Worked Case: Lactation Exposure and Sensitivity Analysis

From Maternal Concentration to Infant Exposure

A lactation PBPK prediction links a maternal concentration profile to the drug amount an infant consumes in milk, and then—if needed—to an estimate of the amount reaching the infant’s systemic circulation. The calculation is compact, but its interpretation depends on careful choices: which maternal concentration represents the feeding period, whether the milk-to-plasma ratio is applicable, how much milk the infant actually consumes, and whether the infant absorbs the drug.

The worked case below uses hypothetical values. Its purpose is to make the calculation and sensitivity analysis transparent, not to assess the safety of a particular medicine. Infant dose estimates are exposure predictions; by themselves, they do not establish a clinical effect or a safety threshold.

Concentration multiplied by volume gives amount: for example, mg/L multiplied by L/day gives mg/day. Dividing by infant weight gives mg/kg/day. Bioavailability is a fraction from 0 to 1, so applying it changes the estimated systemic amount but not the amount swallowed. If milk concentration is measured directly, use that concentration rather than multiplying it by \(K_{m:p}\) again.

When concentrations vary substantially during the day, a single average is only defensible if it represents the milk consumed. More generally, the ingested amount is the integral of milk concentration multiplied by the rate of milk consumption over time. A simple daily-volume calculation approximates this integral by using a consumption-weighted average milk concentration. A time average that gives equal weight to periods when the infant consumes little and periods when the infant consumes much can misrepresent the amount.

Worked Case: A Baseline Prediction

What the Inputs Imply About Sensitivity

For this simplified model, the weight-normalized systemically available daily dose is

This expression assumes that all inputs refer to compatible periods and concentration bases, that the milk-to-plasma ratio can be treated as constant, and that bioavailability is adequately represented by a single fraction. It also assumes the predicted amount is proportional to each input; it does not represent saturation, time-dependent changes in partitioning, or a feeding-specific absorption process.

A useful local sensitivity measure is elasticity: the approximate proportional change in an output for a proportional change in an input near a specified baseline. For a differentiable output \(Y\) and input \(x\), its elasticity is \((\partial Y/\partial x)(x/Y)\). Here the output is a product and ratio, so the elasticities follow directly by differentiation:

Thus, locally—and in this particular product model, exactly—a 10% increase in \(K_{m:p}\), maternal concentration, milk volume, or \(F_i\) increases the predicted weight-normalized systemic dose by 10%, if all other inputs remain fixed. A 10% increase in infant weight reduces that normalized dose by about 9.1%, because the exact multiplier is \(1/1.10\), not \(0.90\). The absolute systemic amount \(D_{\mathrm{sys},24}\) does not depend on infant weight; the weight-normalized value does.

Joint Scenarios and Systemic Exposure

One-at-a-time analysis isolates the effect of each input but does not describe what happens when several inputs change together. A joint scenario evaluates a specified combination. It is useful as a transparent stress test, provided it is not mislabelled as a likely outcome or an uncertainty interval. If inputs are correlated, combinations such as “all inputs at their low values” may be implausible; correlations should inform scenario construction or a probabilistic analysis.

Under linear infant pharmacokinetics, with constant clearance \(CL_i\), the steady-state AUC over a 24-hour dosing interval is the systemically available amount per interval divided by clearance. This relationship assumes clearance is stable over the interval, the daily regimen has reached periodic steady state, and the exposure is in a linear range. It predicts AUC, not the maximum concentration or the timing of a peak.

Choosing and Checking a Sensitivity Analysis

A one-at-a-time analysis is easy to audit and helps identify which individual inputs deserve better measurement. It can be misleading when the model is nonlinear, when inputs interact, or when uncertainty is joint. A scenario analysis makes combinations explicit but depends on the scenarios selected. A probabilistic analysis instead assigns distributions to uncertain inputs, samples combinations, and propagates them through the model to produce an output distribution. That distribution is conditional on the model structure, input distributions, and dependence assumptions; it is not automatically a confidence interval or a guarantee of coverage.

For example, if milk partitioning and maternal concentrations are estimated from the same data, treating their uncertainties as independent could produce combinations that the data do not support. A probabilistic analysis should represent relevant dependence, and input distributions should respect physical limits such as \(F_i\) lying between 0 and 1 and positive concentrations and volumes. In any approach, report the baseline, input values or ranges, what was varied, and whether the result concerns ingested dose, systemic dose, or AUC.

The algebra also provides a quick diagnostic. In this model, doubling \(K_{m:p}\) while fixing everything else must double the predicted milk concentration and both dose estimates. If a calculation does not do so, check the implementation, unit conversion, and which output is being reported. A mismatch may instead indicate that the actual model includes nonlinear or time-varying processes; in that case, use sensitivity calculations on the implemented model rather than assuming the simple elasticity result still applies.

Assumptions, Failure Modes, and Limits

• Partitioning basis: \(K_{m:p}\) and maternal concentration must use compatible concentration definitions and units. A ratio based on total plasma concentration cannot be applied without justification to an unbound concentration.
• Representativeness over feeding: a single ratio and average concentration may not represent milk consumed throughout the day if concentrations vary by time, milk composition, or feeding pattern.
• Consumed volume: use the volume consumed by the infant, not automatically the volume produced. If the infant receives some feeds from another source, only the relevant milk volume belongs in this calculation.
• Bioavailability: \(F_i\) is often uncertain in infants. A value based on another age group or formulation may not be appropriate. Do not apply \(F_i\) to ingested dose when reporting the amount swallowed.
• Infant disposition: converting dose to AUC requires a suitable infant clearance and linear pharmacokinetics. A dose calculation alone cannot establish infant plasma concentrations, peak exposure, or clinical outcome.
• Precision: the arithmetic may yield several decimal places, but the prediction cannot be more reliable than its least certain inputs. Report sensible significant figures and retain units at each step.

A common mistake is to call the one-at-a-time high value an “upper confidence limit.” It is not: it is a conditional model output obtained by setting one input to a selected high value while holding others fixed. A second mistake is to multiply the infant dose by \(F_i\) and then describe that result as the amount consumed. The correction is to label the outputs separately: ingested amount, systemically available amount, and—only with clearance and the stated PK assumptions—AUC.