Distinguishing Maturation From Age-Varying Allometry

Why Separate Maturation From Allometry?

In pediatric pharmacokinetics, clearance may differ with both body size and developmental age. A common formulation assigns these roles separately: body weight acts through a fixed allometric exponent, while a maturation function describes age-related development. An alternative lets the allometric exponent itself vary with age. The distinction matters because these models do not merely use different mathematical notation; they can predict different age contrasts at the same weight, and even opposite directions of change at weights on opposite sides of a reference value.

The earlier material on combining allometry and maturation established the use of a size term and an age-dependent maturation term in one clearance model. Here the focus is narrower: determine what each term implies, how to compare predictions fairly, and when the data can distinguish the explanations.

The Two Formulations and Their Assumptions

A typical fixed-exponent model for clearance is

Here \(\theta_{CL}\) is clearance at the reference weight when maturation is complete and the individual random effect is zero. The exponent \(\theta\) describes the body-weight scaling, and \(\eta_{CL,i}\) is an individual-level random effect on log clearance. The maturation function is dimensionless and ordinarily lies between zero and one when it represents incomplete maturation approaching an adult plateau. A particular functional form, such as a sigmoidal function, is a scientific and modeling choice rather than a requirement of this comparison.

The pure age-varying-exponent alternative is

In this second formulation, age changes the exponent that maps weight to clearance. As written, there is no separate maturation multiplier. Both models can include the same residual error model and random-effects structure; when comparing their structural predictions, hold those features and the reference definitions consistent. The age variable and its units must also be defined consistently, and any random effects must be interpreted under the distribution specified for the model.

The formulas encode different restrictions. The fixed-exponent model assumes that, after accounting for the maturation function, the weight elasticity is constant across ages. The age-varying-exponent model assumes that age acts through a changing weight elasticity, and that its age effect is zero at exactly the reference weight: there the weight ratio is one, and \(1^{\theta(a)}=1\) for every age. Neither restriction is automatically true simply because it is convenient to fit.

Deriving the Difference in Age Contrasts

A useful way to compare these formulations is to take the ratio of predicted clearances at two ages while holding weight and the individual random effect fixed. Define \(r=WT/WT_{\mathrm{ref}}\). Under the fixed-exponent maturation model, the ratio is

The size term cancels. Thus this model predicts the same proportional age contrast at every body weight, provided the maturation function is the same for those individuals. Taking logarithms of the age-varying-exponent model instead gives

so its age contrast is

This contrast is zero at the reference weight, positive or negative according to both the exponent change and the sign of \(\log(r)\), and generally different across weights. That result follows directly from the model equations; it is not a claim that one structure is physiologically correct in every setting.

What the Data Must Support

The mathematical distinction does not guarantee that a particular data set can estimate it. If age and weight move together closely, the observed records may contain few individuals of different ages at similar weights or of similar ages at different weights. In that situation, an apparent age trend can be attributed to maturation, to changing allometry, or to both, with little empirical leverage to distinguish the explanations. Sparse observations at the youngest ages or near the adult plateau make this problem more serious.

A meaningful comparison requires the reference weight and age scale to be explicit, the same endpoint and units in both models, and an age range supported by observations. The maturation function should have its specified reference value and a plausible shape over that range. For the age-varying-exponent model, the chosen function \(\theta(a)\) should be defined over the observed ages and should not be interpreted far outside them without justification. As with other NLME models, individual random effects and residual variability must be specified; they do not remove structural confounding between age and weight.

Residual patterns can reveal a problem but cannot by themselves prove which biological mechanism is responsible. For example, residual trends by age after fitting a fixed-exponent maturation model may indicate an inadequate maturation function, an age-dependent exponent, or another omitted feature. Predictions over a common age-by-weight grid make the model implications visible. Estimates should be treated cautiously where the grid extends beyond the observed covariate support.

The two structures are generally not nested as written, so a likelihood-ratio test that assumes nested models is not automatically appropriate. They can be compared using suitable information criteria or predictive performance, with attention to parameter counts and validation design. A richer hybrid with both \(M(a)\) and \(\theta(a)\) may improve fit, but that improvement does not establish that both mechanisms are separately identifiable or biologically distinct. Uncertainty in the parameters and the quality of covariate support remain central.

Common Mistake and Practical Limits

Wrong reasoning: “Both models include age, so they should make the same predictions; the exponent is just another way to represent maturation.” This is incorrect. In the fixed-exponent model, the proportional age contrast is \(M(a_2)/M(a_1)\), independent of weight. In the pure age-varying-exponent model, the log age contrast is \([\theta(a_2)-\theta(a_1)]\log(WT/WT_{\mathrm{ref}})\), which is zero at reference weight and changes with weight.

Correction: Compare the formulas at common weights, not just the fitted age coefficients or predictions at each individual's observed covariates. If an age effect is intended to remain present at reference weight, a pure exponent-only formulation cannot express it; adding an age multiplier creates a hybrid model and changes the question. Conversely, a maturation multiplier does not allow the weight elasticity to change with age unless that feature is explicitly added.

Neither structure is universally superior. The fixed-exponent maturation model is interpretable when evidence and scientific understanding support a common size scaling modified by developmental function. An age-varying exponent may be useful when the relationship between weight and clearance plausibly changes across development and the data support estimation of that change. Both can fail if the age function is too rigid, reference values are poorly chosen, age and weight are confounded, or predictions are extrapolated beyond observed data. Model fit alone cannot convert an empirical age pattern into a causal physiological explanation.