Choosing Among Linear, Exponential, and Sigmoidal Maturation

Why the Shape of the Age Function Matters

A maturation function describes how a developmental process changes with age relative to a reference level. In a clearance model, it can represent the age-related change in the fraction of a reference clearance that is attained. Choosing a function is therefore not just a mathematical convenience: it determines whether the model predicts steady progress, rapid early development followed by slowing, or a concentrated transition across a particular age range.

This tutorial compares three normalized curves. To focus on age shape, treat size and other covariates as fixed. Let \(a\) denote age on a specified, nonnegative time scale, and let \(F(a)\) be the maturation fraction. The examples use hypothetical values; they are not results from a real study.

Definition: A normalized maturation function \(F(a)\) is an age-dependent fraction between 0 and 1. In the basic forms below, \(F(0)=0\), maturation increases with age, and \(F(a)\) approaches or reaches 1 at the reference mature level. If the model uses a mature reference clearance \(CL_{\mathrm{ref}}\), the age-dependent clearance at fixed size and other covariates is \(CL(a)=CL_{\mathrm{ref}}F(a)\). This setup assumes that the chosen age scale and reference value are meaningful for the population being modeled.

The three forms differ in their implied trajectory. A linear function gives equal absolute increases per unit age until it reaches its cap. An exponential function rises quickly at first and then has progressively smaller increments. A sigmoidal Hill function can represent relatively slow early maturation, a more concentrated transition, and a later plateau. These are distinct biological patterns, not interchangeable labels for the same curve.

Three Candidate Functions

A bounded linear maturation function can be written as:

F_{\mathrm{lin}}(a)=\min\left(\frac{a}{A},1\right), \qquad A>0.

Here \(A\) is the age at which the function reaches 1. Before \(A\), each additional unit of age adds \(1/A\) to the fraction. At and after \(A\), the function is capped at 1. This is convenient when a near-linear increase over the observed age range is plausible, but it imposes a sharp change in slope at \(A\).

An exponential maturation function is:

F_{\mathrm{exp}}(a)=1-\exp(-ka), \qquad k>0.

The parameter \(k\) has units of inverse age. The function begins at zero, rises monotonically, and approaches 1 asymptotically; it does not reach 1 at any finite age. Its derivative is \(k\exp(-ka)\), so the absolute rate of increase is greatest at age zero and declines thereafter. The age at which the function reaches one-half is \(a_{50}=\ln(2)/k\).

A sigmoidal Hill maturation function is:

F_{\mathrm{Hill}}(a)=\frac{a^h}{a_{50}^{\,h}+a^h}, \qquad a\geq 0,\quad a_{50}>0,\quad h>0.

Here \(a_{50}\) is the age at which \(F_{\mathrm{Hill}}(a)=0.5\), and \(h\) is the Hill coefficient. Larger \(h\) concentrates more of the increase around \(a_{50}\), giving a sharper transition. For \(h>1\), the curve has a genuinely sigmoidal shape: its slope is initially shallow, becomes steeper, and then declines. When \(h=1\), the function is a hyperbolic rise rather than a strongly sigmoidal curve.

Interpretation: The parameters do not have identical meanings across functions. \(A\) is the linear function’s finite cap age; \(k\) sets the exponential rate; \(a_{50}\) is the Hill function’s half-maturation age; and \(h\) controls the Hill curve’s transition shape. The exponential half-maturation age is derived from \(k\), rather than being a separate parameter in the stated function.

Worked Calculations

Worked Example: Translating an Exponential Rate to Half-Maturation Age

Suppose a hypothetical exponential maturation curve has half-maturation age \(a_{50}=0.50\) years. Find \(k\), then calculate the maturation fraction at age \(0.25\) years.

At half-maturation, \(F_{\mathrm{exp}}(a_{50})=0.5\). Substituting into the function gives \(0.5=1-\exp(-ka_{50})\), so \(\exp(-ka_{50})=0.5\). Taking logarithms yields \(-ka_{50}=\ln(0.5)=-\ln(2)\), and therefore:

k=\frac{\ln(2)}{a_{50}} =\frac{0.693147}{0.50\ \text{years}} =1.386\ \text{years}^{-1}.

At age \(0.25\) years, substitute this value:

F_{\mathrm{exp}}(0.25) =1-\exp[-(1.386)(0.25)] =1-\exp(-0.3465) \approx 1-0.7072 =0.2928.

Thus, the modeled maturation fraction at \(0.25\) years is about 0.293. If the hypothetical mature reference clearance is \(10\) L/h, then at fixed size the corresponding prediction is \(10(0.2928)=2.928\) L/h. This result follows from the chosen curve; it is not evidence by itself that exponential maturation is biologically correct.

Worked Example: Calculating a Sigmoidal Maturation Fraction

Suppose a hypothetical Hill function has \(a_{50}=0.50\) years and \(h=3\). Calculate its maturation fraction at ages \(0.25\) and \(1.0\) years.

At \(0.25\) years, the age ratio is \(0.25/0.50=0.5\), whose cube is \(0.125\). Dividing numerator and denominator by \(a_{50}^{\,h}\) gives:

F_{\mathrm{Hill}}(0.25) =\frac{(0.25/0.50)^3}{1+(0.25/0.50)^3} =\frac{0.125}{1.125} =0.1111.

At \(1.0\) year, the age ratio is \(1.0/0.50=2\), whose cube is 8:

F_{\mathrm{Hill}}(1.0) =\frac{2^3}{1+2^3} =\frac{8}{9} =0.8889.

The curve predicts about 11.1% of the reference level at \(0.25\) years and 88.9% at \(1.0\) year. It predicts exactly 50% at \(0.50\) years by construction. The rise from a low fraction to a high fraction is concentrated around this age scale, unlike a constant-slope linear increase.

Worked Example: Comparing Predictions Across Age

Compare three hypothetical curves: a linear function with \(A=2.0\) years, an exponential function with \(a_{50}=0.50\) years, and the Hill function with \(a_{50}=0.50\) years and \(h=3\). Calculate their fractions at ages \(0.25\), \(0.50\), \(1.0\), and \(2.0\) years.

For the linear function, the fractions are the age divided by 2.0 until the cap: \(0.25/2.0=0.125\), \(0.50/2.0=0.250\), \(1.0/2.0=0.500\), and \(2.0/2.0=1.000\). For the exponential function, \(k=\ln(2)/0.50=1.386\) years\(^{-1}\); substitution gives fractions of approximately 0.2929, 0.5000, 0.7500, and 0.9375. The Hill fractions are calculated in the preceding example at 0.25 and 1.0 years; at 0.50 years the fraction is 0.5000, and at 2.0 years it is \(4^3/(1+4^3)=64/65=0.9846\).

Age (years)	Linear, \(A=2.0\)	Exponential, \(a_{50}=0.50\)	Hill, \(a_{50}=0.50,\ h=3\)
0.25	0.125	0.2929	0.1111
0.50	0.250	0.500	0.500
1.0	0.500	0.750	0.8889
2.0	1.000	0.9375	0.9846

The functions agree at some ages but differ substantially at others. For example, at \(0.25\) years the exponential fraction is almost three times the Hill fraction. At \(2.0\) years, the linear function has reached its cap while the other two approach, but do not equal, 1. If the reference clearance were \(10\) L/h, multiplying any table entry by \(10\) L/h would give the corresponding clearance prediction; for example, the Hill prediction at \(1.0\) year would be \(10(0.8889)=8.889\) L/h.

Matching a Curve to a Developmental Pattern

A linear function is a reasonable candidate when the data and biological rationale support an approximately constant absolute change in maturation fraction over a defined age interval. It is simple and its slope is easy to explain. Its main constraint is its cap: after \(A\), the derivative changes abruptly from \(1/A\) to zero. If observed ages extend across that point, this corner may be an unrealistic feature of the model rather than a feature of development.

An exponential function is useful when maturation is expected to progress rapidly at young ages and then slow continuously. Its diminishing increments may be plausible for a process approaching a mature level without a distinct transition interval. The curve has no finite-age plateau, however. It can be very close to 1 at later ages, but a prediction of exactly 1 requires an explicit cap or a different function.

A sigmoidal function is useful when a slow early phase, a more rapid transition, and a late flattening are plausible. Its half-maturation age gives a direct age landmark, while \(h\) controls how concentrated the transition is. A very large \(h\) implies a sharp change over a narrow age span; this can fit a sudden transition but may be implausible for gradual biological development. A small \(h\) gives a more gradual curve, and \(h=1\) does not produce the same initial shoulder as a sigmoid with \(h>1\).

All three functions as written impose a zero fraction at age zero and an upper level of 1. These may not be appropriate boundaries. If the modeled biological age clock does not have a meaningful zero at the start of development, a curve starting at \(F(0)=0\) can be inappropriate. Similarly, a nonzero immature level or an age-specific maximum different from the reference requires a model that represents those features. Extensions can add a lower or upper asymptote, but those additional parameters need data over ages that inform them.

Model Choice, Checks, and Limitations

Curve selection should begin with the age coverage and the developmental pattern the model needs to represent. Check the implied values at the youngest and oldest observed ages, the direction of change, whether the plateau is reached or merely approached, and whether the transition occurs over a biologically credible interval. Plotting each candidate over the observed age range is a useful way to reveal implications that parameter names alone can hide.

When fitting alternatives to the same data, compare them under the same observation and residual-error assumptions. Likelihood-based criteria can help assess relative fit while accounting for model complexity, but a better fit statistic does not establish that the curve is biologically correct. Sparse data at young ages may fail to distinguish a steep sigmoid from an exponential rise; data concentrated in a narrow age range may make all three curves look nearly linear there. Predictions outside the observed age range are especially dependent on the chosen functional form.

Check the age scale. Confirm that age units are consistent with the parameter units. For example, \(k\) in years\(^{-1}\) must be paired with age in years.
Check the endpoints and range. The basic functions are monotone and bounded between 0 and 1, but the linear function reaches 1 at a finite age while the others approach it asymptotically.
Check the transition implied by the parameters. A Hill coefficient that forces nearly all change into a narrow age interval deserves scrutiny, as does an exponential rate that predicts most maturation before the earliest measured age.
Check whether the data cover the shape. Without observations spanning early, transitional, and later ages, the apparent preference among curves may reflect extrapolation rather than information in the data.

Flexible alternatives, such as a piecewise curve or a spline, can represent shapes that do not match these three forms. They can also require more data and may behave unpredictably near or beyond the observed ages unless constrained. A mechanistic representation may be preferable when a specific developmental process supports it, but it requires appropriate biological structure and data. The three simple functions are useful candidates, not a complete catalog of possible age models.

Common Mistake

Common mistake: “The Hill function has \(a_{50}=0.50\) years, so it must predict half of mature clearance at \(0.50\) years, regardless of the model.” In fact, that interpretation applies to the Hill definition given here. The exponential function also predicts half at \(a_{50}=0.50\) only when its rate is set to \(k=\ln(2)/0.50\), while the linear function in the comparison predicts only \(0.25\) at that age. Parameter meanings are function-specific; equal-looking age values do not make different models equivalent.

The practical consequence is to compare predicted fractions over the relevant age range, not just compare parameter labels or fit statistics. Also distinguish a mathematical plateau from a biological claim: a function approaching 1 says how the model behaves at high age, but does not demonstrate that the reference level is actually attained in the available data.

Check Your Understanding

Answer each question, then compare with the solutions below.

For an exponential curve with \(k=0.6931\) years\(^{-1}\), calculate its half-maturation age.
A Hill curve has \(a_{50}=1.0\) year and \(h=2\). What is its fraction at age \(1.0\) year?
Which of the three basic functions reaches a fraction of exactly 1 at a finite positive age?
A fitted Hill curve has a very large \(h\). What pattern does this imply, and what should be checked before accepting it?

Solutions. (1) \(a_{50}=\ln(2)/k=0.693147/0.6931=1.000\) year, to three significant figures. (2) At \(a=a_{50}\), numerator and the second denominator term are equal, so \(F=1/(1+1)=0.500\). (3) The bounded linear function reaches exactly 1 at \(a=A\); the exponential and Hill functions approach 1 but remain below it at finite age. (4) A very large \(h\) implies a sharp, narrow transition around the half-maturation age. Check whether the observed ages resolve that transition and whether such rapid development is biologically plausible.