From a Population Input to a Subject-Specific Parameter
A population model does not usually assign every subject the same clearance, volume, or absorption rate. Instead, it starts with a population-level typical value and combines that value with a subject’s random-effect realization to determine the parameter used for that subject. Keeping these quantities distinct is essential: the population value is an input to the parameter model, whereas the individual parameter is the subject-specific value used in the pharmacokinetic model.
The exact transformation from random effect to parameter depends on the model’s chosen parameterization. This tutorial focuses on the common structure of that calculation, not on prescribing a particular transformation. Whenever a numerical example uses a mapping, that mapping is stated as part of the hypothetical model.
The symbols describe different things. The population typical value is shared across subjects at the population level, while \(\eta_i\) varies by subject and \(\theta_i\) is the parameter that results for subject \(i\). A random effect is not itself the individual parameter, and it is not an observed concentration error.
A random effect is treated as random before a subject’s value is realized. Once the model is applied to a particular subject, its realized value is used to calculate that subject’s parameter. A model may specify a probability distribution for random effects and assumptions about dependence across subjects or parameters. Those assumptions belong to the model; the generic mapping alone does not determine them.
The Calculation and Its Assumptions
The calculation is direct once the mapping is known: insert the population value and the subject’s random-effect realization into the specified function. The structural pharmacokinetic model then uses the resulting individual parameter. If a subject has several parameters, each parameter has its own applicable mapping; a shared random effect or correlated random effects can be used only if the model specifies that structure.
A basic consistency condition is that setting the random effect to zero returns the population typical value. This identifies what the model means by “typical.” It does not, by itself, say that the typical value equals the arithmetic mean or median across subjects. Such distributional interpretations require additional assumptions about the random-effect distribution and the mapping.
For an individual parameter calculation to be interpretable, check the following assumptions and inputs:
- The parameterization \(h\) is stated, and the random effect supplied is on the scale expected by that mapping.
- The population value and the resulting individual parameter have compatible units. The random effect’s units, if any, must also be consistent with the mapping.
- The random-effect realization belongs to the correct subject and parameter. Do not substitute an observation-level residual into a subject-level parameter equation.
- The mapping returns a value permitted by the structural model. For example, a model that requires a positive parameter must not be given a nonpositive value.
- If a population prediction is intended, the calculation must use the population-level convention specified for that prediction; an individual prediction instead uses that subject’s random-effect realization.
Worked Calculations
The following mappings are hypothetical and are stated only to make each calculation explicit. They are not universal rules for converting a random effect into a parameter.
Worked Example: Individual Clearance From a Specified Multiplier
Suppose a hypothetical clearance model has a population typical clearance of \(5.00\ \mathrm{L/h}\). For this example, the model specifies \(\theta_i=\theta_{\mathrm{pop}}r(\eta_i)\), with \(r(0)=1.00\) and \(r(1)=1.25\). Subject A has the realized random effect \(\eta_A=1\). Find the subject’s clearance.
The parameter model gives
Subject A’s clearance is \(6.25\ \mathrm{L/h}\). The population input remains \(5.00\ \mathrm{L/h}\); the individual result is different because this subject’s random effect maps to a multiplier of \(1.25\). If \(\eta_A\) were zero, the specified condition \(r(0)=1.00\) would return \(5.00\ \mathrm{L/h}\).
Worked Example: Using an Individual Volume in a Prediction
Consider a hypothetical one-compartment model after an intravenous bolus, with concentration \(C=D/V\). The population typical volume is \(40.0\ \mathrm{L}\). The model specifies \(V_i=V_{\mathrm{pop}}q(\eta_i)\), where \(q(0)=1.00\) and \(q(-1)=0.80\). Subject B has \(\eta_B=-1\) and receives a \(400\ \mathrm{mg}\) dose. Calculate the individual volume and the corresponding concentration under this simplified model.
First calculate the individual volume:
Then use that individual parameter—not the population typical volume—in the subject-specific structural prediction:
For comparison, using the population typical volume gives \(400\ \mathrm{mg}/40.0\ \mathrm{L}=10.0\ \mathrm{mg/L}\). The difference arises from the specified individual volume. This calculation assumes the stated one-compartment bolus relationship and does not include any additional sources of variation.
Worked Example: Applying Separate Effects to Clearance and Volume
Suppose a hypothetical model has \(CL_{\mathrm{pop}}=6.00\ \mathrm{L/h}\) and \(V_{\mathrm{pop}}=30.0\ \mathrm{L}\). It specifies \(CL_i=CL_{\mathrm{pop}}r(\eta_{CL,i})\) and \(V_i=V_{\mathrm{pop}}s(\eta_{V,i})\). For one subject, the supplied mappings give \(r(0.5)=1.10\) and \(s(-1)=0.80\). Find the individual clearance, volume, and elimination rate constant \(k_i=CL_i/V_i\).
Apply each parameter’s mapping to its own population value:
Now use the individual parameters in the stated derived quantity:
The population ratio would be \(CL_{\mathrm{pop}}/V_{\mathrm{pop}}=(6.00\ \mathrm{L/h})/(30.0\ \mathrm{L})=0.200\ \mathrm{h^{-1}}\). It is not the individual rate constant here, because both individual parameters differ from their population inputs. The two mappings must be applied separately before calculating the ratio.
Checks, Alternatives, and Common Pitfalls
A useful diagnostic is to evaluate the mapping at zero random effect. If \(h(\theta_{\mathrm{pop}},0)\ne\theta_{\mathrm{pop}}\) when the model intends the population value to be typical, check whether the equation has been transcribed correctly or whether the parameterization defines “typical” differently. Also check the units and plausible range of every calculated parameter before using it in a structural prediction. A numerical result can be arithmetically correct but incompatible with the model.
The main alternatives differ in what they represent. A population-only calculation uses a common value and does not represent subject-to-subject parameter differences. A model with subject-specific parameters calculates each \(\theta_i\) using that subject’s random effect. Observation-level residual variation, by contrast, concerns departures of observed data from a model’s predictions; it does not replace the subject-level parameter calculation. These approaches answer different modeling questions and should not be interchanged.
| Quantity or approach | Role in the calculation | Common check |
|---|---|---|
| \(\theta_{\mathrm{pop}}\) | Population-level input to the parameter model | Confirm its units and parameter identity |
| \(\eta_i\) | Subject-specific random-effect realization | Confirm it belongs to subject \(i\) and the stated mapping |
| \(\theta_i\) | Individual parameter used by the structural model | Check its units and permitted range |
| Population-only prediction | Uses a population-level parameter convention | Do not describe it as a subject-specific prediction |
| Residual variation | Represents observation-level departures, as specified by the model | Do not substitute it for \(\eta_i\) |
Common mistake: “The random effect is the individual clearance, so I can add it directly to the population clearance.” This reasoning confuses a model input with the parameter produced by the model and assumes a particular transformation without checking it. The correction is to write the specified equation \(\theta_i=h(\theta_{\mathrm{pop}},\eta_i)\), then calculate the result using that mapping. Unless the model explicitly defines direct addition, adding the two quantities is not justified; their units may not even match.
Another frequent error is to use the population typical value in every subject-level prediction. That suppresses the modeled subject-specific parameter differences. Conversely, using one subject’s realized effect for another subject assigns that second subject the wrong parameter. Keep the index \(i\) attached consistently: both the random-effect realization and the resulting parameter refer to the same subject.
Check Your Understanding
For each question, use the stated model information. Solutions are included for checking your reasoning.
- A model defines \(\theta_i=h(\theta_{\mathrm{pop}},\eta_i)\). What does \(\theta_i\) represent, and what does \(\eta_i\) represent?
- A hypothetical model specifies \(\theta_i=\theta_{\mathrm{pop}}r(\eta_i)\), with \(\theta_{\mathrm{pop}}=8.00\ \mathrm{mg/h}\) and \(r(0)=1.00\). What is \(\theta_i\) for a subject with \(\eta_i=0\)?
- For a different subject, the same model gives \(r(\eta_i)=0.75\). Calculate the individual parameter and state its units.
- Why should an observation-level residual not be substituted for \(\eta_i\) in the parameter equation?
- A model calculates individual clearance as \(7.20\ \mathrm{L/h}\) and individual volume as \(36.0\ \mathrm{L}\). What is the individual ratio \(CL_i/V_i\), and why is it preferable to calculate it from the individual values?
Solutions. (1) \(\theta_i\) is the individual parameter for subject \(i\); \(\eta_i\) is that subject’s random-effect realization. (2) \(\theta_i=(8.00\ \mathrm{mg/h})(1.00)=8.00\ \mathrm{mg/h}\). (3) \(\theta_i=(8.00\ \mathrm{mg/h})(0.75)=6.00\ \mathrm{mg/h}\). (4) The residual represents observation-level variation, while \(\eta_i\) is the subject-level input to the parameter mapping; they have different roles. (5) \(CL_i/V_i=(7.20\ \mathrm{L/h})/(36.0\ \mathrm{L})=0.200\ \mathrm{h^{-1}}\). The ratio should use the individual values because it is the derived quantity for that subject.