Consortium Tree Model Explainability and TRIPOD Reporting — consortium

Provides model-agnostic and tree-based feature attribution, direction of effect, TRIPOD Item 13b / STROBE Item 15a compliant variable importance tables, and cross-site feature importance drift auditing across federated health systems. Requires zero heavy external dependencies.

Usage

consortium_explainability(
  model = NULL,
  X,
  y = NULL,
  feature_names = NULL,
  site_labels = NULL,
  method = "auto",
  n_permutations = 5L,
  top_n = NULL,
  ...
)

Arguments

model: A fitted predictive model (e.g. from randomForest, rpart, glm, xgboost), or NULL.
X: A data frame or numeric matrix of predictors.
y: Optional numeric or factor vector of true outcomes/labels.
feature_names: Optional character vector of feature names if X is a matrix without column names.
site_labels: Optional vector of site identifiers for multi-site importance drift analysis.
method: Attribution method: "auto" (default), "gain", "permutation", or "coefficient".
n_permutations: Number of shuffle iterations for permutation importance (default 5L).
top_n: Optional integer to limit the exported importance table to the top N features.
...: Additional arguments passed to methods.

Value

A consortium_explainability S3 object with elements:

importance: Data frame of ranked features, attributions, standard errors, and directions.
tripod_summary: Publication-ready table adhering to TRIPOD Item 13b / STROBE Item 15a guidelines.
site_drift: Cross-site ranking concordance table if site_labels provided, otherwise NULL.
shap_values: Matrix of sample-level marginal attribution values.
feature_names: Character vector of feature names.
method: Attribution method used.