Integrated Case: Building a Size-and-Maturation Model

Why Combine Size and Maturation?

A pediatric patient’s clearance can differ from an adult’s for more than one reason. Body size affects the scale of clearance, while physiological maturation can change clearance capacity independently of weight. A model that accounts for only one of these features may leave a systematic age or size trend in its predictions. The integrated model developed here represents both effects explicitly, so that each prediction can be traced to a weight multiplier and a maturation multiplier.

The form below is a useful modeling framework, not a universal biological law. The reference weight, allometric exponent, maturation function, and parameter values must be appropriate to the drug and data. All numerical values in the examples are hypothetical and are used only to illustrate the calculation.

Definition: Let $CL_i$ denote the typical clearance prediction for an individual with body weight $WT_i$ in kilograms and age $AGE_i$ in years. The integrated size-and-maturation model is $$ CL_i=CL_{70}\left(\frac{WT_i}{70\ \mathrm{kg}}\right)^{0.75}MAT(AGE_i), $$ where $CL_{70}$ is clearance at the 70 kg reference weight when maturation is complete, and $$ MAT(AGE_i)=\frac{AGE_i^{Hill}}{TM_{50}^{Hill}+AGE_i^{Hill}}. $$ Here $TM_{50}$ is the age at which $MAT=0.5$, and $Hill$ controls the steepness of the maturation curve. The model’s predictions have units of clearance, such as L/h; the size and maturation multipliers are dimensionless.

Assumptions and Interpretation of the Components

The model assumes that weight acts through a power function with exponent 0.75, and that age-dependent maturation acts as a multiplicative fraction between zero and one. In this specification, size and maturation effects combine without an additional interaction: at a given age, changing weight changes clearance by the allometric ratio; at a given weight, changing age changes it by the maturation ratio. The exponent 0.75 is a modeling choice here, not a value that must apply to every drug or population.

The reference weight is a scaling anchor. At 70 kg, the size multiplier equals one. When age is large relative to $TM_{50}$, the maturation fraction approaches one, so $CL_{70}$ represents the limiting clearance at the reference weight under complete maturation. At finite ages, the maturation fraction is generally below one; therefore, a patient’s clearance at 70 kg need not equal $CL_{70}$.

The maturation function is a saturating Hill function. For nonnegative age and positive $TM_{50}$ and $Hill$, it lies between zero and one, increases with age, equals 0.5 at $AGE=TM_{50}$, and approaches one as age increases. In the examples, age is measured in years, $TM_{50}=0.50$ years, and $Hill=2$. This is a hypothetical specification; using age in a different unit would require expressing $TM_{50}$ in that same unit.

The equation gives a typical structural prediction. It does not, by itself, represent unexplained between-individual variability, assay or sampling error, or uncertainty in the parameter values. Those features require additional model components and data. The examples focus on the structural prediction so that the roles of size and maturation remain clear.

Building and Reading the Model

A useful way to interpret a prediction is to keep its three parts visible: the reference clearance, the size multiplier, and the maturation fraction. For a particular patient, first calculate the weight ratio relative to 70 kg and raise it to 0.75; then calculate the maturation fraction from age; finally multiply these factors by $CL_{70}$. Do not combine the two covariates into an unlabelled adjustment, because that makes it harder to check whether an implausible prediction comes from weight, age, or both.

The ratio of two predictions makes the model’s implications especially clear. For two patients under the same parameter values, the common reference clearance cancels. The ratio is the weight ratio raised to the allometric exponent, multiplied by the ratio of their maturation fractions. Taking logarithms expresses this multiplicative relationship as a sum. These ratio identities assume $WT_A,WT_B>0$, $AGE_A,AGE_B>0$, and $CL_{70}\ne0$. At age zero, $MAT=0$ and predicted clearance is zero; the clearance ratio is undefined if patient B is age zero, while the logarithmic identity is undefined if either patient is age zero.

\frac{CL_A}{CL_B} = \left(\frac{WT_A}{WT_B}\right)^{0.75} \frac{MAT(AGE_A)}{MAT(AGE_B)}

\log\left(\frac{CL_A}{CL_B}\right) = 0.75\log\left(\frac{WT_A}{WT_B}\right) + \log\left(\frac{MAT(AGE_A)}{MAT(AGE_B)}\right)

This identity follows by substituting the model for each clearance, dividing, and cancelling the shared $CL_{70}$. It shows that the predicted difference is not attributable to weight alone when maturation fractions differ. It also provides a check on calculations: the ratio of separately calculated predictions should agree, up to rounding, with the product of the weight and maturation ratios.

Worked Example: Predicting Clearance for a Young Child

Suppose a hypothetical model uses $CL_{70}=5.00\ \mathrm{L/h}$, $WT=22.719\ \mathrm{kg}$, $AGE=1.50$ years, $TM_{50}=0.50$ years, and $Hill=2$. Calculate the typical clearance prediction.

First, calculate the size multiplier:

\left(\frac{22.719}{70}\right)^{0.75} \approx 0.43000.

Next, calculate the maturation fraction:

MAT(1.50)=\frac{1.50^2}{0.50^2+1.50^2} =\frac{2.25}{2.50}=0.9000.

Substituting both factors into the clearance model gives

CL =5.00\ \mathrm{L/h} \left(\frac{22.719}{70}\right)^{0.75} (0.9000) \approx 1.9350\ \mathrm{L/h}.

Thus, under these hypothetical model assumptions, the predicted clearance is approximately $1.935\ \mathrm{L/h}$. The weight multiplier reduces the reference clearance, and the maturation fraction reduces it further. The result is a model prediction, not a measured clearance for a real child.

Worked Example: Separating Size and Maturation at a Lower Age

Use the same hypothetical model parameters for a patient weighing $4.00\ \mathrm{kg}$ at age $0.10$ years. Calculate the prediction, then compare it with the prediction for the same weight at age $0.50$ years.

The size multiplier is

\left(\frac{4.00}{70}\right)^{0.75} \approx 0.11687.

At age $0.10$ years, maturation is

MAT(0.10) =\frac{0.10^2}{0.50^2+0.10^2} =\frac{0.01}{0.26} \approx 0.03846154.

Using the unrounded size multiplier in the calculation,

CL(0.10) =5.00\ \mathrm{L/h} \left(\frac{4.00}{70}\right)^{0.75} \left(\frac{0.01}{0.26}\right) \approx 0.02248\ \mathrm{L/h}.

At age $0.50$ years, age equals $TM_{50}$, so maturation is exactly 0.5. The prediction is

CL(0.50) =5.00\ \mathrm{L/h} \left(\frac{4.00}{70}\right)^{0.75} (0.5) \approx 0.2922\ \mathrm{L/h}.

The weight is identical in the two calculations, so the difference is entirely due to maturation in this model. The ratio of the maturation fractions is $0.5/(1/26)=13$, and the clearance predictions differ by the same factor before rounding. This is a consequence of the multiplicative model, not a claim that every real patient’s clearance increases thirteenfold over this age interval.

Worked Example: Comparing Two Patients with Both Covariates Changing

Compare the hypothetical patient weighing $22.719\ \mathrm{kg}$ at age $1.50$ years with the patient weighing $4.00\ \mathrm{kg}$ at age $0.10$ years. Use the same model parameters as above and calculate the clearance ratio from the covariate ratios.

The weight contribution to the ratio is

\left(\frac{22.719}{4.00}\right)^{0.75}\approx 3.679.

The maturation contribution is

\frac{0.9000}{0.038461538\ldots}=23.40.

Therefore, using unrounded component values,

\frac{CL_A}{CL_B} = \left(\frac{22.719}{4.00}\right)^{0.75} \frac{0.9000}{0.038461538\ldots} \approx 86.09.

The model predicts about 86.09 times as much clearance for patient A as for patient B under these assumptions. Both the larger weight and the greater maturation fraction contribute to the ratio. The ratio is not an estimate of a causal effect of age or weight: the model describes predictions conditional on its chosen functional forms and parameters.

Checking the Model and Recognizing Failure

A model can produce precise arithmetic and still be poorly suited to the data. Assessment should therefore examine whether the combined structure is plausible across the observed ages and weights, and whether the covariate patterns left after fitting are consistent with random variation rather than systematic misspecification. The earlier material on evaluating maturation curves and diagnosing residual size trends provides useful context for these checks.

Inspect predictions across age and weight. Tabulate or plot predictions over the observed covariate range. Check that the maturation curve transitions in a region supported by data and that the resulting clearances are physiologically plausible.
Look for residual covariate patterns. If prediction errors or residuals trend with weight at comparable ages, the size function or exponent may be inadequate. If they trend with age at comparable weights, the maturation function may be inadequate. A combined pattern may indicate that the assumed separable effects do not capture the data.
Check covariate overlap and identifiability. If nearly all younger patients are small and nearly all older patients are large, age and weight effects may be difficult to distinguish. A fitted model can then give unstable or strongly dependent estimates even if its overall predictions appear reasonable.
Compare observed support with prediction range. Predictions far outside the weights or ages represented in the data are extrapolations. They depend strongly on the assumed power law or maturation curve and should not be treated as equally well supported.

Alternative approaches answer somewhat different questions. A size-only model omits maturation and may leave an age pattern in the residuals; a maturation-only model omits size and may leave a weight pattern. Age categories can be easier to summarize but impose abrupt boundaries and discard within-category age differences. A different smooth maturation function may better represent particular data, but adding flexibility requires enough informative observations to estimate its parameters. The integrated model is attractive when a compact, interpretable representation is justified; it is not automatically preferable to every alternative.

Limitations and a Common Pitfall

The model assumes a fixed allometric exponent and a specific maturation curve. If either assumption is wrong, the model can be systematically biased, especially at the extremes of age or weight. A simple multiplicative structure also assumes that the weight effect does not change with age beyond its combination with the maturation fraction. Data that do not span overlapping ages and weights may not support a reliable separation of these effects. Finally, $CL_{70}$, $TM_{50}$, and $Hill$ are not interchangeable: the first sets the clearance scale, while the latter two govern the age pattern.

Common mistake: It is tempting to call $CL_{70}$ “the clearance of a 70 kg child” or to treat the maturation fraction as a second size adjustment. Neither interpretation is correct. $CL_{70}$ is the model’s clearance at the reference weight under complete maturation, and $MAT$ is a dimensionless age-dependent fraction. For a 70 kg patient at age $AGE$, the model predicts $CL_{70}\times MAT(AGE)$, not necessarily $CL_{70}$.

Another avoidable error is rounding intermediate multipliers too early. For example, using a rounded size factor in several subsequent calculations can shift the last reported digits or make two methods of calculating a ratio appear inconsistent. Retain full precision through the calculation and round only the reported results.

Key takeaway: A size-and-maturation prediction is the reference clearance multiplied by a weight-based allometric factor and an age-based maturation fraction. Interpret each component separately, check predictions across the data range, and treat the result as conditional on the model’s assumptions.

Check Your Understanding

Answer each question using the hypothetical model $CL_{70}=5.00\ \mathrm{L/h}$, $TM_{50}=0.50$ years, and $Hill=2$, unless the question specifies otherwise.

What is the maturation fraction at age $0.50$ years, and why?
For a patient weighing $10.0\ \mathrm{kg}$ at age $0.50$ years, calculate the predicted clearance.
At a fixed weight, how does the predicted clearance at age $1.50$ years compare with that at age $0.50$ years?
Why might the size and maturation effects be difficult to distinguish if younger patients are almost all smaller than older patients?
What does $CL_{70}$ represent, and what would the model predict for a 70 kg patient at age $0.50$ years?

Solutions. (1) $MAT=0.50^2/(0.50^2+0.50^2)=0.5$, because age equals $TM_{50}$. (2) The size factor is $(10.0/70)^{0.75}\approx0.232368$; therefore $CL=5.00(0.232368)(0.5)\approx0.5809\ \mathrm{L/h}$. (3) At age 1.50 years, maturation is 0.9000, compared with 0.5000 at age 0.50 years. At fixed weight, the predicted clearance ratio is $0.9000/0.5000=1.800$. (4) If age and weight change together with little overlap, the observed data provide limited information for separating an age effect from a weight effect; different combinations of covariate effects may explain similar patterns. (5) $CL_{70}$ is the clearance at the 70 kg reference weight when maturation is complete. At age 0.50 years, the weight factor is one but maturation is 0.5, so the prediction is $5.00(1)(0.5)=2.50\ \mathrm{L/h}$.