Visual Predictive Check (VPC)

Examples demonstrating VPC generation for model validation.

What is VPC?

VPC compares simulated data from the model to observed data: 1. Simulate n replicates (e.g., 500) using final model estimates 2. Compute quantiles (5th, 50th, 95th) of simulated data at each time 3. Overlay observed data 4. Check if observed quantiles fall within simulated confidence intervals

VPC Types

Type	Description	Use Case
Standard VPC	Basic VPC	Initial validation
pcVPC	Prediction-corrected	Variable dosing
Stratified	By covariate	Covariate effects
VPC with BLQ	Below LOQ handling	Assay limits

Examples

Example	Description	Directory
Standard VPC	Basic implementation	01_standard_vpc
pcVPC	Prediction-corrected	02_prediction_corrected_vpc

Usage

using NeoPKPD

# Generate VPC
vpc_result = compute_vpc(
    observed_data,
    pop_spec,
    n_simulations = 500,
    quantiles = [0.05, 0.5, 0.95],
    seed = 12345
)

# Access results
sim_quantiles = vpc_result.simulated_quantiles  # Simulated CI
obs_quantiles = vpc_result.observed_quantiles   # Observed data

Interpretation

Good VPC: - Observed median close to simulated median - 90% of observed data within 90% simulated PI - No systematic deviations

Poor VPC signals: - Median bias → Structural model misspecification - Over-prediction → IIV too large - Under-prediction → IIV too small