Skip to contents

Compute propensity and prognostic scores (treatment- or censoring-based)

ComputeScores.Rd

Computes (i) a propensity score and (ii) prognostic score(s) for use in matching/weighting pipelines. Supports SuperLearner-based fitting or glm/glmnet-based alternatives. Can compute treatment PS + treatment-specific prognostic scores (pg0/pg1), or censoring PS/PG (pscens/pgcens) when censmod=TRUE.

Usage

ComputeScores(
  data,
  id,
  Y,
  event,
  X,
  A,
  Xtrt = NULL,
  doublepg = TRUE,
  outer_CV = 5,
  inner_CV = NULL,
  stratifyCV = FALSE,
  cores = 5,
  tau = NULL,
  sl.seed = 100,
  A.SL.library = c("SL.mean", "SL.glm", "SL.glmnet", "SL.ranger", "SL.xgboost"),
  Y.SL.library = c("LIB_COXen", "LIB_AFTggamma"),
  A.method = "method.AUC",
  Y.method = "auc",
  param.tune = NULL,
  ngrid = 2000,
  param.weights.fix = NULL,
  param.weights.init = NULL,
  optim.method = "Nelder-Mead",
  penalty = NULL,
  pgcens = FALSE,
  pscens = TRUE,
  censmod = TRUE,
  maxit = 1000,
  model.pg = "cox",
  standardize = FALSE,
  superLearn = TRUE,
  pslink = "logit",
  pglink = "lognormal"
)

Arguments

data

data.frame with all variables.

id

Character scalar. Subject ID column name.

Y

Character scalar. Survival time column name.

event

Character scalar. Event indicator column name (1=event, 0=censored).

X

Character vector. Covariate column names for prognostic model(s).

A

Character scalar. Treatment indicator column name (coded 0/1 or -1/1).

Xtrt

Optional character vector. Covariates for treatment model if different from X.

doublepg

Logical. If TRUE and censmod=FALSE, fit separate prognostic models by treatment arm.

outer_CV

Integer. Outer CV folds.

inner_CV

Optional integer. Inner CV folds for nested CV (SuperLearner).

stratifyCV

Logical. Whether to stratify CV folds.

cores

Integer. Requested cores for parallel fit (multicore only on non-Windows).

tau

Optional numeric. Truncation horizon for mean survival.

sl.seed

Integer. Seed for SuperLearner.

A.SL.library

Character vector. SL learners for propensity/censoring models.

Y.SL.library

Character vector. Learners for survivalSL prognostic models.

A.method

Character. CV risk method for propensity SL.

Y.method

Character. Metric for survivalSL.

param.tune, ngrid, param.weights.fix, param.weights.init, optim.method, penalty, maxit

Control survivalSL fitting and prediction.

pgcens, pscens, censmod

Logical flags controlling censoring scores.

model.pg

Character. "cox" or "aft" for non-SL prognostic modeling.

standardize

Logical. Standardize covariates for glmnet.

superLearn

Logical. Use SuperLearner/survivalSL branches if TRUE.

Character. "logit" or "probit".

Character. AFT distribution for flexsurvreg.

Value

A data.frame with stable columns:

  • id: ID values

  • ps: treatment propensity score (if computed)

  • pg0, pg1: treatment-specific prognostic scores (if computed)

  • pscens: censoring propensity score (if computed)

  • pgcens: censoring prognostic score (if computed)