Gestational Age and Postnatal Age in Neonatal Dosing

Why neonatal age needs careful definition

For neonatal pharmacokinetics, “age” is not a single unambiguous quantity. Two infants measured on the same day after birth may have different developmental histories, and two infants at the same postmenstrual age may have spent different amounts of time outside the uterus. Those distinctions can matter when clearance is changing rapidly during early life.

A model cannot resolve this by using an age variable without defining it. The analyst must specify the reference point, units, and time at which age is evaluated. A value recorded at birth is not interchangeable with a value calculated at a later concentration sample. Nor does an association between age and clearance, by itself, establish which biological process caused the change.

In clinical records, “gestational age” may be used informally to mean gestational age at birth or current PMA. A quantitative analysis should avoid that ambiguity: label the variable explicitly and record how it was calculated. For example, PMA of 31 weeks and 4 days is not the same as PNA of 31 weeks and 4 days.

Calculating and recording the age measures

Weeks and days are often more precise than rounded weeks. To add ages, convert to a common unit first, or add days while carrying each complete 7-day period into a week. Keep the original precision where possible; rounding an age near a steep part of a maturation curve can change the model prediction.

For a longitudinal model, the age value should correspond to the time of the observation or dose event to which it applies. If several samples are collected over time, PNA and PMA both increase as the infant ages. Gestational age at birth, in contrast, is a baseline characteristic and remains fixed for that infant. A dataset should distinguish these time-varying and fixed quantities rather than use one column labelled simply “age.”

Choosing an age representation for clearance

Maturation models represent how a clearance-related process changes with development. The choice of age input is a scientific and modelling choice: it asserts which time scale best represents the developmental change of interest. PMA can be useful when the process is thought to track development accumulated before and after birth. PNA can be useful when a process is associated with time since birth or with adaptation to life outside the uterus. These are possible representations, not universal biological rules.

The earlier material on maturation functions describes how an age-dependent maturation factor can be incorporated in clearance. The key issue here is the age supplied to that function. As a simple illustration, let the maturation factor be \(F(a)=a/(a+20)\), where \(a\) is age in weeks. This hypothetical function increases with age and approaches 1, but it does not assert that all drugs or developmental processes follow this shape.

If the maturation factor multiplies a size-scaled clearance, the age definition affects the resulting clearance prediction even when weight and all other covariates are unchanged. For instance, a model may have the form \(CL=CL_{\mathrm{ref}}(WT/WT_{\mathrm{ref}})^{0.75}F(a)\). The exponent and reference values belong to the specified model; replacing \(a\) with another age measure changes the model, not merely the way its output is labelled.

A further possibility is that clearance depends on both a developmental time scale and a distinct postnatal process. That possibility needs a model whose parameters can be estimated from the available observations. It should not be represented by adding every available age variable to a regression and assuming that each coefficient has a separate biological interpretation.

An exact relationship with estimation consequences

PMA, PNA, and gestational age at birth are not three independent measurements. When expressed in the same units, PMA is exactly the sum of the other two. This has a direct consequence for a model containing unrestricted linear effects of all three.

This is an algebraic limitation, not a question of having a large enough sample. Substituting \(M=G+P\) into a linear predictor gives:

Only the sums of coefficients on the right can be determined from the predictor. To see that there are infinitely many equivalent coefficient sets, choose any number \(c\) and define \(\beta_G'=\beta_G-c\), \(\beta_P'=\beta_P-c\), and \(\beta_M'=\beta_M+c\). Then \(\beta_G'+\beta_M'=\beta_G+\beta_M\) and \(\beta_P'+\beta_M'=\beta_P+\beta_M\). The predictions are unchanged, so the three original coefficients cannot be uniquely separated.

This result concerns the simultaneous inclusion of all three as unrestricted linear terms. Nonlinear age functions or interactions may not produce the same exact linear dependence among their model columns. However, estimating distinct effects still requires informative variation in the data and a defensible model. A fitted coefficient alone does not show that a biological effect is uniquely identified.

Assumptions, diagnostics, and limitations

An age-based clearance model relies on more than correct arithmetic. Its interpretation depends on a clear age definition, consistent units, appropriate timing, and a functional form supported by the observations. If age is used as a proxy for maturation, the model also assumes that the selected age scale adequately represents the developmental process for the population and drug being studied. These assumptions should be considered before using predictions to support a dosing analysis.

Useful checks include:

• Check the age construction. Recalculate PMA from gestational age at birth and PNA for a sample of records. Look for mixed units, missing birth dates, inconsistent rounding, or an “age” field whose meaning changes across records.
• Check the timing. Confirm that the age attached to a concentration or dose corresponds to that event, rather than to study entry or birth alone.
• Check support and overlap. Examine whether the data include infants with different combinations of gestational age at birth and PNA. Sparse combinations can make comparisons between age representations depend heavily on model assumptions.
• Check the fit by age. Look for systematic differences between observed data and model predictions across age ranges. A smooth average age effect can still miss a poorly represented region or individual variability.
• Check the specification. Do not fit all three age measures as unrestricted linear terms. If a complex model has unstable estimates, large standard errors, or strong sensitivity to small specification changes, inspect for redundant or weakly supported age information.

A useful comparison is between age representations that have distinct scientific interpretations. A PMA-based model describes change as a function of total postmenstrual age. A PNA-based model describes change with time since birth. A model using gestational age at birth and PNA asks whether those two recorded quantities help represent the outcome under its chosen structure. Compare models using appropriate fit and predictive checks, but do not choose an age measure solely because it produces a preferred coefficient or prediction.

Age is also an imperfect proxy for development. Infants of the same PMA may differ in clinical condition and developmental history; an age function cannot automatically account for those differences. Conversely, a model fitted over a narrow age range may not support reliable extrapolation to substantially earlier or later ages. The purpose of an age term is to represent a supported relationship, not to guarantee that every developmental difference has been captured.

Common mistake

A frequent error is to treat “gestational age,” PMA, and PNA as interchangeable labels, or to include all three as independent linear covariates. The first mistake changes the meaning and numerical value of the input without changing the model description. The second creates exact redundancy because PMA equals gestational age at birth plus PNA. The correction is to define each quantity, calculate it at the relevant time, choose an age representation consistent with the scientific question, and state the resulting interpretation explicitly.