Gestational Age, Prematurity, and Developmental Age

Why Age at Birth Matters

An infant's age in a pharmacokinetic data set can refer to different clocks. A baby born at 28 weeks' gestation and observed four weeks later is not at the same developmental age as a baby born at 40 weeks' gestation and observed four weeks later. If a maturation model uses postnatal age for both, it assigns them the same age despite their 12-week difference in time since conception. That choice can affect the estimated maturation curve and the clearance predictions derived from it.

The key modeling decision is not simply whether age matters, but which age scale represents the developmental process being modeled. For many developmental processes, postmenstrual age (PMA) is used as a practical index of development. That is a modeling choice, not a claim that every aspect of drug disposition depends on PMA alone. Birth itself can introduce changes that a single developmental-age curve does not represent.

Definition: Gestational age (GA) is the duration of gestation at birth, usually recorded in weeks. Postnatal age (PNA) is elapsed time since birth. Postmenstrual age (PMA) is the sum of gestational age at birth and postnatal age: \( \mathrm{PMA}=\mathrm{GA}+\mathrm{PNA} \). Corrected age, when used for an infant born preterm, is postnatal age minus the number of weeks by which birth preceded 40 weeks' gestation. These quantities have different origins and are not interchangeable.

For example, an infant born at 28 weeks' gestation and observed 12 weeks after birth has a PMA of 40 weeks and a corrected age of zero weeks. The PMA describes elapsed developmental time from the start of gestation, while corrected age expresses how far the infant is beyond the expected due date. Corrected age can be useful for some clinical descriptions, but it is not automatically the appropriate age variable for a PK maturation function.

Using Age in a Maturation Model

A common sigmoid maturation function maps age to a fraction between zero and one. For a PMA-based model, let \(a\) denote PMA in weeks, \(M(a)\) the maturation fraction, \(TM_{50}\) the age at which the fraction is one-half, and \(\gamma\) the Hill coefficient. Assume \(TM_{50}>0\), \(\gamma>0\), and \(a\geq 0\).

M(a)=\frac{a^\gamma}{a^\gamma+TM_{50}^{\gamma}}

The fraction \(M(a)\) can scale a mature clearance component, for example, but its interpretation depends on the PK model in which it is used. The equation assumes that maturation is governed by the chosen age axis, that the curve increases from zero toward one, and that one \(TM_{50}\) and one \(\gamma\) adequately describe the population over the modeled age range. It does not by itself represent a birth-related adaptation effect or other independent developmental processes.

When \(a=TM_{50}\), the numerator and the \(TM_{50}^{\gamma}\) term are equal, so \(M(a)=1/2\). A larger \(\gamma\) makes the transition around \(TM_{50}\) steeper; it does not change the definition of \(TM_{50}\). Age units must be consistent: if \(a\) is entered in weeks, \(TM_{50}\) is in weeks as well.

Key interpretation: In a PMA-based maturation model, two infants with the same PMA receive the same maturation fraction, regardless of their different GA and PNA values. This is the implication of the model's age coding, not proof that their physiology is identical in every respect.

Worked Examples: Age Coding and Maturation

Worked Example: Different Gestational Ages at the Same Postnatal Age

Suppose a hypothetical PK model uses the maturation function above with \(TM_{50}=50\) weeks and \(\gamma=3\). Compare two hypothetical infants, each observed four weeks after birth: Infant A was born at 28 weeks' gestation, and Infant B at 40 weeks' gestation.

First calculate PMA. For Infant A, \(28+4=32\) weeks. For Infant B, \(40+4=44\) weeks. Their PMAs differ even though their PNAs are equal.

For Infant A, the model gives \(M(32)=32^3/(32^3+50^3)=32{,}768/(32{,}768+125{,}000)=32{,}768/157{,}768=0.2077\).

For Infant B, \(M(44)=44^3/(44^3+50^3)=85{,}184/(85{,}184+125{,}000)=85{,}184/210{,}184=0.4053\).

Under this stipulated PMA model, the predicted maturation fractions are about 0.208 and 0.405. If PNA were incorrectly substituted for PMA, both infants would be assigned \(a=4\) weeks, giving \(M(4)=4^3/(4^3+50^3)=64/125{,}064=0.0005117\). That coding would erase the modeled developmental difference between the infants and produce a very different prediction. The calculation illustrates the consequence of the assumed age axis; it does not establish which axis is biologically correct for a particular drug.

Worked Example: Same PMA, Different Birth Histories

Consider two hypothetical infants at PMA 40 weeks. Infant C was born at 28 weeks and is 12 weeks postnatal; Infant D was born at 32 weeks and is 8 weeks postnatal. Use \(TM_{50}=50\) weeks and \(\gamma=3\).

Both have the same PMA: \(28+12=40\) weeks and \(32+8=40\) weeks. The PMA model therefore predicts the same maturation fraction for each: \(M(40)=40^3/(40^3+50^3)=64{,}000/(64{,}000+125{,}000)=64{,}000/189{,}000=0.3386\).

Their corrected ages differ: Infant C has \(12-(40-28)=0\) weeks, whereas Infant D has \(8-(40-32)=0\) weeks. In this example both are at corrected age zero, as expected at 40 weeks PMA. More generally, corrected age is \( \mathrm{PMA}-40 \) weeks for preterm infants, so it is merely a shifted age scale in this setting. If a model uses corrected age, its zero point and valid domain must be defined explicitly; the PMA equation above cannot be applied mechanically to a negative corrected age because its fractional powers may not be defined for all \(\gamma\).

The shared prediction is a consequence of a PMA-only model. If observations at the same PMA systematically showed different PK behavior according to time since birth or GA at birth, that would be evidence that PMA alone may not describe the data adequately. It would not justify silently changing the age definition after seeing the results.

How Age Coding Changes Apparent Maturation Parameters

The relation between an age value and a maturation fraction can be rearranged to show how an age coding choice affects the implied \(TM_{50}\). Starting from the model, multiply both sides by the denominator and collect the terms involving \(TM_{50}\):

M(a)\{a^\gamma+TM_{50}^{\gamma}\}=a^\gamma

M(a)TM_{50}^{\gamma}=a^\gamma\{1-M(a)\}

For \(a>0\) and \(0<M(a)<1\), both sides can be divided by \(M(a)\), and taking the positive \(1/\gamma\) power gives:

TM_{50}=a\left(\frac{1-M(a)}{M(a)}\right)^{1/\gamma}

The conditions matter. At \(a=0\), the model gives \(M(0)=0\) for every positive \(TM_{50}\). That observation cannot identify \(TM_{50}\), and the inversion is not valid because it would divide by zero. Also, a single nonzero-age observation cannot determine both \(TM_{50}\) and \(\gamma\); the formula only recovers \(TM_{50}\) when \(\gamma\) is specified.

Worked Example: Apparent TM50 Under the Wrong Age Axis

Suppose the hypothetical observations in the previous example are consistent with a PMA-based maturation fraction of 0.3386 at PMA 40 weeks, with \(\gamma=3\). The infants have the same PMA but different PNAs: 12 and 8 weeks. What \(TM_{50}\) would be implied if each observation were incorrectly interpreted using PNA as the age variable?

Using the unrounded fraction \(M=64/189\), the ratio is \((1-M)/M=(125/189)/(64/189)=125/64\). Its cube root is 1.25. For the infant with PNA 12 weeks, the incorrectly coded apparent value would be \(12(1.25)=15\) weeks. For the infant with PNA 8 weeks, it would be \(8(1.25)=10\) weeks.

The two apparent values differ even though the infants have the same PMA and the stipulated PMA model assigns them the same maturation fraction. These values are not valid estimates of a common biological \(TM_{50}\); they demonstrate how substituting PNA for PMA can produce birth-history-dependent parameters. In actual population PK estimation, maturation fractions are typically linked to PK parameters through a model rather than observed directly, so this inversion is an illustration, not a general estimation procedure.

Model Choice, Diagnostics, and Limitations

A PMA-based model is often a reasonable starting point when the developmental process is expected to progress with time since conception. A PNA-based model instead places the origin at birth and may be appropriate for a process tied specifically to postnatal exposure or adaptation. A model with both GA and PNA, or with an explicit birth-related component, may be considered when there is a scientific rationale and enough informative data. These approaches answer different questions; they should not be treated as interchangeable parameterizations of the same curve.

A separate GA term is not automatically a remedy for poor age coding. GA, PNA, and PMA are mathematically related, so a model containing several age terms needs sufficient information to distinguish their effects. The data must include relevant combinations of birth gestational ages and postnatal follow-up, and the proposed terms must have interpretable roles. The choice should be guided by the developmental mechanism, study design, and predictive performance rather than by selecting whichever coding gives the most convenient parameter.

Useful checks include verifying the source and units of every age field, recomputing PMA as GA plus PNA, and checking that the model uses the intended age at each concentration record. Compare observed PK patterns or individual parameter estimates with predictions across both PMA and PNA, and examine whether residual trends remain by GA at birth or prematurity status. If patients with similar PMA show systematic discrepancies associated with time since birth, a PMA-only model may be insufficient. Conversely, a trend by PNA alone does not establish that PNA is the correct developmental clock; it may arise because PNA and PMA differ systematically across birth groups.

The main limitations are practical as well as biological. GA may be recorded imprecisely, age variables may be rounded, and some studies may contain too few infants across the relevant gestational-age range to assess alternative age codings. A maturation function also simplifies potentially distinct organ processes into a compact description. Its fitted parameters summarize the assumed model and available data; they should not be interpreted as direct measurements of a single biological event.

Common Mistake

Common mistake: “Both infants are four weeks old, so they should have the same maturation.” This reasoning silently treats PNA as the developmental age and ignores GA at birth. If the scientific model is indexed by PMA, calculate \( \mathrm{PMA}=\mathrm{GA}+\mathrm{PNA} \) for each record and use that age consistently. A different age scale may be justified, but it must be stated, supported, and evaluated rather than assumed.

Key takeaway: Age coding is part of the developmental PK model. Record GA and PNA distinctly, derive PMA transparently when it is the selected age axis, and interpret maturation parameters only in relation to that axis and its assumptions.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

An infant is born at 30 weeks' gestation and observed 6 weeks later. What are the infant's PMA and corrected age?
Under a PMA-only maturation model, should two infants at PMA 36 weeks receive different maturation fractions solely because their GA and PNA differ? Why?
With \(TM_{50}=48\) weeks, \(\gamma=2\), and PMA \(a=48\) weeks, calculate the maturation fraction.
Why can an observation at age zero not identify \(TM_{50}\) from the sigmoid maturation function?
A model fit using PNA gives different apparent maturation parameters for infants with different GA at birth. Give one reason this could happen and one diagnostic check.

Solutions. (1) PMA is \(30+6=36\) weeks. Corrected age is \(6-(40-30)=-4\) weeks, or equivalently \(36-40=-4\) weeks. (2) No. The PMA-only model assigns the same \(a\) to both, so its equation gives the same fraction; this is a model implication, not proof of identical physiology. (3) \(M(48)=48^2/(48^2+48^2)=2304/4608=0.5000\). (4) At \(a=0\), \(M(0)=0\) for every positive \(TM_{50}\), so the observation contains no information to distinguish its values; the nonzero-age inversion also cannot be used because it divides by \(M=0\). (5) PNA offsets the age origin at birth, whereas a PMA-based developmental curve includes GA at birth. Check age records and compare residual or prediction patterns by PMA, PNA, and GA at birth to assess whether the selected age coding captures the data adequately.