Skip to contents

Extract treatment effects or treatment-specific outcomes for the nested subgroups in a binary, count or survival catefit object.

Usage

subgroup_estimates(object, method = NULL, measure = NULL)

Arguments

object

A binary, count or survival catefit object.

method

One scoring method used to fit the object, such as "contrast_regression" or a T-learner outcome method such as "poisson". The same method selectors as predict() are accepted; see catefit-extract. If omitted, it is inferred when exactly one method is available.

measure

For binary outcomes, "risk_difference" (risk under treatment 1 minus treatment 0). For count outcomes, "rate_ratio". For survival outcomes, one of "rmtl" (RMTL ratio), "cumulative_hr" (marginal cumulative hazard ratio), "conditional_hr" (conditional Cox hazard ratio), "risk0", "risk1", "risk_difference" (risk under treatment 1 minus treatment 0), "absolute_risk_reduction" (the opposite contrast), "rmst0", or "rmst1" at tau0. The default is "risk_difference" for binary, "rate_ratio" for count, and "rmtl" for survival.

Value

A named numeric vector, with one estimate per nested subgroup in increasing order of prop.cutoff. Treatment-specific risks and RMSTs are outcomes, not treatment effects. Undefined estimates are NA.

Details

The subgroups are ranked by the selected score and estimated using the same data used to fit it; these results do not validate the score. The cumulative hazard ratio and risk estimates use outcome, treatment, and censoring models; a causal interpretation requires appropriate treatment and censoring assumptions. Older fits without the requested estimates must be refitted. No confidence intervals are returned. The measure names are also accepted in dotted form.

Author

Thomas Debray