means is \(13.2-11.2=2.0\) mg/L.
The paired changes are 2, 2, 1, 2, and 3 mg/L. Their mean is \((2+2+1+2+3)/5=10/5=2.0\) mg/L, and their median is 2 mg/L. In this complete paired example, the mean paired change equals the difference in visit means. The visit 1 median is 10 mg/L and the visit 2 median is 13 mg/L; these medians describe the two marginal visit distributions and are not a calculation of the median paired change.
The observed summaries indicate higher concentrations at visit 2 for these five participants at this time point, with a mean paired increase of 2.0 mg/L. They do not establish why the concentrations differ, nor do they alone describe differences at other postdose times.
Why Changing Data Coverage Can Change a Visit Summary
Visit summaries are calculated from the observations available at that visit. If the participants contributing to one visit differ from those contributing to another, a difference in the summaries may reflect that composition as well as any changes within participants. Reporting the count at each visit and time, and identifying whether participants are paired, helps make this limitation visible.
The algebra makes the issue precise. Let \(C_{iv}(t)\) denote participant \(i\)’s concentration at time \(t\) and visit \(v\), and let \(S_v\) be the set of participants observed at that visit. The visit mean is
If the same \(n\) participants are observed at both visits, subtracting the means gives
This equality follows by distributing the common factor \(1/n\) across the two sums and subtracting term by term. It says that with complete, matched observations, the difference in arithmetic means equals the mean of the participant-level changes. If the sets \(S_1\) and \(S_2\) differ, the two sums do not contain the same participants, so that paired-change interpretation does not follow from subtracting the visit means.
Worked Example: A Visit Mean Shift From Participant Composition
Hypothetical data. At 2 hours after dose, four participants have unchanged concentrations between visits: A has 4 mg/L, B has 6 mg/L, C has 10 mg/L, and D has 12 mg/L. A is not observed at visit 2; B, C, and D are observed at both visits. Assume the recorded concentrations for the observed participants are unchanged.
At visit 1, the mean across all four participants is \((4+6+10+12)/4=32/4=8.00\) mg/L. At visit 2, only B, C, and D contribute, so the mean is \((6+10+12)/3=28/3=9.33\) mg/L. The visit means differ by \(9.33-8.00=1.33\) mg/L.
For the three participants observed at both visits, the paired changes are \(0,0,0\) mg/L, and their mean change is 0 mg/L. Thus, in these hypothetical data, the higher visit 2 mean results from the changed set of participants contributing to the summaries, not from a concentration increase among the matched participants. Without the participant-level and availability information, the two visit means alone would not reveal this distinction.
Plotting Choices, Diagnostics, and Limitations
A useful display makes both the profile and its support inspectable. For dense individual data, aligned visit panels can show subject-level observations and lines, with visit-specific summary curves added only when they do not obscure the underlying data. If observations are sparse, a summary at every time point may be unstable or misleading; show the counts and avoid implying a smooth, well-supported curve between observed times. Do not treat an interpolated line as an observed concentration.
Common options answer different descriptive questions:
| Approach | Useful for | Important limitation |
|---|---|---|
| Aligned visit panels | Comparing shapes while keeping observations grouped by visit | Differences can be hard to see unless axes and scales are comparable |
| Overlaid visit summaries | Quickly comparing central profiles on a shared axis | Can hide individual spread, missingness, and unequal support |
| Paired subject trajectories or changes | Showing within-participant patterns among those observed at both visits | May represent only participants with complete paired observations |
| Visit-specific distribution summaries | Comparing spread and central tendency at selected times | Do not show the full time-course shape by themselves |
Useful diagnostics include checking that the plotted times were derived consistently, that units and axes agree, and that sample counts are visible at each visit and time. Look for changes in sampling density, gaps in the time window, outlying observations, and visits with few contributors. Compare participant-level trajectories where available: a visit-level curve can move even when matched participants show little change, as the preceding example demonstrates.
A plot cannot resolve all reasons for a difference. Differences may coincide with changes in dose conditions, study procedures, participant availability, or other features not represented on the axes. These are possibilities to investigate, not conclusions supported by the plot alone. Descriptive displays also do not substitute for a model when the objective is to estimate effects while accounting for multiple factors or repeated observations.
A Practical Comparison Sequence
A disciplined comparison separates the display from the interpretation. First align observations using the intended time scale. Next inspect visit-specific data coverage and check whether the relevant study conditions are comparable. Then choose a display and summary that match the data density and question. Finally state what the observed differences describe and what they cannot establish.
Specify the visits, concentration units, time scale, and whether the intended description is marginal by visit or paired within participant.
Use comparable elapsed-time values, common axes where appropriate, and visible observations or counts to judge coverage.
Choose centers and measures of spread suited to the observed distributions; report the number of records behind each summary.
Distinguish a difference between visit summaries from a pattern of paired changes, and do not assign a cause from the display alone.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- At visit 1, a sample is collected 90 minutes after dose; at visit 2, a sample is collected 2 hours after dose. What must be done before comparing their concentrations on a postdose-time profile?
- Five participants have paired visit changes of 1, 0, 2, 1, and 1 mg/L. Calculate the mean paired change.
- Why should the number of observations behind a visit-specific summary be reported?
- A visit 2 mean is higher than a visit 1 mean, but different participants contribute at the two visits. What can be concluded, and what cannot be concluded from those means alone?
- Name one advantage and one limitation of aligned visit panels compared with an overlaid summary curve.
Solutions. (1) Plot each observation at its elapsed time since dose; the first is at 1.5 hours and the second at 2 hours, so they are not measurements at the same postdose time. (2) The mean is \((1+0+2+1+1)/5=5/5=1.00\) mg/L. (3) The count shows how much data supports the summary and can reveal sparse or unequal coverage across visits or times. (4) The visit 2 summary is higher for the observations contributing to it, but the difference cannot be attributed to within-participant increases because the contributing sets differ. (5) Aligned panels separate observations by visit and make individual patterns easier to inspect; they may make direct comparison less immediate unless axes and scales are consistent. An overlaid curve facilitates direct comparison but can conceal spread, data coverage, and individual changes.