Why Parameter Names Are Not Enough
Moving a parameter from an estimation model into a simulation model is not simply a matter of copying a number into a similarly named field. Two tools may use different names for the same quantity, the same name for quantities defined differently, or different units and transformations. A copied value can therefore be numerically plausible but represent the wrong quantity.
A parameter crosswalk makes those relationships explicit. For each quantity, it records what the parameter means, how it is named in each model, its units, any transformation needed, and the reference conditions under which the value applies. This tutorial develops that approach using clearance and volume in a one-compartment model. It builds on the earlier material on translating clearance and rate parameterizations and on interpreting estimation and simulation outputs, without assuming that output labels alone define a parameter.
The basic checks are semantic: does each value represent the same quantity, on the same scale, in compatible units, and at the same reference conditions? The crosswalk is not complete until those questions have answers. For a simple one-compartment model, a useful starting point is
Here \(CL_i\) is individual clearance, \(V_i\) is individual volume, \(WT_i\) is an individual’s weight, \(WT_{\mathrm{ref}}\) is the weight at which the typical values are defined, and the \(\eta\) terms are individual random effects on the log scale. The equations are included to specify the quantities being mapped; they assume positive clearance and volume, the stated covariate relationships, and the same definitions of weight and reference weight in both implementations. A value of \(TVCL\), for example, is not necessarily an individual’s clearance: it is the typical clearance at the reference weight and at \(\eta_{CL}=0\).
Build the Crosswalk Around Meaning and Conditions
A practical crosswalk separates the quantity from the labels used by the two models. It also states what computation, if any, is needed to get from the estimation output to the value expected by the simulation model. The entries below use hypothetical names; the names themselves are not universal.
| Quantity | Estimation-model name | Simulation-model name | Units and scale | Reference or mapping condition |
|---|---|---|---|---|
| Typical clearance | TVCL | CL | Volume/time; natural scale | Typical value at \(WT_{\mathrm{ref}}\), with \(\eta_{CL}=0\) |
| Typical volume | TVV | V | Volume; natural scale | Typical value at \(WT_{\mathrm{ref}}\), with \(\eta_V=0\) |
| Weight effect on clearance | THETA_WT_CL | WTCL | Dimensionless exponent | Must act on the same weight ratio |
| Weight effect on volume | THETA_WT_V | WTV | Dimensionless exponent | Must act on the same weight ratio |
| Elimination rate constant | Derived as \(CL/V\) | K | Reciprocal time | Derived from clearance and volume using consistent units |
The table distinguishes estimated quantities from derived ones. If both models calculate \(k\) from clearance and volume, mapping \(CL\) and \(V\) consistently should also map \(k\). If one model instead stores \(k\) directly, the crosswalk must identify it as a rate constant, not treat it as another name for clearance.
Worked Example: Converting Clearance and Volume Units
Suppose an estimation model reports hypothetical typical values \(TVCL=5.200\ \mathrm{L/h}\) and \(TVV=42.00\ \mathrm{L}\). A simulation model expects clearance in \(\mathrm{mL/h}\) and volume in \(\mathrm{mL}\). Assume both values refer to the same one-compartment model, with no covariate or random-effect adjustment for this calculation.
Clearance conversion. Since \(1\ \mathrm{L}=1000\ \mathrm{mL}\),
Volume conversion. Using the same volume-unit conversion,
As a check, the elimination rate constant in the original units is
The converted values give the same result: \(5200\ \mathrm{mL/h}/42000\ \mathrm{mL}=0.1238\ \mathrm{h^{-1}}\). This check supports the unit mapping, but it does not by itself establish that the models use the same parameter definitions or covariate assumptions.
Why Consistent Unit Conversion Preserves the Rate
The rate constant \(k=CL/V\) has units of reciprocal time. If clearance and volume are both expressed using the same volume-unit conversion factor \(c\), then the numerical value of their ratio is unchanged:
For example, converting both liters to milliliters uses \(c=1000\), so that factor cancels. The cancellation depends on converting both quantities consistently. It does not mean that the units of \(k\) disappear: they remain reciprocal time. If the time unit changes as well, the numerical rate changes accordingly. For instance, \(0.1238\ \mathrm{h^{-1}}\) is approximately \(0.002063\ \mathrm{min^{-1}}\), because an hour contains 60 minutes and \(0.1238/60=0.002063\). Thus, a crosswalk must record time units as well as volume units.
Worked Example: Applying Covariate and Individual-Effect Transformations
Consider a hypothetical estimation model with \(TVCL=5.200\ \mathrm{L/h}\), \(TVV=42.00\ \mathrm{L}\), \(WT_{\mathrm{ref}}=70\ \mathrm{kg}\), \(\theta_{WT,CL}=0.75\), and \(\theta_{WT,V}=1.00\). For one hypothetical individual, let \(WT_i=84\ \mathrm{kg}\), \(\eta_{CL,i}=0.10\), and \(\eta_{V,i}=-0.05\). Assume the simulation model expects natural-scale individual values in \(\mathrm{mL/h}\) and \(\mathrm{mL}\), and implements no additional covariate or random-effect adjustment to these supplied values.
First, calculate the weight ratio: \(84/70=1.2\). For clearance, apply both the weight exponent and the exponential individual effect:
The final multiplication is \(5.200\times1.1465\times1.1052\approx6.589\), to the stated precision. Converting the result for the simulation model gives \(6.589\times1000=6589\ \mathrm{mL/h}\).
For volume, apply its own weight exponent and random effect:
Thus, the value supplied for volume is approximately \(47940\ \mathrm{mL}\). The individual rate constant provides another consistency check:
The crosswalk must record that these are individual values, not typical values, and that the covariate and individual effects have already been applied. If the simulation model applies those effects again, it will double-adjust the parameters. Alternatively, the crosswalk could map the typical values and effects separately, provided the simulation model uses the same equations, reference weight, and effect conventions.
Track Parameter Scale and Reference Values
A parameter’s scale is part of its definition. For example, an estimation result may report a parameter on a logarithmic scale while a simulation model expects a positive natural-scale value. The crosswalk should identify the scale explicitly, including the logarithm base, rather than relying on a label such as “theta.”
Worked Example: Mapping a Log-Scale Typical Clearance
Suppose a hypothetical estimation model reports \(\log(TVCL)=1.64866\), using the natural logarithm, while a simulation model expects \(TVCL\) in \(\mathrm{L/h}\). Assume the reported value refers to typical clearance at the same reference conditions as the simulation parameter.
The required transformation is exponentiation:
If the simulation model instead expects \(\log(TVCL)\) and performs the exponentiation internally, the supplied value should remain 1.64866 on the log scale. Applying the exponential before supplying it would then be a scale mismatch. Likewise, if an estimation output already reports \(TVCL=5.200\ \mathrm{L/h}\) on the natural scale, exponentiating that value would be incorrect.
A reference value matters in the same way. If one model defines typical clearance at \(70\ \mathrm{kg}\) and another at \(1\ \mathrm{kg}\), identical labels and units do not imply identical typical values. The crosswalk must either align the reference values or calculate the corresponding value under the model’s covariate equation. A mapped number without its reference condition is incomplete.
Diagnostics, Alternatives, and Limitations
A crosswalk is useful because it forces checks that direct name matching can miss. Before using mapped values, verify that each row has an unambiguous scientific meaning, consistent units, a stated scale, and compatible reference conditions. For derived quantities such as \(k\), calculate them independently from the mapped values. Where covariate or random-effect terms are present, check whether they are supplied separately or already included in an individual value.
A small hand calculation is often an effective diagnostic. Use one hypothetical parameter set and one stated reference individual, then calculate the expected clearance, volume, and rate constant independently. Compare these calculations with the values the receiving model is configured to use. Disagreement can reveal a unit mismatch, an omitted or duplicated transformation, or a different reference condition. It does not, by itself, identify which model is wrong; inspect the definitions and equations.
| Approach | Strength | Main risk |
|---|---|---|
| Copy values by matching names | Quick when definitions are demonstrably identical | Misses differences in units, scale, or reference conditions |
| Map quantities through a crosswalk | Makes meanings and required transformations explicit | Depends on correct documentation of both models |
| Reparameterize one model to match the other | Can make later model use more direct | May introduce errors if equations or reference values are changed incorrectly |
Common failure signals include implausibly large or small converted values, a rate constant that changes after a volume-only conversion, a covariate effect applied twice, and a value labeled “typical” that was calculated for a particular individual. These are reasons to recheck the crosswalk, not proof that the underlying model is invalid. The method also has limits: it documents and checks parameter mappings, but it does not establish that two full models have equivalent event handling, residual-error behavior, or predictions. Those require separate model-specific checks.
- Wrong reasoning: “The source and destination fields are both named CL, so the value can be copied unchanged.”
- Correction: Confirm that both fields mean clearance, use compatible volume and time units, are on the same scale, and refer to the same covariate and individual-effect conditions. Convert or calculate as needed, then record the mapping.
Check Your Understanding
Answer each question using the stated units and conditions. Solutions follow the questions.
- A model reports \(CL=3.6\ \mathrm{L/h}\) and \(V=30\ \mathrm{L}\). A receiving model expects \(\mathrm{mL/h}\) and \(\mathrm{mL}\). What values should be supplied, and what is \(k\) in \(\mathrm{h^{-1}}\)?
- Why is \(TVCL\) not necessarily the same as an individual’s clearance?
- A model reports \(\log(TVCL)=\log(4.0)\) using natural logarithms. What is \(TVCL\), and what additional check is needed before mapping it?
- A mapped clearance is converted from \(\mathrm{L/h}\) to \(\mathrm{mL/h}\), but volume remains in liters. Why is calculating \(CL/V\) from those two supplied values invalid without further conversion?
Solutions. (1) \(CL=3.6\times1000=3600\ \mathrm{mL/h}\) and \(V=30\times1000=30000\ \mathrm{mL}\). Then \(k=3.6/30=0.120\ \mathrm{h^{-1}}\), or equivalently \(3600/30000=0.120\ \mathrm{h^{-1}}\). (2) \(TVCL\) is the typical value at specified reference conditions; an individual’s value can differ because of covariates and an individual random effect. (3) Exponentiating gives \(TVCL=\exp(\log 4.0)=4.0\) in the units associated with clearance. Check that the receiving model expects natural-scale clearance and uses the same units and reference conditions. (4) The numerator and denominator then use different volume units, so their numerical ratio is not a valid rate in \(\mathrm{h^{-1}}\). Convert volume to milliliters or clearance to liters per hour before dividing.