Why a PBPK Model Needs a Reference Profile
A PBPK model is parameterized for an organism, not for an abstract “average person” or “average animal.” Before assigning physiological inputs, it is necessary to say what kind of individual those inputs describe. Age, sex, body mass, species, and other recorded characteristics can define the context in which a model is intended to apply. If that context is left vague, two modelers can use the phrase “reference adult” while silently referring to different individuals.
A reference physiological profile makes that context explicit. It is a documented set of covariate values chosen to represent a specific modeling baseline. It is not automatically the arithmetic average of a study sample, nor does it assert that every person or animal has those characteristics. The distinction matters when interpreting model predictions, comparing parameterizations, or deciding whether a baseline is relevant to a particular population.
For clarity, write a profile as a vector of covariates, \(\mathbf{c}=(c_1,c_2,\ldots,c_k)\). Components might include age in years, body mass in kilograms, sex as a recorded category, and species or strain where relevant. This notation identifies the contents of a profile; it does not imply that every component is numeric or that the components can all be averaged. The reference context should also say what population or setting the profile is meant to represent, such as a healthy adult human baseline or a specified laboratory-animal group.
What Makes a Profile Usable?
A useful profile is specific enough that another modeler can tell what each entry means. For numeric covariates, state the unit and the basis of the value. For categorical covariates, state the category labels and what they mean in the data source or model. If age is recorded in years, specify whether the value is exact age or an age category. If sex is recorded, name the recorded category rather than replacing it with a number that might be mistaken for a quantitative measurement.
The profile should also identify its intended context and time reference. An individual's age changes with time, and some covariates can depend on when or how they were recorded. A profile assembled from records collected at different times may not describe one coherent individual. Likewise, “adult,” “healthy,” or “control” is not a complete definition by itself: those labels need operational meaning for the application. The goal here is not to settle how each physiological input is measured, but to make the model's reference subject unambiguous.
A profile can include both covariates that define the baseline and notes about the population from which the baseline was selected. These should not be conflated. For example, “70 kg” may be the selected mass for a reference profile, while “the median mass in the target population was 70 kg” is a statement about a population summary. The numerical values happen to match, but their meanings and justifications differ.
Worked Example: Writing Down a Reference Profile
Suppose a modeling team needs a hypothetical baseline for an adult human. It chooses age 35 years, body mass 72 kg, and recorded sex category “female.” The values are invented for this example and are not presented as reference standards.
The profile can be written as \(\mathbf{c}=(35\text{ years}, 72\text{ kg}, \text{female})\), with the component definitions stated alongside it. The first two entries are numeric covariates with units; the third is categorical. The team should also document that this is a selected baseline adult human profile, rather than implying that it describes all adults or the average of a particular sample.
This specification is more reproducible than “typical adult.” Another modeler can identify the chosen age, mass, and category and decide whether that context matches a modeling question. It does not, by itself, establish that the values are suitable for every application; suitability depends on the intended target population and the model use.
Representative Baseline Versus Population Average
A representative baseline is a choice of a complete profile. A population average is a summary of observations. A mean can be computed for a numeric covariate, but the resulting value may not correspond to an actual individual. A categorical covariate such as sex cannot be averaged in the same sense as age or mass. It can instead be described by category counts or proportions.
Even when every component is numeric, averaging each component separately may yield a combination that was not observed. If covariates are related within the population, a coordinate-by-coordinate average can also hide those relationships. A vector of mean values therefore should not be assumed to be a realistic, jointly representative individual without checking the context. One alternative is to select a documented individual or profile that is relevant to the intended use; another is to describe the target population using suitable summaries or distributions. Those are different modeling choices, not interchangeable labels for the same object.
Worked Example: Why Component Means May Not Define an Individual
Consider five hypothetical adults with recorded age, body mass, and sex category:
| Person | Age (years) | Body mass (kg) | Sex category |
|---|---|---|---|
| A | 25 | 60 | Female |
| B | 40 | 90 | Male |
| C | 35 | 70 | Female |
| D | 50 | 80 | Male |
| E | 30 | 50 | Female |
The mean age is
The mean body mass is
Three of the five records are female and two are male, so the observed proportions are \(3/5=0.600\) female and \(2/5=0.400\) male. The numeric summaries are 36 years and 70 kg, while sex is summarized by category proportions. Nothing in those calculations supplies one sex category for a single profile. Calling “36 years, 70 kg” the average profile would therefore leave a categorical component unresolved, and it would still not establish that a specific individual has the combined numeric values.
Population Composition Changes an Average
A population average depends on which individuals are included and how they are represented. A sample with equal numbers from two age groups may not reflect a target population in which one group is more common. If group means and target-population proportions are known, a population mean can be calculated as a weighted mean. For two groups with means \(\mu_1,\mu_2\) and population proportions \(w_1,w_2\), where \(w_1+w_2=1\), the overall mean is \(w_1\mu_1+w_2\mu_2\). This is a population summary, not a recipe for selecting an individual baseline.
Worked Example: Sample Average Versus Target-Population Average
Assume a hypothetical target population has two age groups. The younger group makes up 70% of the population and has a mean age of 30 years; the older group makes up 30% and has a mean age of 50 years. A convenience sample instead contains 10 people from each group. All numbers are illustrative.
The sample's unweighted mean, using the two group means and equal sample counts, is
The target-population mean uses the target proportions:
The values differ because the sample gives equal weight to the groups, while the target population does not. The 36-year target mean is a summary of that defined population. It does not mean that a 36-year-old is the uniquely correct reference individual, nor that the target population consists of individuals aged 36.
Practical Checks and Common Failure Modes
Before using a profile as a model baseline, check that the covariates are defined, that their units are stated, and that the values describe a coherent reference context. Confirm that the profile does not quietly combine values from incompatible populations or time points. If a value is unavailable, mark it as missing rather than silently substituting a population mean. Any imputation or selection rule should be explicit.
A useful diagnostic is to ask whether another analyst could reconstruct the intended profile from its description. If “adult,” “average,” or “standard” is the only information provided, the answer is usually no. A second check is to distinguish individual-level values from group summaries: “the recorded age of this reference individual” is different from “the mean age in the target population.” A third is to confirm that categorical values remain categories and are not treated as measurements merely because software encodes them numerically.
The main alternatives serve different purposes. A single specified baseline is straightforward to interpret and compare across model runs. A mean profile provides a compact summary of numeric covariates but can combine values in a way that does not describe any actual person. A distribution or set of stratified summaries retains information about population heterogeneity but requires more than one profile to represent. The choice depends on whether the question concerns a defined baseline or variation across a group.
Common mistake: “The target population's mean age is 36 years and its mean mass is 70 kg, so the reference individual must be a 36-year-old person weighing 70 kg.” This reasoning turns separate population summaries into an individual without justification and may ignore categorical covariates and relationships among values. The correction is to label those values as population means, then separately specify and justify a complete reference profile if a single baseline is needed.
Name the organism and the intended reference setting or target population.
Give each value, its unit or category meaning, and any relevant time reference.
Say whether the values specify one baseline profile or summarize a population.
Look for missing definitions, mixed sources or times, and averages that do not describe a joint individual.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A hypothetical profile lists age as 42, body mass as 68, and sex as 1. What additional information is needed before the profile is unambiguous?
- A sample contains 20 people aged 20–39 and 20 aged 40–59. The target population is 75% in the younger group and 25% in the older group. If group mean ages are 30 and 50 years, calculate the sample mean based on these group means and the target-population mean.
- Why is it generally inappropriate to report the average of a categorical sex variable as though it were an age-like measurement?
- A model description says only “standard adult.” Name two reasons this is insufficient to reproduce the reference profile.
- Can a reference profile and a population average have the same numeric age? Explain whether that makes them the same kind of quantity.
Solutions. 1. State units for age and body mass, define what the value 1 encodes (or use the category label directly), and identify the reference context. 2. The sample mean is \([20(30)+20(50)]/40=1{,}600/40=40\) years. The target mean is \(0.75(30)+0.25(50)=22.5+12.5=35\) years. The difference follows from the different group weights. 3. The category is a label, not a quantitative measurement with meaningful differences; report category counts or proportions instead. 4. The description does not specify exact covariate values and does not define what “standard” means or which population is intended. 5. Yes. The numeric value can coincide, but a profile's age is one component of a specified individual baseline, whereas a population mean summarizes values across a group.