NEWS

---

pkident 0.1.0

First release. Decisions are local and are stated as such in every result.

Model definition

• pkpd_model() reads ODE models written for rxode2 (code, compiled models or model functions) and records outputs, doses (bolus_dose(), infusion_dose()), known input signals, known constants and initial conditions.
• reference_models() and reference_model() provide ten acceptance models with expected results fixed from primary literature or analytical derivation.

Structural identifiability

• structural_identifiability() decides local structural identifiability with Lie derivatives evaluated at the known initial state, checked by the rank and null space of output sensitivities solved with rxode2. A decision is made only if all random points and both methods agree; otherwise the result is unresolved. Identifiable monomial combinations (for example V/F) are reported.

Practical identifiability and design

• practical_identifiability() computes the expected Fisher information for a sampling design with additive, proportional, combined or exponential residual error (residual_error()), with doses from the model or from an event table.
• compare_designs() and experiment() compare user-supplied candidate designs: added outputs, sampling times or dose levels.
• classify_profiles() classifies likelihood profiles of nlmixr2 fits (bounded, one-sided, flat, shallow; Raue et al. 2009) using nlmixr2extra::profileFixed() with warm-started re-estimation, and sets them against the structural result.

Validation

• All ten acceptance models (and a variant) agree with their expected results at four seeds, and the two structural methods agree at every point.
• Expected RSEs agree with the spread of 1,000 repeated estimates within 0.96–1.08 in four scenarios; limitations at RSEs of 30–40% are documented.
• Profile classification is correct for all ten parameters of three synthetic nlmixr2 fits.
• Vignettes: reference models, nimotuzumab TMDD (public nimoData), PSA decline and regrowth.