pkpd-clinical-response

name	pkpd-clinical-response
description	Pharmacokinetic/pharmacodynamic modeling, exposure-response analysis, clinical response classification (RECIST/iRECIST), dose-response curves, population PK covariate modeling, PD biomarker tracking, PK-PD hysteresis, and time-to-event endpoints
version	1.1.0
tags	["pharmacokinetics","pharmacodynamics","PK/PD","exposure-response","RECIST","iRECIST","dose-response","clinical-response","pd-biomarker","hysteresis","NCA","population-pk","lifelines","survival-analysis"]

PK/PD & Clinical Response

This skill guides pharmacokinetic/pharmacodynamic data analysis and clinical response evaluation in Phase 1/2 trial-adjacent contexts.

Clinical Response Classification

RECIST 1.1 (Response Evaluation Criteria in Solid Tumors)

Reference: references/clinical-response-criteria.md — read it for full RECIST 1.1 tables, overall response derivation matrix, recist_classify_target(), best_overall_response(), and iRECIST pseudoprogression rules.

Category	Abbreviation	Definition
Complete Response	CR	Disappearance of all target lesions
Partial Response	PR	>=30% decrease in sum of diameters
Stable Disease	SD	Neither PR nor PD criteria met
Progressive Disease	PD	>=20% increase in sum + absolute increase >=5mm, or new lesion

When response data is present:

• Binary: responder (CR+PR) vs non-responder (SD+PD)
• Ordinal: CR > PR > SD > PD (use ordinal logistic regression)
• Duration: time to best response, duration of response

iRECIST (for immunotherapy)

Adds "unconfirmed" states (iUPD → must confirm with iCPD at next assessment). For IO trials, always check if iRECIST was used.

Pharmacokinetic Analysis

Non-Compartmental Analysis (NCA)

Reference: references/nca-analysis.md — read it for complete NCA functions (auc_linear_trapezoidal, auc_log_linear_trapezoidal, auc_extrapolated, terminal_half_life, derive_pk_params), concentration-time plots, and quality checks.

Always report: Cmax, Tmax, AUC(0-last), AUC(0-inf), t1/2, CL/F, Vd/F. Plot: concentration-time profile (linear and semi-log scale).

Population PK Covariates

Reference: references/population-pk.md — read it for statsmodels MixedLM syntax, random effects specification, forward_selection(), backward_elimination(), standard PK covariates table, allometric scaling, and diagnostic plots.

When patient demographic/clinical data is available alongside PK data:

• Test covariates: age, weight, sex, renal function (eGFR), hepatic function (Child-Pugh), genotype (CYP polymorphisms)
• Use mixed-effects models (statsmodels MixedLM or lme4 via rpy2)
• Forward addition / backward elimination for covariate selection

Exposure-Response Analysis

Efficacy Exposure-Response

Reference: references/dose-exposure-response.md — read it for four_param_logistic() fitting, exposure_response_binary(), safety_exposure_response() (time-to-AE by exposure quartile), and identify_therapeutic_window().

Link drug exposure (AUC, Cmax, Cmin) to efficacy outcomes. Use logistic regression for binary response vs exposure, and derive EC50 (exposure at 50% response probability) from the fitted model.

Safety Exposure-Response

Reference: references/lifelines-pkpd.md — read it for safety_exposure_response() with KM + log-rank, cox_exposure_safety() for multivariate hazard models, and at-risk table formatting.

Link exposure to adverse event probability:

• Logistic regression: P(AE grade >= 3) ~ AUC_ss
• Kaplan-Meier: time to first grade 3+ AE by exposure quartile
• Identify therapeutic window: exposure range where efficacy is maintained but safety is acceptable

Dose-Response Curves

Fit a 4-parameter logistic (Hill equation) to dose-response data. See the reference for 4PL/3PL fitting functions and plotting.

Always plot: observed data points + fitted curve + 95% CI band. Report: EC50/IC50, Hill coefficient, Emax, R-squared.

Responder vs Non-Responder Analysis with Omics

When paired PK/response + omics data is available:

1. Define groups — responder vs non-responder per RECIST or custom criteria. Document the exact definition.
2. DE analysis — standard differential expression between groups. Use the response classification as the contrast, not treatment (treatment is constant in single-arm trials).
3. Exposure-stratified analysis — split by exposure quartiles, then compare omics profiles. This separates pharmacokinetic effects (dose-dependent transcriptional changes) from pharmacodynamic effects (target-dependent biology).
4. Predictive biomarker analysis — if treatment and control arms exist, test marker x treatment interaction (see statistical-modeling skill for predictive vs prognostic testing).

Waterfall and Spider Plots

Standard clinical trial visualizations. See PD biomarker tracking reference for waterfall plot functions.

• Waterfall plot: Best % change from baseline per patient, sorted ascending, colored by response category (green = PR threshold at -30%, red = PD threshold at +20%).
• Spider plot: Per-patient tumor burden over time (individual traces), with RECIST thresholds overlaid.

Pharmacodynamic Biomarker Tracking

Reference: references/pd-biomarker-tracking.md — read it for complete PD assessment functions (compute_pd_change, classify_pd_response, plot_pd_timecourse, plot_pd_by_dose), hysteresis diagnosis, PD-exposure mixed-effects modeling, and waterfall plots.

Longitudinal PD Marker Assessment

PD biomarkers measure target modulation and downstream biological effects over time. Track them at defined timepoints relative to dosing:

Phase	Typical Timepoints	Purpose
Pre-treatment (baseline)	C1D1 pre-dose	Reference values
On-treatment early	C1D2, C1D8, C1D15	Onset of PD effect
On-treatment steady-state	C2D1, C3D1	Sustained modulation
Post-treatment washout	1w, 2w, 4w post-last-dose	Recovery kinetics

C = cycle, D = day, w = week.

PD Response Criteria

Define PD response categories based on marker change thresholds:

Category	Typical Threshold	Example
PD responder	>= 50% decrease from baseline	Phospho-protein inhibition
PD partial responder	30-50% decrease	Cytokine reduction
PD non-responder	< 30% decrease	No target modulation
PD rebound	Return to >= 80% of baseline after initial decrease	Resistance or washout

Always document: which marker, what threshold, at which timepoint, and the biological rationale for the threshold.

PK-PD Hysteresis

When plotting PD effect vs drug concentration, a hysteresis loop indicates temporal dissociation between PK and PD:

• Counter-clockwise hysteresis — PD effect lags behind concentration (common: indirect response mechanisms, signal transduction delays).
• Clockwise hysteresis — PD effect precedes concentration decline (rare: tolerance, receptor desensitization).

PD Marker-Exposure Correlation

Link PD marker changes to drug exposure using mixed-effects models to account for repeated measures. Always report whether the PD-exposure relationship is:

• Monotonic — more drug = more PD effect (expected for on-target)
• Saturating — PD effect plateaus at high exposure (target fully engaged)
• Bell-shaped — PD effect decreases at very high exposure (off-target or toxicity)

References

Detailed API references, ready-to-use functions, and domain-specific gotchas. Read the reference file itself before writing code in that area.

Reference	File	Contents
NCA Analysis	references/nca-analysis.md	AUC (linear/log-linear/extrapolated), terminal half-life, full NCA parameter derivation, concentration-time plots, quality checks
Dose & Exposure-Response	references/dose-exposure-response.md	4PL/3PL dose-response fitting, exposure-response logistic regression, safety E-R (time-to-AE by quartile), therapeutic window identification
Population PK	references/population-pk.md	statsmodels MixedLM for pop PK, random effects specification, forward/backward covariate selection, standard covariates table, allometric scaling, diagnostic plots
Clinical Response Criteria	references/clinical-response-criteria.md	RECIST 1.1 full tables (target/non-target/overall), iRECIST pseudoprogression rules, response endpoint derivation (ORR, DCR, DOR, TTR, PFS)
PD Biomarker Tracking	references/pd-biomarker-tracking.md	PD change from baseline, PD response classification, PK-PD hysteresis (plot + diagnosis), PD-exposure correlation, waterfall plots
Lifelines for PK/PD	references/lifelines-pkpd.md	KM for DOR/PFS/TTR, safety exposure-response, Cox PH for multivariate analysis, at-risk tables, competing risks guidance

Do NOT

• Fit compartmental PK models without sufficient timepoints (need >= 2x the number of parameters)
• Report PK parameters without specifying the dose, route, and whether steady-state was reached
• Use NCA terminal half-life if the terminal phase has fewer than 3 points or R-squared < 0.8
• Conflate exposure-response with dose-response — exposure accounts for PK variability, dose does not
• Call a biomarker "predictive" without a treatment x marker interaction test (this is the single most common error)
• Assume single-arm trial biomarkers are predictive — they can only be prognostic without a control arm