Interpreting Covariate Effects During Rapid Growth

Why Clearance Can Change Rapidly During Growth

A child’s modeled clearance may increase over time because body weight increases, because maturation progresses, or because both occur together. These contributions are easy to conflate when looking only at the beginning and ending predictions. Separating them makes the model’s meaning clearer: it distinguishes a size-related change from an age-related change, while recognizing that the observed longitudinal change combines both.

The comparisons in this tutorial use an illustrative size-and-maturation model. Its numerical values are hypothetical and are chosen to make the decomposition transparent; they are not estimates from a real study. The maturation curve and allometric exponent would need to be justified for a particular drug and data set.

Definition: Let $CL_i(t)$ denote the modeled clearance for individual $i$ at time $t$, $WT_i(t)$ the individual’s body weight, and $AGE_i(t)$ the age used by the model. In the illustrative model, $$ CL_i(t)=CL_{\mathrm{ref}}\left(\frac{WT_i(t)}{WT_{\mathrm{ref}}}\right)^{0.75}F_{\mathrm{mat}}\!\left(AGE_i(t)\right), \qquad F_{\mathrm{mat}}(A)=\frac{A^2}{A^2+TM_{50}^2}. $$ Here $CL_{\mathrm{ref}}$ is clearance at reference weight $WT_{\mathrm{ref}}$ when maturation is complete, and $TM_{50}$ is the age at which this particular maturation function equals 0.5. The age $A$ and $TM_{50}$ must use the same units. The function $F_{\mathrm{mat}}(A)$ lies between 0 and 1 for positive age and approaches 1 as age increases.

For the examples, take $CL_{\mathrm{ref}}=5$ L/h, $WT_{\mathrm{ref}}=70$ kg, and $TM_{50}=0.5$ years. The clearance value and reference weight simply set the scale; they cancel from the ratios used to separate contributions. The exponent 0.75 and the maturation function are stipulated model choices, not universal biological laws.

Three Comparisons That Answer Different Questions

A fixed-age comparison asks how the model prediction changes with weight when age is held constant. A fixed-weight comparison asks how it changes with age when weight is held constant. Both are conditional model contrasts: they isolate a term mathematically, but need not describe a pair of children who could be observed at exactly the same age or weight.

A longitudinal comparison follows the model from one age-and-weight combination to another. It includes both changes. In a model with separate multiplicative terms, the ratio between the ending and starting predictions can be written as the product of a weight ratio and a maturation ratio:

\frac{CL_2}{CL_1} = \left(\frac{WT_2}{WT_1}\right)^{0.75} \frac{F_{\mathrm{mat}}(AGE_2)}{F_{\mathrm{mat}}(AGE_1)}.

The subscripts 1 and 2 refer to the starting and ending points. This factorization assumes the same reference values and model parameters at both points. If additional time-varying covariates or model terms are present, their contributions must also be included rather than attributed to weight or maturation.

Worked Example: Weight Change at a Fixed Age

Consider a hypothetical child aged 0.5 years whose weight changes from 6 kg to 9 kg. What is the modeled clearance change attributable to weight if age is held fixed?

At age 0.5 years, the maturation function is the same in both predictions: $F_{\mathrm{mat}}(0.5)=0.5^2/(0.5^2+0.5^2)=0.5$. Thus it cancels from the ratio, as do the reference clearance and reference weight:

\frac{CL(9\ \mathrm{kg},0.5\ \mathrm{y})}{CL(6\ \mathrm{kg},0.5\ \mathrm{y})} = \frac{5(9/70)^{0.75}(0.5)}{5(6/70)^{0.75}(0.5)} = \left(\frac{9}{6}\right)^{0.75} = 1.5^{0.75} \approx 1.355.

The prediction at 9 kg is therefore approximately 1.355 times the prediction at 6 kg, or approximately 35.5% higher, at the same age. This is the model-implied weight contrast; it does not say that weight alone caused a real child’s clearance to change by that amount.

Worked Example: Maturation Change at a Fixed Weight

Now hold weight at 9 kg and compare ages 0.25 and 0.5 years. The maturation values are:

F_{\mathrm{mat}}(0.25) = \frac{0.25^2}{0.25^2+0.5^2} = \frac{0.0625}{0.3125} = 0.2, \qquad F_{\mathrm{mat}}(0.5) = \frac{0.25}{0.25+0.25} = 0.5.

Because weight is fixed, the allometric term cancels. The age-related ratio is $0.5/0.2=2.5$. Thus, under this model, the prediction at 0.5 years is 2.5 times the prediction at 0.25 years at the same weight, an increase of 150%. This is the maturation component specified by the assumed curve, not an estimate of a causal effect of age.

Worked Example: Combining Weight and Maturation During Growth

Suppose a hypothetical child changes from 6 kg at age 0.25 years to 9 kg at age 0.5 years. Both covariates change, so use both terms in the longitudinal ratio. The weight factor is the one calculated in the first example, $1.5^{0.75}\approx1.355403$. The maturation factor is $0.5/0.2=2.5$. Therefore:

\frac{CL_2}{CL_1} = \left(\frac{9}{6}\right)^{0.75} \frac{0.5}{0.2} \approx 1.355403\times2.5 = 3.388508.

The ending prediction is approximately 3.389 times the starting prediction, or approximately 238.9% higher. The 35.5% weight contrast and 150% maturation contrast must not be added: their ratio factors multiply. This result is conditional on the illustrative model and the stated covariate values.

Why the Decomposition Works

The ratio formula follows by writing the model at each point and dividing. At point 1, clearance is $CL_{\mathrm{ref}}(WT_1/WT_{\mathrm{ref}})^{0.75}F_{\mathrm{mat}}(AGE_1)$; at point 2 it has the same form with $WT_2$ and $AGE_2$. Division cancels the common reference quantities. The quotient of the weight terms is $(WT_2/WT_1)^{0.75}$, and the quotient of the maturation terms is $F_{\mathrm{mat}}(AGE_2)/F_{\mathrm{mat}}(AGE_1)$. Multiplying these gives the stated result.

The same separation can be expressed on the log scale. Taking natural logarithms of the ratio converts the product into a sum:

\log\left(\frac{CL_2}{CL_1}\right) = 0.75\log\left(\frac{WT_2}{WT_1}\right) + \log\left(\frac{F_{\mathrm{mat}}(AGE_2)}{F_{\mathrm{mat}}(AGE_1)}\right).

This additive expression is useful for describing relative contributions: the log change has one term for weight and one for maturation. It does not mean that percentage changes in clearance can be added. To return to the clearance ratio, exponentiate the total log change; equivalently, multiply the two component ratios.

Interpretation, Diagnostics, and Limitations

For a careful interpretation, state what is held fixed and what changes. “At the same age, the model predicts a 35.5% higher clearance at 9 kg than at 6 kg” is a weight-only contrast. “At the same weight, the model predicts a 2.5-fold change between these ages” is a maturation-only contrast. “Along the specified growth path, the model predicts a 3.389-fold change” includes both. These are model-based statements; they do not establish that either covariate independently causes a change.

Before interpreting a longitudinal prediction, check that the age definition matches the model and that age and weight units are consistent with the fitted covariates. Confirm that the child’s values lie within, or are reasonably supported by, the data used to estimate the model. A mathematically valid ratio can still be an unreliable extrapolation. Also verify that the same clearance definition and model parameters are being compared at both times.

A useful diagnostic is to calculate each component ratio separately and then multiply them. If that product does not equal the ratio of the two model predictions, a term may have been omitted or a value substituted incorrectly. Plotting the weight factor, maturation factor, and combined prediction over age can also reveal whether one term dominates unexpectedly or whether a rapid rise is driven by the assumed maturation curve. The curve’s shape and sensitivity to its parameters should be examined rather than treated as known with certainty.

The fixed-age and fixed-weight contrasts are not alternatives to the longitudinal prediction; they answer different questions. A standardized clearance comparison can help describe size-adjusted values, while direct model predictions describe clearance under specified covariates. Neither removes uncertainty in the model, parameter estimates, or individual covariate measurements. A further limitation is that weight and age commonly change together in children, so an isolated contrast may represent a model-based conditional comparison rather than a directly observed contrast. The course’s earlier material on age-weight collinearity and identifiability explains why the available data may not distinguish these effects precisely.

A Practical Reading Sequence

Write the model terms.
Identify which term depends on weight and which depends on age or maturation. Record the units and reference values.

Specify the comparison.
For a weight-only contrast, hold age fixed; for a maturation-only contrast, hold weight fixed. For a child’s trajectory, allow both to change.

Calculate component ratios.
Substitute the starting and ending values in each changing term. Common reference terms cancel when the model parameters are unchanged.

Combine and interpret.
Multiply component ratios, then express the result as a fold change or as a percent change using $(\text{ratio}-1)\times100\%$. State the comparison and its model-based nature.

Common Mistakes

Adding percentage changes. A 35.5% weight contrast and a 150% maturation contrast do not combine to a 185.5% change. The modeled factors multiply; in the combined example the total increase is approximately 238.9%.
Calling a fixed-covariate comparison an observed effect. Holding age or weight fixed is a way to isolate a model term. It is not automatically a feasible observed comparison or a causal conclusion.
Reporting only the combined change. A large longitudinal ratio can hide which modeled term drives it. Report the weight and maturation factors separately when the goal is interpretation.
Ignoring the curve’s assumptions. A maturation multiplier is a chosen functional form with estimated or stipulated parameters. Its predictions, especially where data are sparse, depend on those choices.

Key takeaway: For a multiplicative size-and-maturation model, the modeled clearance ratio along a growth path is the product of the weight ratio and maturation ratio. State whether the comparison holds age fixed, holds weight fixed, or allows both to change.

Check Your Understanding

Answer each question using the illustrative model and conventions in this tutorial.

At a fixed age, weight rises from 4 kg to 8 kg. What is the predicted weight-only clearance ratio?
At fixed weight, maturation changes from $F_{\mathrm{mat}}=0.2$ to $F_{\mathrm{mat}}=0.5$. What is the ratio and percent increase?
Why is it incorrect to add the separate percentage increases in the combined worked example?
What should be checked if the product of the two component ratios differs from the ratio of the model’s endpoint predictions?

Solutions. 1. The ratio is $(8/4)^{0.75}=2^{0.75}\approx1.682$, so the prediction is approximately 68.2% higher. 2. The ratio is $0.5/0.2=2.5$, a 150% increase. 3. The model is multiplicative: the weight and maturation factors multiply, so the separate percent changes do not add. 4. Check the substituted covariate values, units, model terms, and whether any additional changing terms were omitted; with the stated model and unchanged parameters, the component product must equal the endpoint prediction ratio.