FDA Population Pharmacokinetics Guidance for Industry (2019)
概述 Overview
Document: Population Pharmacokinetics — Guidance for Industry
Agency: FDA (CDER / CBER)
Published: July 2019 (Final)
Source: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/population-pharmacokinetics
PDF (Final): https://www.fda.gov/media/128793/download
Population pharmacokinetics (popPK) is a cornerstone of ADC clinical pharmacology packages submitted in NDA/BLA. This guidance defines FDA’s expectations for popPK model development, reporting, and application to labeling decisions including dose adjustments for special populations.
核心要点 Key Points
1. Purpose of PopPK in ADC Development
PopPK models serve multiple functions that FDA explicitly expects to be addressed:
| Application | ADC-Specific Relevance |
|---|---|
| Quantify PK variability | ADC PK is highly variable (DAR heterogeneity, TMDD, ADA, antigen expression) |
| Identify covariates explaining variability | Weight, sHER2 (for HER2 ADCs), ADA status, hepatic function |
| Support dose selection | Flat vs. weight-based; dose adjustments |
| Enable E-R analysis | Derive individual AUC/Cmax estimates for each patient for E-R modeling |
| Inform special population dosing | Hepatic/renal impairment dose adjustment decisions |
| Support pediatric dose selection | Allometric scaling from adult popPK |
| Simulate alternative dosing regimens | Q2W vs. Q3W; loading dose strategies |
2. Model Development Requirements
Data requirements:
- Combine PK data across all clinical studies (Phase 1–3) into a single dataset when feasible
- Include both intensive PK sampling (Phase 1) and sparse sampling (Phase 2/3)
- Tag each observation with study, dose, collection time (actual, not nominal), and matrix
Structural model:
- For ADC antibody PK: typically 2-compartment with first-order elimination
- For free payload: typically 1-compartment (shorter half-life, faster distribution)
- Include TMDD if relevant (demonstrated by non-linear PK at low doses or by target expression data)
- If stochastic conjugation creates DAR heterogeneity: consider whether to model TAb or cAb; justify which analyte is modeled
Statistical model:
- Inter-individual variability (IIV): exponential error model on CL, Vd, ka
- Residual variability (RV): additive + proportional; proportional alone if concentration range is large
- NONMEM (first-choice for regulatory submissions), Monolix, Phoenix NLME all acceptable
Covariate model:
- Exploratory: plot covariate vs. empirical Bayes estimates (EBEs) of PK parameters; assess correlation
- Screening: stepwise forward addition (p<0.05) + backward elimination (p<0.001) OR full covariate model approach
- Evaluate clinical significance: does the covariate change AUC by ≥30% for any patient subgroup? If yes → consider dose adjustment in label
3. Covariate Analysis for ADCs
Standard covariates expected in all ADC popPK submissions:
| Covariate | Effect on PK Parameter | Clinical Decision Threshold |
|---|---|---|
| Body weight | CL, Vd (primary covariate) | If weight explains <20% of CL variability → flat dosing may be justified |
| Hepatic function (ALT, AST, bilirubin, albumin, Child-Pugh) | Payload CL (CYP3A4 metabolism) | ALT/AST >3× ULN: moderate HI; Bilirubin >1.5× ULN: severe HI |
| Renal function (CrCl, eGFR) | Payload CL | Generally minor; assess if payload metabolites are renally cleared |
| ADA status | ADC total CL (immune complex clearance) | ADA+ subgroup: compare exposure (AUC) to ADA− group |
| Soluble antigen (sTAA) | TMDD-driven CL of cAb | Elevated sTAA → increased CL; explore as prognostic for exposure |
| Baseline tumor burden | Target-mediated CL | Relevant for hematologic ADCs (CD33, CD19 on circulating tumor cells) |
| Albumin | TAb clearance (FcRn competition) | Low albumin (<3 g/dL): may accelerate IgG clearance |
| Race/ethnicity | Payload CL (CYP polymorphisms) | Significant for CYP2D6/CYP2C19 if payload metabolized by these |
| Sex | Minor impact on ADC PK typically | Explore; rarely clinically significant |
| Age | Renal/hepatic function-mediated; explore as covariate | Elderly subgroup analysis required |
4. Model Evaluation Requirements
FDA requires comprehensive model diagnostics in the popPK report:
Goodness-of-fit (GOF) plots (mandatory):
- Observed vs. population predicted (PRED); should be symmetric around identity line
- Observed vs. individual predicted (IPRED); tighter scatter around identity
- Conditional weighted residuals (CWRES) vs. PRED and vs. time; should be randomly distributed ±3
Visual Predictive Check (VPC) (mandatory):
- Simulate ≥1000 replicates from the final model
- Overlay simulated 5th, 50th, 95th percentiles with observed data
- For ADC: conduct stratified VPC by dose, study, analyte (TAb vs. cAb if both modeled)
- Prediction-corrected VPC (pcVPC) when multiple dose levels are combined
Bootstrap (recommended):
- Non-parametric bootstrap (≥500 replicates) to estimate uncertainty on final parameter estimates
- 95% CI from bootstrap should be reported for all structural and variance parameters
5. Reporting Requirements for NDA/BLA
Per FDA guidance, the popPK report (Module 2.7.2 and 5.3.3.5 of CTD) must include:
- Dataset description: All clinical studies contributing PK data; number of subjects, observations, BLQ (below LLOQ) handling strategy
- Software and algorithm: NONMEM version, estimation method (FOCE-I, SAEM), minimization status
- Model selection rationale: OFV comparison, AIC/BIC where applicable; scientific rationale over statistical criteria
- All diagnostic plots (GOF, VPC, ETA distributions, ETA correlation)
- Final model parameter estimates with RSE% (relative standard error); RSE >30% for fixed effects warrants justification
- Simulation outputs: Dose-AUC relationship at recommended dose; special population simulations
- Sensitivity analysis: Impact of BLQ handling, covariate bounds, alternative structural models
常见问题和挑战，具体案例和解决方案
Challenge 1: TAb vs. cAb — Which Analyte to Use for PopPK?
Problem: Both TAb and cAb are measured in clinical studies. Should the popPK model use TAb or cAb? Both? FDA expects clarity.
FDA expectation: FDA ADC ClinPharm 2024 guidance recommends modeling both TAb and cAb separately, with a deconjugation parameter linking them. However, many sponsors model only cAb (primary efficacy/safety driver) with TAb as a supporting analysis.
Solution approaches:
- Separate models: Independent TAb and cAb two-compartment models; deconjugation rate kdc estimated by fitting both analytes simultaneously
- cAb-only model: Acceptable if the goal is E-R analysis (cAb drives efficacy/safety); report TAb separately for antibody PK characterization
- Mean DAR model: Model average DAR as a function of time (DAR₀ × exp(−kdc × t)) to characterize the rate of payload loss
Recommendation: Model TAb + cAb simultaneously for an NDA/BLA with the deconjugation parameter reported. Use cAb individual AUC estimates for E-R analysis.
Challenge 2: Handling BLQ Data for Free Payload
Problem: Free payload concentrations are below LLOQ in 40–60% of PK samples (especially at late time points). BLQ handling strategy affects AUC estimation.
FDA expectation: The guidance recommends specifying the BLQ handling strategy prospectively in the analysis plan. M3 (maximum likelihood approach), LAPLACIAN-extended, or conditional weighted least squares can handle BLQ explicitly in NONMEM.
Solution:
- If ≤20% of observations are BLQ: LLOQ/2 substitution acceptable (FDA BMV 2018 convention)
- If >20% BLQ: use M3 method (likelihood-based BLQ treatment in NONMEM)
- Sensitivity analysis: compare NCA-derived AUC₀–last with popPK model-predicted AUC₀–∞ to confirm truncation does not materially alter E-R conclusions
Challenge 3: Sparse Sampling in Phase 3 — Sufficient for PopPK?
Problem: Phase 3 oncology trials typically allow only 2–3 PK samples per patient (sparse PK) due to patient burden. Is this sufficient for an informative popPK model?
FDA expectation: Sparse data from Phase 3 (large N) combined with rich data from Phase 1 is acceptable. The Phase 1 intensive PK data defines the structural model; Phase 3 sparse data refines covariate estimates in the target population.
Solution:
- Design Phase 3 sparse PK sampling to cover Cmax (end of infusion) and trough (pre-dose, next cycle) as minimum — captures AUC range without full profile
- Use a D-optimal design (or similar) to select sparse sampling time points that maximize information content given the Phase 1 structural model
- Apply the full combined dataset (Phase 1 + 2 + 3) in the final popPK model; demonstrate shrinkage of IIV estimates reduces from Phase 1 alone
相关条目 Related Entries
- FDA ADC ClinPharm 2024 — E-R analysis framework built on popPK
- ICH M10 BMV 2022 — BMV for cAb/TAb/free payload data feeding the popPK model
- 07_pkpd_modeling — Full PK/PD modeling framework including popPK approaches
- AAPS ADC White Papers — Industry consensus on ADC analyte selection for popPK