Developmental PK-PD Models of Pediatric Response

Why Pharmacodynamic Maturation Matters

A pediatric dose can produce a concentration similar to that in an older patient without producing the same response. One explanation is that pharmacodynamic sensitivity differs with age: a given concentration may have a smaller or larger effect. Another is that the response system turns over at a different rate, so response takes longer to change even when its eventual level is similar. A model that carries age effects only in pharmacokinetics can miss either pattern.

The goal is to represent how age changes the concentration-response relationship or the time course connecting concentration to response. Concentrations may be supplied by an established pharmacokinetic model; the question here is how to describe age-dependent pharmacodynamics without interpreting a concentration difference as a sensitivity difference.

Definition: Developmental pharmacodynamic maturation is an age-related change in a parameter governing the effect of drug concentration on response, or in the dynamics of the response system. An age-dependent potency parameter changes the concentration-response relationship; an age-dependent turnover parameter changes how quickly response approaches its drug-influenced level.

A useful model should distinguish these mechanisms, use age in a biologically and statistically defensible way, and match the sampling times and concentration range in the data. Its interpretation also depends on assumptions: concentrations must be measured or predicted adequately, the response must be defined consistently across ages, and the chosen functional form must be plausible over the ages being modeled.

Age-Dependent Sensitivity in a Direct-Response Model

For a response that can be related to current concentration without an explicit delay, a simple model is an Emax relationship with age-dependent EC50. Here \(C\) is concentration, \(R_0\) is the response at zero concentration, and \(\Delta R_{\max}\) is the maximum drug-related change from baseline. For a stimulatory effect, one form is:

R(C,a)=R_0+\Delta R_{\max}\frac{C}{EC_{50}(a)+C}

The age \(a\) is expressed in years. The parameter \(EC_{50}(a)\) is the concentration at which the modeled drug-related change is half of \(\Delta R_{\max}\). A smaller EC50 means greater modeled sensitivity. For an inhibitory response, the effect term can instead be subtracted from baseline. The direction and scale of the response must be stated in context.

One parsimonious way to describe maturation is to let a maturity function rise from zero in early life toward one at older ages:

M(a)=\frac{a^h}{T_{50}^{h}+a^h}, \qquad EC_{50}(a)=EC_{50,\mathrm{adult}}\left[1+\theta\{1-M(a)\}\right]

Here \(T_{50}\) is the age at which \(M(a)=0.5\), \(h\) controls the steepness of the transition, and \(\theta\) is a nonnegative parameter for the specific hypothesis that immature patients have a higher EC50, hence lower sensitivity. As \(M(a)\) approaches one, EC50 approaches its adult value. This is a candidate model, not a universal biological law; the direction and shape of maturation need evidence and should not be imposed merely because the parameter is called developmental.

Worked Example: Calculating an Age-Dependent Concentration Response

Suppose a hypothetical response is measured in points and follows the stimulatory model above. Let \(R_0=10\) points, \(\Delta R_{\max}=40\) points, \(EC_{50,\mathrm{adult}}=2.0\) mg/L, \(T_{50}=2.0\) years, \(h=2\), and \(\theta=1.6\). Calculate the predicted response at \(C=3.0\) mg/L for ages 1 and 8 years.

At age 1 year, \(M(1)=1^2/(2^2+1^2)=1/5=0.200\). Thus \(EC_{50}(1)=2.0[1+1.6(1-0.200)]=4.56\) mg/L. The predicted response is \(10+40(3.0)/(4.56+3.0)=10+120/7.56=25.9\) points.

At age 8 years, \(M(8)=8^2/(2^2+8^2)=64/68=0.941\). Thus \(EC_{50}(8)=2.0[1+1.6(1-0.941)]=2.19\) mg/L. The predicted response is \(10+40(3.0)/(2.19+3.0)=10+120/5.19=33.1\) points.

Under this hypothetical model, the same concentration predicts a larger response at age 8 than at age 1 because the younger age has the higher EC50. These are model predictions, not evidence that this pattern occurs for any particular drug.

Age-Dependent Turnover and Response Timing

Some responses do not follow concentration immediately. In a turnover model, drug changes the rate at which a response is produced, while the response is removed at a separate rate. For an inhibitory drug effect on production, a simple model is:

\frac{dR}{dt}=k_{\mathrm{in}}(a)\left[1-I_{\max}\frac{C}{IC_{50}+C}\right]-k_{\mathrm{out}}(a)R

Here \(k_{\mathrm{in}}(a)\) is the zero-concentration production rate, \(k_{\mathrm{out}}(a)\) is the first-order loss rate, \(I_{\max}\) is the maximum fractional inhibition, and \(IC_{50}\) is the concentration producing half of that maximum inhibition. With constant age and no drug, the baseline is \(R_0(a)=k_{\mathrm{in}}(a)/k_{\mathrm{out}}(a)\). Age may affect either rate, or both. If baseline response changes with age, that also needs to be represented or otherwise accounted for; changing turnover rates does not automatically preserve the same baseline.

Derivation: If concentration is held constant, the production term is a constant \(P=k_{\mathrm{in}}(a)[1-I_{\max}C/(IC_{50}+C)]\). The differential equation becomes \(dR/dt=P-k_{\mathrm{out}}(a)R\). Its steady state is \(R_{\mathrm{ss}}=P/k_{\mathrm{out}}(a)\). Starting from \(R(0)\), subtracting \(R_{\mathrm{ss}}\) gives \(d(R-R_{\mathrm{ss}})/dt=-k_{\mathrm{out}}(a)(R-R_{\mathrm{ss}})\); integrating yields \(R(t)=R_{\mathrm{ss}}+[R(0)-R_{\mathrm{ss}}]e^{-k_{\mathrm{out}}(a)t}\). Thus \(k_{\mathrm{out}}\) governs the speed of approach, while production and removal together determine the steady state.

Worked Example: Comparing Response Timing Across Ages

Consider a hypothetical inhibitory drug at a constant concentration of 2.0 mg/L. Let \(I_{\max}=0.80\), \(IC_{50}=2.0\) mg/L, and the baseline response be 100 units at both ages. Suppose \(k_{\mathrm{out}}=0.20\) per day at age 1 year and \(0.50\) per day at age 8 years. For each age, set \(k_{\mathrm{in}}=100k_{\mathrm{out}}\), so the zero-concentration baseline is 100 units. Calculate the steady-state response and response after 2 days, assuming the drug concentration steps to 2.0 mg/L at time zero.

The fractional inhibition is \(0.80(2.0)/(2.0+2.0)=0.40\), so production under drug is \(0.60k_{\mathrm{in}}\). The steady-state response is therefore \(0.60k_{\mathrm{in}}/k_{\mathrm{out}}=60\) units at either age.

At age 1 year, \(R(0)=100\), so \(R(2)=60+(100-60)e^{-0.20(2)}=60+40e^{-0.4}=86.8\) units. At age 8 years, \(R(2)=60+40e^{-0.50(2)}=60+40e^{-1}=74.7\) units.

The predicted response at 2 days is lower at age 8 because the response turns over faster in this example. The steady states are identical. The difference is in response timing, not in equilibrium sensitivity to concentration.

What the Data Can Distinguish

An age-dependent potency model predicts age differences in the concentration-response curve. An age-dependent turnover model predicts age differences in the response trajectory, which can persist even when the steady-state concentration-response relationship is shared. These explanations cannot generally be distinguished from one response measurement at one concentration and one time. Repeated response measurements, observations over a useful concentration range, and coverage across ages provide more information.

Worked Example: Estimating Potency at Two Ages

Suppose a hypothetical study measures the drug-related increase above baseline, \(\Delta R\), at two concentrations. Assume for this calculation that the Emax is known to be 40 response points and that the Hill coefficient is one. At age 1 year, the observed increases are 13.33 points at 2.0 mg/L and 24.0 points at 6.0 mg/L. At age 8 years, they are 20.0 points and 30.0 points at those concentrations. Calculate the implied EC50 at each age from each concentration.

Rearranging \(\Delta R=40C/(EC_{50}+C)\) gives \(EC_{50}=C(40/\Delta R-1)\). At age 1, the 2.0 mg/L observation gives \(2.0(40/13.33-1)\), approximately 4.00 mg/L; the 6.0 mg/L observation gives \(6.0(40/24.0-1)=4.00\) mg/L. At age 8, the two calculations are \(2.0(40/20.0-1)=2.00\) mg/L and \(6.0(40/30.0-1)=2.00\) mg/L.

The observations are internally consistent with EC50 values of 4.00 mg/L and 2.00 mg/L, respectively, under the stated model assumptions. This suggests higher modeled sensitivity at age 8 in this hypothetical data set. In practice, measurement error and uncertainty in Emax must also be considered; if Emax is not known, a single concentration does not separately determine Emax and EC50.

Assumptions, Diagnostics, and Model Choices

Before interpreting developmental effects, check whether the model is supported by the design and data. The age range must be broad enough to estimate a maturation pattern, and the data need concentration and response information that can distinguish age effects from other sources of variation. The concentration scale and units must be consistent across participants. If concentrations are model-predicted, uncertainty in the pharmacokinetic model can carry into the pharmacodynamic estimates.

Useful diagnostics include plotting observed and predicted response against concentration separately by age range, and plotting residuals against age and time since dosing. Age-stratified predictive checks can reveal systematic underprediction or overprediction in particular groups. For turnover models, inspect the time course: a model that fits late observations but misses early response changes may have an incorrect turnover rate or an omitted delay. For direct-response models, systematic curvature across concentration suggests that the assumed Emax form or Hill coefficient may be inadequate.

A practical comparison is between an age-invariant model, an age-dependent sensitivity model, and an age-dependent turnover model. Age bands can be useful for summaries, but they discard the ordering and within-band variation of continuous age and can create artificial jumps at band boundaries. A continuous maturation function avoids those jumps but imposes a chosen shape. The richer model is not automatically preferable: use it only when the data identify its parameters and its predictions improve relevant diagnostics or decisions.

Confounding concentration with sensitivity: different concentrations across ages can create apparent age-related response differences. Compare responses at overlapping concentrations or use a joint model that represents concentration adequately.
Confounding potency with maximum effect: observations concentrated in a narrow range may not identify both EC50 and Emax. Include information spanning the curve, or acknowledge which parameter is fixed or weakly estimated.
Confounding turnover with potency: a response observed only at one time after dosing may reflect delayed turnover rather than a different concentration-response curve. Collect observations at multiple times.
Overinterpreting the maturity function: a fitted midpoint or steepness is conditional on the selected function and sampled ages. Do not extrapolate it far beyond the data without justification.

The model remains a simplification. Age is a proxy for biological development, and children of the same chronological age may differ in relevant physiology. A single parameter may not capture multiple developmental mechanisms. The estimated relationship also describes the studied response and conditions; it does not by itself establish that a dose adjustment improves clinical outcomes.

Common Mistake

A common error is to see a smaller response in younger children and conclude that they are less pharmacodynamically sensitive. That conclusion does not follow if their concentrations were lower, if response turnover is slower, or if the measurements were taken at different times after dosing. The correction is to evaluate concentration and response jointly, specify whether the hypothesis concerns potency, maximum effect, baseline, or turnover, and use observations that can distinguish those components. A fitted age coefficient is evidence for the model's conditional relationship, not proof of a biological mechanism.

Check Your Understanding

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

In an Emax model, what does a lower EC50 imply when Emax and the response direction are unchanged?
A response has \(k_{\mathrm{in}}=30\) units per day and \(k_{\mathrm{out}}=0.30\) per day. What is its zero-concentration baseline?
In a turnover model with \(R_{\mathrm{ss}}=50\) units, \(R(0)=90\) units, and \(k_{\mathrm{out}}=0.25\) per day, what is the predicted response after 2 days?
Why can a single response measurement at one concentration and one time be insufficient to distinguish lower sensitivity from slower turnover?

Solutions. (1) A lower EC50 means that half of the same maximum drug-related effect is reached at a lower concentration, so modeled sensitivity is greater. (2) The baseline is \(k_{\mathrm{in}}/k_{\mathrm{out}}=30/0.30=100\) units. (3) \(R(2)=50+(90-50)e^{-0.25(2)}=50+40e^{-0.5}=74.3\) units. (4) Both mechanisms can produce a smaller observed response at that particular observation time: lower sensitivity reduces the concentration-driven effect, while slower turnover delays the response. Measurements over concentrations and times are needed to distinguish their predictions.