Data Selection (IGNORE / ACCEPT)

This example shows the [data_selection] block, which filters dataset rows at read time without modifying the CSV. The complete model file is examples/warfarin_data_selection.ferx.

When to use

Use [data_selection] when you want to: - Exclude observations below a lower limit of quantification (LLOQ surrogate) - Restrict the analysis to a specific study part, occasion, or compartment - Mirror a NONMEM $DATA IGNORE= or ACCEPT= statement in the model file instead of pre-processing the data

Model file

This is the contents of examples/warfarin_data_selection.ferx:

# One-compartment oral PK model (warfarin) with data-selection filtering

[parameters]
  theta TVCL(0.134, 0.001, 10.0)
  theta TVV(8.1, 0.1, 500.0)
  theta TVKA(1.0, 0.01, 50.0)

  omega ETA_CL ~ 0.07
  omega ETA_V  ~ 0.02
  omega ETA_KA ~ 0.40

  sigma PROP_ERR ~ 0.01 (sd)

[individual_parameters]
  CL = TVCL * exp(ETA_CL)
  V  = TVV  * exp(ETA_V)
  KA = TVKA * exp(ETA_KA)

[structural_model]
  pk one_cpt_oral(cl=CL, v=V, ka=KA)

[error_model]
  DV ~ proportional(PROP_ERR)

[data_selection]
  # Drop observations below a surrogate LLOQ of 1.0 mg/L.
  ignore = DV < 1.0

[fit_options]
  method   = foce
  maxiter  = 300
  gradient = fd

The [data_selection] block accepts two optional directives:

Key Effect
ignore = <expr> Drop any row for which <expr> evaluates to true
accept = <expr> Keep only rows for which <expr> evaluates to true

Only one of ignore / accept may be specified per model.

The filter expression can reference any column in the dataset (DV, TIME, OCC, CMT, covariates, …). Dose rows (EVID = 1) are never dropped regardless of the filter, preserving dose history for all subjects.

Running the fit

ferx examples/warfarin_data_selection.ferx --data data/warfarin.csv

Or via the Rust API:

let result = fit_from_files("examples/warfarin_data_selection.ferx", "data/warfarin.csv")?;
println!("Excluded observations: {:?}", result.warnings);

Interpreting output

The console summary reports how many observations were excluded:

Data: 32 subjects, 192 observations (12 excluded by data_selection)

Excluded rows do not contribute to the OFV, EBE optimisation, or sdtab output. Dose rows are always retained.

Tips

  • Prefer ignore for LLOQ exclusion — ignore = DV < lloq is the direct analogue of NONMEM IGNORE=(DV.LT.lloq).
  • Column names are case-insensitive — ignore = dv < 1.0 and ignore = DV < 1.0 are equivalent.
  • Combine with [data_selection] for study-design subsets: accept = OCC == 1 restricts the fit to the first occasion without touching the data file.
  • Dose rows are protected: the engine never excludes EVID=1 rows, so PK history is always complete even when observations are filtered.
  • LOQ-censored observations: if you use the M3 method ([fit_options] bloq = m3) alongside [data_selection], set the filter so censored rows (CENS != 0) are not excluded — the M3 likelihood uses them. Use a tighter filter or remove the ignore entirely and rely on M3 alone.