Assess Model Eligibility for Quantification — assess_model

Applies a set of identifiability and precision gates to determine whether a fitted model is reliable enough to be considered for selection as the calibration model used for sample quantification. The same gates are applied in both the frequentist and Bayesian frameworks, with framework-specific gates automatically skipped when the required inputs are not available.

Usage

assess_model_eligibility(
  model_name,
  parameters,
  constraints = NULL,
  pcov_profile = NULL,
  grid_x = NULL,
  pcov_threshold = 20,
  bound_tol = 1e-04,
  max_rel_se = 5,
  min_dynamic_range_log10 = 0.5,
  vcov_matrix = NULL
)

Arguments

model_name

Character. Name of the model being assessed.

parameters

Data frame. Must contain term and either estimate

std_error (frequentist) or mean + sd (Bayesian).

constraints

Named list with $lower and $upper named numeric vectors for the free parameters, or NULL (Bayesian / fixed-a).

pcov_profile

Numeric vector of pcov values (%) on the prediction grid, or NULL if the grid has not been computed for this model.

grid_x

Numeric vector of log10_concentration values matching pcov_profile, or NULL.

pcov_threshold

Numeric. The pcov (%) threshold used to define the LLOQ and ULOQ. Default 20.

bound_tol

Numeric. A parameter estimate within this absolute distance of a constraint bound is treated as "at bound". Default 1e-4.

max_rel_se

Numeric. Maximum permitted relative SE (SE / |estimate|) for any parameter. Default 5.0.

min_dynamic_range_log10

Numeric. Minimum required LLOQ-to-ULOQ span in log10 concentration units. Default 0.5 (~3-fold).

vcov_matrix

Numeric matrix. The parameter covariance matrix from vcov(fit), or NULL. Used for the condition-number gate.

Value

A named list:

model_name: Character.
eligible: Logical. TRUE if all applicable gates pass.
gates: Data frame with columns gate, passed, detail.
dynamic_range_log10: Numeric or NA.
lloq: Numeric or NA (log10 scale).
uloq: Numeric or NA (log10 scale).