Population PK and dosing-interval comparison
Published population PK models are reconstructed, and summary or individual concentration data are fitted, to compare exposure when the dose, the interval or the step-up schedule changes: the bispecific T-cell engagers mosunetuzumab and tarlatamab, and nimotuzumab. Each note states what public data can and cannot support.
Reimplementing a published mosunetuzumab population PK model: step-up dosing exposure and the limits of extrapolating to receptor occupancy and CRS
Mosunetuzumab PK recalculated from the published paper and its supplement. A paired analysis of 2,000 virtual patients examines the effects of dose and interval and how far exposure can be translated to CD20 occupancy and CRS. Figures, Python code and CSV files included.
- The published PopPK model (two compartments with time-varying CL) was reimplemented in Python, recalculating Cycle 1, 42-day and Cycle 4 AUC within about 1% of published values (Cycle 1 Cmax −4.0%); this confirms the implementation, not external validity.
- In the same 2,000 virtual patients, raising the second dose from 2 to 4 mg increased the concentration before the first 60 mg dose by 77.9%, and lengthening step-up intervals from 7 to 14 days lowered 42-day AUC by 30.8%; changing only the first or second dose moved 42-day AUC by less than 2%.
- The published RO–CRS regression takes the maximum RO over 42 days and outputs about 6.7% even when RO is near 0; it cannot predict CRS incidence or individual risk for modified regimens.
How complex does a population PK model need to be for exposure comparison? Measured nimotuzumab PK
Three population PK structures estimated from individual PK data, with internal validation, convergence and exposure differences across dosing conditions used to examine the limits of model selection.
- Linear one-compartment, linear two-compartment and nonlinear-elimination PopPK models were fitted to public data (12 subjects, 321 concentrations). AIC fell with complexity (790.42, 751.25, 747.47), but leave-one-subject-out RMSE improved by only about 4.2% and 3.3%, short of the prespecified 10%.
- The 400 to 200 mg AUC ratio was 2 in the linear models and about 2.35 in the nonlinear model, a difference above 10%: the models ranked dosing conditions alike, but absolute exposure and dose ratios depended on the model.
- Only 4.5–22% of bootstrap runs passed the convergence screen, so no 95% CIs were reported. Adopting the nonlinear model and declaring the linear model sufficient are both deferred; original infusion and sampling records and independent validation data are needed next.
How far can public summary PK data evaluate a change in the tarlatamab dosing interval?
A two-compartment model estimated from published aggregate PK values, comparing Q2W, Q3W and Q4W exposure in light of identifiability and uncertainty.
- A two-compartment model fitted to six Cycle 2 PK summary values from the label explained them with a maximum error of 10.0%, but profiles showed CL, Q and Vp were poorly identified, and the terminal half-life (about 19.0 days) did not match the NCA median of 6.21 days.
- Extending the interval at the same 10 mg gives average-concentration ratios of 2/3 for Q3W and 1/2 for Q4W (fixed by mass balance under linear PK); the Q4W trough ratio was 0.371 with a wide conditional range of 0.171–0.498.
- The efficacy Emax relationship is based on first-cycle average concentration and cannot be reused as a threshold for maintenance troughs, and the per-dose fall in CRS cannot be explained by current exposure alone.
Can matching average exposure alone set the tarlatamab dosing interval?
Between-subject variability simulated from a published population PK model, with efficacy exposure–response, period-specific CRS and maintenance exposure examined separately. Matching average exposure does not preserve peaks and troughs, and the analysis shows the limits of deciding from public information.
- Recalculating the published PopPK model (FDA Tables 66–67) in 5,000 virtual subjects matched initial-dose exposure to the label means, but the steady-state trough was about 16–41% higher, so population exposure is not fully reproduced.
- Matching dose intensity to 10 mg Q2W does not preserve the profile: 20 mg Q4W kept average exposure but gave a median peak of about 1.78× and a median trough of about 0.60×.
- Efficacy saturation (Cycle 1 Cavg), CRS by period and low maintenance concentrations need separate evidence; matching average exposure alone cannot establish equivalence of maintenance regimens.
From public T-cell engager evidence to the next experiment
Public PopPK and efficacy reconstruction, Bayesian CRS sensitivity, and hypothetical measurement value and sampling design.
- Using the published tarlatamab PopPK and Emax models in 5,000 virtual subjects, initial exposure matched label summaries, but steady-state troughs did not fully agree.
- Bayesian models of three aggregate CRS windows fitted the data, but the exposure-history timescale depended on the prior, so no optimal step-up among 14 schedules could be chosen.
- Matching dose intensity preserved average exposure but changed peaks and troughs, and early-only sampling did not stabilize CL; the framework specifies which measurements could change a decision, not a validated regimen.