Skip to content

Theophylline theo_sd Real-World Validation

Dataset

  • Source: theo_sd from nlmixr2data (GPL >= 3)
  • Columns: ID, TIME, DV, AMT, EVID, CMT, WT

Design Realism Addressed

  1. Missing DV rows: Ignored in RMSE calculation (kept for audit, excluded from metric).

  2. Duplicate observation times: Supported via unique grid with mapping back to all observations.

  3. AMT ambiguity: Handled explicitly with deterministic rule:

  4. If AMT < 50 (typical mg/kg scale): dose_mg = AMT * WT
  5. Otherwise: dose_mg = AMT

  6. Rule is recorded in artifact metadata as dose_unit_rule

  7. WT covariate scaling scenario: Two simulation modes:

  8. Fixed: CL = 2.75 L/hr, V = 31.8 L (population typical)
  9. WT-scaled:
    • CL_i = CL_ref * (WT/70)^0.75
    • V_i = V_ref * (WT/70)^1.0

This is simulation-only validation (no fitting).

Model Parameters

Parameter Value Unit
Ka 1.59 1/hr
CL_ref 2.75 L/hr
V_ref 31.8 L
WT_ref 70.0 kg

Outputs

Per subject (12 subjects total): - subj_<id>_fixed.json - Fixed parameter artifact - subj_<id>_wt_scaled.json - WT-scaled parameter artifact

Metrics JSON with per-subject: - id, wt, dose_mg - dose_unit_rule - How AMT was interpreted - rmse_fixed - RMSE for fixed params - rmse_wt_scaled - RMSE for WT-scaled params

Running

# From repository root
julia docs/examples/real_world_validation/datasets/theophylline_theo_sd/run.jl

# Copy to expected (first time or when regenerating)
rm -f docs/examples/real_world_validation/studies/theophylline_theo_sd/expected/*.json
cp docs/examples/real_world_validation/studies/theophylline_theo_sd/output/*.json \
   docs/examples/real_world_validation/studies/theophylline_theo_sd/expected/

# Validate
julia docs/examples/real_world_validation/datasets/theophylline_theo_sd/validate.jl

Validation

  • Metrics compared with strict tolerance (1e-12)
  • All artifacts replayed and compared
  • Both fixed and WT-scaled scenarios validated