Skip to contents

Estimate CATE scores for binary, count and survival outcomes. Binary scores include pooled logistic regression, doubly robust two-regression estimation and boosting. Count and survival methods include both outcome-regression and doubly robust scores. Continuous outcomes are not yet implemented.

Usage

catefit(
  response,
  data,
  score.method,
  cate.model,
  ps.model,
  ps.method = "glm",
  init.model = NULL,
  initial.predictor.method = NULL,
  ipcw.model = NULL,
  ipcw.method = "breslow",
  minPS = 0.01,
  maxPS = 0.99,
  followup.time = NULL,
  tau0 = NULL,
  higher.y = TRUE,
  prop.cutoff = seq(0.5, 1, length = 6),
  surv.min = 0.025,
  xvar.smooth.score = NULL,
  xvar.smooth.init = NULL,
  tree.depth = 2,
  n.trees.rf = 1000,
  n.trees.boosting = 200,
  B = 3,
  Kfold = 5,
  error.maxNR = 0.001,
  max.iterNR = 150,
  tune = c(0.5, 2),
  seed = NULL,
  plot.gbmperf = TRUE,
  verbose = 0,
  contrast.control = list(),
  rmst.control = list(),
  ps.control = list(),
  rf.control = list(),
  ...
)

Arguments

response

A string describing the type of outcome in the data. Allowed values include "binary", "count" (see catecvcount()), "survival" (see catecvsurv()) and "continuous" (see catecvmean()).

data

A data frame containing the variables in the outcome, treatment, and censoring models (if specified); a data frame with n rows (1 row per observation).

score.method

A vector of one or multiple methods to estimate the CATE score. Allowed values are: 'boosting', 'twoReg', 'contrastReg', 'poisson' (count and survival outcomes only), 'randomForest' (survival, continuous outcomes only), negBin (count outcomes only), 'gam' (continuous outcomes only), 'gaussian' (continuous outcomes only). For binary outcomes, use 'logistic', 'twoReg', or 'boosting'.

cate.model

A formula describing the outcome model to be fitted. The outcome must appear on the left-hand side. For survival outcomes, a Surv object must be used to describe the outcome.

ps.model

A formula describing the treatment model (propensity-score model) to be fitted. The treatment must appear on the left-hand side. For binary outcomes, logical treatment values FALSE and TRUE are mapped to 0 and 1, respectively. Numeric 0/1 values or two distinct treatment labels are also accepted. If data are from a randomized controlled trial, specify ps.model = trt ~ 1 as an intercept-only model.

ps.method

A character value for the method to estimate the propensity score. Allowed values include one of: 'glm' for logistic regression with main effects only (default), or 'lasso' for a logistic regression with main effects and LASSO penalization on two-way interactions (added to the model if interactions are not specified in ps.model). Relevant only when ps.model has more than one variable.

init.model

A formula describing the initial predictor model. The outcome must appear on the left-hand side. It must be specified when score.method = contrastReg or twoReg, except for binary outcomes, which use cate.model.

initial.predictor.method

A character vector for the method used to get initial outcome predictions conditional on the covariates specified in cate.model. Only applies when score.method includes 'twoReg' or 'contrastReg'.Allowed values include one of 'randomForest' (survival outcomes only), 'boosting', 'logistic' (survival outcomes only, fast), 'poisson' (count outcomes only, fast), 'gaussian' (continuous outcomes only) and 'gam' (count and continuous outcomes only). Default is NULL, which assigns 'boosting' for count outcomes and 'randomForest' for survival outcomes.

ipcw.model

A formula describing the censoring model used for inverse probability of censoring weighting (IPCW). The left-hand side must be empty. Only applies for survival outcomes. Default is NULL, which uses a censoring model with the same covariates as the outcome model cate.model, plus the treatment.

ipcw.method

A character value for the censoring model. Only applies for survival outcomes. Allowed values are: 'breslow' (Cox regression with Breslow estimator of t he baseline survivor function), 'aft (exponential)', 'aft (weibull)', 'aft (lognormal)' or 'aft (loglogistic)' (accelerated failure time model with different distributions for y variable). Default is 'breslow'.

minPS

A numerical value (in `[0, 1]`) below which estimated propensity scores should be truncated. Default is 0.01.

maxPS

A numerical value (in `(0, 1]`) above which estimated propensity scores should be truncated. Must be strictly greater than minPS. Default is 0.99.

followup.time

A column name in data specifying the maximum follow-up time, interpreted as the potential censoring time. Only applies for survival outcomes. Default is NULL, which corresponds to unknown potential censoring time.

tau0

The truncation time for defining restricted mean time lost. Only applies for survival outcomes. Default is NULL, which corresponds to setting the truncation time as the maximum survival time in the data.

higher.y

A logical value indicating whether higher (TRUE) or lower (FALSE) values of the outcome are more desirable. Default is TRUE.

prop.cutoff

A vector of numerical values (in `(0, 1]`) specifying percentiles of the estimated log CATE scores to define nested subgroups. Each element represents the cutoff to separate observations in nested subgroups (below vs above cutoff). The length of prop.cutoff is the number of nested subgroups. An equally-spaced sequence of proportions ending with 1 is recommended. Default is seq(0.5, 1, length = 6).

surv.min

Lower truncation limit for the probability of being censored. It must be a positive value and should be chosen close to 0. Only applies for survival outcomes. Default is 0.025.

xvar.smooth.score

A vector of characters indicating the name of the variables used as the smooth terms if score.method = 'gam'. The variables must be selected from the variables listed in cate.model.

xvar.smooth.init

A vector of characters indicating the name of the variables used as the smooth terms if initial.predictor.method = 'gam'. The variables must be selected from the variables listed in init.model. Default is NULL, which uses all variables in init.model.

tree.depth

A positive integer specifying the depth of individual trees in boosting (usually 2-3). Used only if score.method = 'boosting' or if initial.predictor.method = 'boosting' with score.method = 'twoReg' or 'contrastReg'. Default is 2.

n.trees.rf

A positive integer specifying the number of trees in random forest. Used if score.method = 'randomForest' or if initial.predictor.method = 'randomForest' with score.method = 'twoReg' or 'contrastReg'. Only applies for survival outcomes. Default is 1000.

n.trees.boosting

A positive integer specifying the maximum number of trees in boosting (usually 100-1000). Used if score.method = 'boosting' or if initial.predictor.method = 'boosting' with score.method = 'twoReg' or 'contrastReg'. Default is 200.

B

A positive integer specifying the number of time cross-fitting is repeated in score.method = 'twoReg' and 'contrastReg'. Default is 3.

Kfold

A positive integer specifying the number of folds used in cross-fitting to partition the data in score.method = 'twoReg' and 'contrastReg'. Default is 5.

error.maxNR

A numerical value > 0 indicating the minimum value of the mean absolute error in Newton Raphson algorithm. Used only if score.method = 'contrastReg'. Default is 0.001.

max.iterNR

A positive integer indicating the maximum number of iterations in the Newton Raphson algorithm. Used only if score.method = 'contrastReg'. Default is 150.

tune

A vector of 2 numerical values > 0 specifying tuning parameters for the Newton Raphson algorithm. tune[1] is the step size, tune[2] specifies a quantity to be added to diagonal of the slope matrix to prevent singularity. Used only if score.method = 'contrastReg'. Default is c(0.5, 2).

seed

An optional integer specifying an initial randomization seed for reproducibility. Default is NULL, corresponding to no seed.

plot.gbmperf

A logical value indicating whether to plot the performance measures in boosting. Used only if score.method = 'boosting' or if score.method = 'twoReg' or 'contrastReg' and initial.predictor.method = 'boosting'. Default is TRUE.

verbose

An integer value indicating whether intermediate progress messages and histograms should be printed. 1 indicates messages are printed and 0 otherwise. Default is 0.

contrast.control

A named list of options for count or survival contrast regression. Supported names are solver ("C", the default, or "R"), error.maxNR, max.iterNR, tune, and optim.control (a named list passed to stats::optim on fallback). List entries override the corresponding top-level NR arguments. Options are kept separate from ..., which is forwarded to boosting. Count and survival outcomes support this list; continuous outcomes retain their existing solver.

rmst.control

Named options for Cox restricted mean survival integration. Supported entry: solver = "C" (default) or solver = "R". The C backend accumulates each patient's integral without allocating a patient-by-event-time matrix. Cox model fitting is unchanged. Available only for survival outcomes; independent of contrast.control.

ps.control

Named options for LASSO propensity-score interaction screening. Supported entry: solver = "C" (default) or solver = "R". Only affects screening when ps.method = "lasso"; the LASSO fit and interaction selection rule are unchanged. Kept separate from boosting's ... and other numerical controls.

rf.control

Named list of randomForestSRC options: mtry, nodesize, nsplit, nodedepth, and splitrule. Applies to both treatment arms and survival initial-predictor forests. Use n.trees.rf for the number of trees.

...

Additional boosting arguments; see boosting-control.

For response = "binary", score.method = "logistic" fits a logistic model with treatment interactions. "twoReg" cross-fits separate logistic outcome models and a treatment model, then projects the augmented inverse-propensity outcomes for each arm onto the covariates in cate.model using linear regression. The score is the difference of these projections, averaged over B repetitions of Kfold folds. It estimates the linear projection of the conditional risk difference if either the treatment model or both outcome models are correct, under identification and positivity assumptions. Propensity score truncation can affect this property. The projection need not lie between -1 and 1 and equals the conditional risk difference only if that difference is linear in the specified design. The difference between arm coefficients is returned in coefficients$twoReg and by coef(); the separate arm coefficients are stored in fit$twoReg$coefficients. Logistic model coefficients are on the log-odds scale and are stored in fit$logistic; they are not linear risk-difference score coefficients. Boosting does not have a linear score coefficient vector. With neither twoReg nor another linear binary score, coef() returns NULL. "boosting" fits separate Bernoulli gradient-boosting models in each treatment arm, with the number of trees selected by five-fold cross-validation. All three scores are predicted risk differences; subgroup effects are doubly robust risk differences. Binary boosting requires at least one outcome-model covariate and both outcome values in each arm. Its fitted models and selected tree counts are stored in fit$boosting. Scores and subgroup effects are estimated on the same data; they are not independent validation. For extra boosting controls, see boosting-control. Subgroup fits without both arms or with rank-deficient models return NA.

Value

Fitted objects retain treatment.labels in treatment 0/1 order for score plot labels. For count response, see description of other outputs in catefitcount(). For survival response, see description of outputs in catefitsurv().

Details

For count response, see details in catefitcount(). For survival response, see details in catefitsurv().

References

Yadlowsky, S., Pellegrini, F., Lionetto, F., Braune, S., & Tian, L. (2020). Estimation and validation of ratio-based conditional average treatment effects using observational data. Journal of the American Statistical Association, 1-18. DOI: 10.1080/01621459.2020.1772080.

Hoogland, J., Debray, T. P. A., Desachy, G., & Nguyen, T.-L. (2026). Introduction to methods for personalizing medicine. In T. P. A. Debray, T.-L. Nguyen, & R. W. Platt (Eds.), Comparative Effectiveness and Personalized Medicine Research Using Real-World Data (Chapter 17, pp. 421–463). Chapman & Hall/CRC. doi:10.1201/9781003300809-17

Hoogland, J., IntHout, J., Belias, M., et al. (2021). A tutorial on individualized treatment effect prediction from randomized trials with a binary endpoint. Statistics in Medicine, 40(26), 5961–5981. doi:10.1002/sim.9154

Jiang, X., Simoneau, G., Pellegrini, F., Shen, C., de Moor, C., & Tian, L. (2026). Visualization and interpretation of individualized treatment rule results. In T. P. A. Debray, T.-L. Nguyen, & R. W. Platt (Eds.), Comparative Effectiveness and Personalized Medicine Research Using Real-World Data (Chapter 20, pp. 513–536). Chapman & Hall/CRC. doi:10.1201/9781003300809-20

See also

Examples

# Count outcome
fit_1 <- catefit(response = "count",
                 data = countExample,
                 score.method = "poisson",
                 cate.model = y ~ age + female + previous_treatment +
                                  previous_cost + previous_number_relapses +
                                  offset(log(years)),
                 ps.model = trt ~ age + previous_treatment,
                 higher.y = TRUE,
                 seed = 999)
#> Variable trt was recoded to 0/1 with drug0->0 and drug1->1.

coef(fit_1)
#>                              poisson
#> (Intercept)              -0.58185218
#> age                      -0.01117701
#> female                    0.58829733
#> previous_treatmentdrugB   0.73724361
#> previous_treatmentdrugC   0.15545985
#> previous_cost            -0.13035642
#> previous_number_relapses -0.05339112

# \donttest{
# Survival outcome
library(survival)
tau0 <- with(survivalExample,
                 min(quantile(y[trt == "drug1"], 0.95), quantile(y[trt == "drug0"], 0.95)))

fit_2 <- catefit(response = "survival",
                 data = survivalExample,
                 score.method = c("poisson", "boosting", "randomForest"),
                 cate.model = Surv(y, d) ~ age + female + previous_cost +
                                           previous_number_relapses,
                 ps.model = trt ~ age + previous_treatment,
                 initial.predictor.method = "logistic",
                 ipcw.model = ~ age + previous_cost + previous_treatment,
                 tau0 = tau0, higher.y = TRUE, seed = 999, n.cores = 1)
#> Variable trt was recoded to 0/1 with drug0->0 and drug1->1.


#>     Warning: Warning(s) occurred when estimating the ATEs in the nested subgroup using "poisson" 
#> Warning: Warning: Warning(s) occurred when estimating the ATEs in the nested subgroup using "poisson"; see 'errors/warnings'.

coef(fit_2)
#>                              poisson
#> (Intercept)               0.46109243
#> age                      -0.24335209
#> female                    0.12532711
#> previous_cost             0.89803249
#> previous_number_relapses -0.02470863

# }