Workshop on Selecting a Pediatric Dose Regimen

From Candidate Doses to a Decision

A pediatric regimen is not selected by finding one dose that gives a desirable predicted exposure for an average child. The recommendation must also be feasible to administer across age and weight, meet the intended exposure objective, and retain an acceptable safety margin when predictions are uncertain. These requirements can point in different directions: a higher dose may improve target attainment while increasing the chance of crossing a safety limit, and a mathematically attractive dose may not be deliverable with the available formulation.

The earlier material on pediatric exposure targets, dose selection across pediatric age and weight bands, formulation and administration effects, and pediatric safety exposure-response provides the relevant component concepts. This workshop combines them in one hypothetical case. The numerical target and safety limit below are stipulated for this case; they are not general clinical thresholds.

Case, Assumptions, and Candidate Regimens

Suppose a fictional oral medicine is given once daily at steady state. The intended exposure measure is the 24-hour area under the concentration-time curve, \(AUC_{24}\), in mg·h/L. For this exercise, the desired exposure range is 40–60 mg·h/L and the prespecified safety ceiling is 80 mg·h/L. The ceiling is an upper constraint, not a target to approach.

The pharmacokinetic model predicts exposure as proportional to daily dose over the range considered. For each representative age-weight profile, the model provides a coefficient \(k\), so that the median predicted exposure is \(kD\), where \(D\) is the daily dose in mg. The coefficient reflects the model's age- and weight-specific prediction; it is not assumed to be the same for every child.

The hypothetical representative profiles and coefficients are below. These are example profiles, not claims that one weight represents every child in an age band.

The available formulation is a 10 mg granule sachet, administered as whole sachets, with a maximum of 16 sachets per dose. Thus the dose must be a multiple of 10 mg from 10 mg through 160 mg. Assume the sachets can be administered as intended in each profile; real administration acceptability would need to be checked separately. We compare three implementable regimens:

• Regimen A: approximately 5 mg/kg, rounded to available 10 mg sachets: 40, 80, and 160 mg daily.
• Regimen B: age-band doses of 50, 80, and 150 mg daily.
• Regimen C: lower age-band doses of 30, 60, and 120 mg daily.

All three obey the stated sachet-size and maximum-dose constraints. Their predicted median exposures follow by multiplying each dose by the matching coefficient.

Why the Median Is Not the Whole Decision

To represent uncertainty in this case, suppose the individual predicted exposure at each profile is lognormally distributed around the model-predicted median, with a coefficient of variation of 20%. This is a simplifying assumption for the worked calculations. It combines the uncertainty and variability represented in this hypothetical prediction; it should not be taken as a universal description of pediatric exposure.

If \(Y=\log(AUC_{24})\) is normally distributed, then the standard deviation on the log scale corresponding to coefficient of variation \(CV\) is \(\sigma_{\log}=\sqrt{\log(1+CV^2)}\). A central 90% prediction interval for an individual exposure is the median multiplied and divided by \(\exp(1.645\sigma_{\log})\). Here \(\sigma_{\log}=\sqrt{\log(1+0.20^2)}=0.198\), giving a multiplicative factor of \(\exp(1.645\times0.198)=1.385\).

This is a prediction interval for an individual exposure under the stipulated model, not a confidence interval for the population median. Its upper endpoint can be compared with the safety ceiling. A regimen whose median is below the ceiling may still have an unacceptably large upper tail.

Worked Regimen Comparison

Applying the same multiplication to all candidates produces the following median exposures. Each entry is in mg·h/L.

Regimen A has medians within the target range for all three profiles. Regimen B's youngest profile has a median above the target range, although it remains below the safety ceiling. Regimen C gives a median just below the target range in the 16 kg profile. These statements describe the medians; the predictive distribution is needed to compare individual-level target attainment and safety risk.

Deriving a Decision Rule and Applying It

A transparent recommendation starts by separating hard constraints from preferences. For this case, a simple hard safety rule is that the 90% upper prediction limit should be below 80 mg·h/L in each evaluated profile. Among regimens meeting that rule and the formulation constraints, prefer one with predicted exposure more consistently within the target range. Other decision rules could be appropriate if clinical priorities differ, but they should be specified before selecting a regimen.

The logic behind the prediction-limit check follows directly from the assumed lognormal model. If \(Z=\log(AUC_{24})\) is normally distributed with median exposure \(\widetilde{AUC}_{24}\), then its mean on the log scale is \(\log(\widetilde{AUC}_{24})\). The 95th percentile of \(Z\) is \(\log(\widetilde{AUC}_{24})+1.645\sigma_{\log}\). Exponentiating gives the 95th percentile of exposure, \(\widetilde{AUC}_{24}\exp(1.645\sigma_{\log})\), which is the upper endpoint of the central 90% prediction interval. Comparing this endpoint with 80 mg·h/L therefore checks whether the model-based 95th percentile is below the stated ceiling.

Under this rule, Regimen A's upper endpoints are \(54.0\times1.385=74.8\), \(52.0\times1.385=72.0\), and \(54.4\times1.385=75.3\) mg·h/L. All are below 80 mg·h/L. Regimen B fails the rule in the youngest profile because its upper endpoint is 93.5 mg·h/L. Regimen C passes the upper-limit rule at the three representative profiles, but its 16 kg median is below the target range, and its lower prediction limits are substantially below that range.

On these criteria, Regimen A is the preferred candidate in this hypothetical comparison. The recommendation is conditional: it depends on the linear exposure-dose relationship, the stated predictive variability, the stipulated target and ceiling, and the representativeness of the evaluated profiles. It is not a claim that a 5 mg/kg regimen is suitable for another medicine or population.

Diagnostics, Alternatives, and Limitations

Before turning this comparison into a dose recommendation, check whether its assumptions are adequate for the intended population and product. A useful diagnostic review asks whether:

• Observed or simulated exposure predictions are calibrated by age and weight, including at the edges of each proposed band.
• The exposure-dose relationship is sufficiently close to proportional across the candidate dose range; dose-dependent clearance or absorption could invalidate the simple multiplication used here.
• The predictive uncertainty reflects the relevant sources of variation and uncertainty, rather than only uncertainty in a population-average estimate.
• The proposed sachet doses can be administered reliably, and formulation-specific effects or measurement limitations do not undermine the assumed exposure predictions.
• The exposure target and safety ceiling are appropriate for the treatment objective and the relevant safety endpoint.

A poor age-stratified predictive check, systematic residual pattern across age or weight, or a prediction interval that widens markedly in a subgroup would weaken the comparison. So would strong sensitivity of the selected regimen to small changes in the exposure coefficient, the uncertainty estimate, or the safety ceiling. Such findings are reasons to revisit the model, data coverage, candidate doses, or decision rule—not to conceal the uncertainty behind a single median.

The main alternatives answer different practical questions. A single weight-based rule is simple and can vary smoothly with weight, but rounding and formulation limits can create uneven delivered doses. Age-band dosing can be straightforward to prescribe, but children of different weights within a band may have different predicted exposures. A lower fixed or banded dose may reduce upper-tail exposure, while increasing the chance of falling below the intended exposure range. A more finely divided regimen may improve predicted exposure matching, but add operational complexity and may not be deliverable with the available dosage forms. The comparison should therefore consider clinical performance and implementation together, rather than treating the most precise model prediction as automatically best.

Three representative profiles cannot establish coverage over every age and weight in a proposed regimen. The final review should assess a grid or simulated population spanning the intended range, with particular attention to band boundaries, very low weights, and any ages for which data are sparse. The simple hypothetical lognormal uncertainty model also cannot replace evaluation of the actual model's predictive performance. Where evidence is inadequate, the appropriate conclusion is a qualified recommendation or a need for more information, not an unjustified claim of certainty.

Common Pitfall