Robust Multivariable Odds Ratios with L2 / Ridge Inversion — consortium_multivariable_risk

Fits multivariable logistic regression using L2 ridge penalization and QR orthogonalization to eliminate collinear columns and prevent Hessian singularity on high-dimensional clinical feature matrices.

Usage

consortium_multivariable_risk_factors(
  X,
  y,
  binary_cols = NULL,
  regularize = TRUE,
  alpha = 1,
  domain_mapping = NULL,
  standardize_continuous = TRUE,
  drop_collinear = TRUE
)

Arguments

X: Data frame or numeric matrix of predictors.
y: Binary numeric vector of outcome indicators (0 or 1).
binary_cols: Optional character vector of binary column names. If NULL, auto-detected.
regularize: Logical; whether to apply L2 ridge regularization (default TRUE).
alpha: Numeric; L2 regularization penalty parameter lambda (default 1.0).
domain_mapping: Optional named character vector mapping feature names to clinical domains.
standardize_continuous: Logical; whether continuous features are scaled to 1-SD units (default TRUE).
drop_collinear: Logical; whether linearly dependent columns are dropped via QR decomposition (default TRUE).

Value

An S3 object of class consortium_risk_factors with Table 5 data frame, coefficients, standard errors, and methods.