zebu (version 0.1.3.0)

subgroups: Local Association Subgroup Analysis

Description

Identifies if the local association between variables (named associated variables) is dependent on the value of an another variable (named interacting variable). Associated variables are specified by las. Interacting variable(s) values are specified by x.

Usage

subgroups(
  las,
  x,
  select,
  continuous,
  breaks,
  default_breaks = 4,
  thresholds = c(-0.05, 0.05),
  significance,
  alpha = 0.01
)

Arguments

las

lassie S3 object. Corresponds to associated variables.

x

data.frame or matrix. Corresponds to interacting variable(s) specified by select.

select

optional vector of column numbers or column names specifying a subset of data to be used. By default, uses all colnames in x except those in las object.

continuous

optional vector of column numbers or column names specifying continuous variables that should be discretized. By default, assumes that every variable is categorical.

breaks

numeric vector or list passed on to cut to discretize continuous variables. When a numeric vector is specified, break points are applied to all continuous variables. In order to specify variable-specific breaks, lists are used. List names identify variables and list values identify breaks. List names are column names (not numbers). If a continuous variable has no specified breaks, then default_breaks will be applied.

default_breaks

default break points for discretizations. Same syntax as in cut.

thresholds

vector specifying respectively the negative and the positive association threshold. Local association values between these thresholds are considered independent. Values not contained in this range are classified as independent.

significance

optional logical value specifying if only non-significant local association values should be considered as independent. Only available if las is also a permtest object.

alpha

alpha error level. Local association with p-values above this value are considered as independent. Only available if las is also a permtest object.

Value

An instance of S3 class lassie.

Details

Associated variables events are recoded into a subgroup variable according to local association values (and eventually significance) into 'positive', 'negative' and 'independent'. This is specified by the thresholds, significance and alpha arguments. The local (and global) association between the new subgroup variable and the interacting variable is then estimated using lassie.

See Also

Significance can be accessed using a permutation test: permtest.

Examples

# NOT RUN {
# In this example, we will use the zebu 'trial' dataset.
# See vignette example for more detailed explanation

# 'trial' corresponds to a simulated clinical trial where patient recovery
# is dependent on drug intake ('drug') and resistance status ('resistance').
# Patient recovery is monitored by a biomarker (continuous variable from 0 to 1)
# Patients with post-treatment biomarker ('postbiom') above 0.7 is have recovered.

# Load 'trial' dataset
data(trial)

# Compute the association between drug intake and patient recovery
las <- lassie(trial,
              select = c("drug", "postbiom"),
              continuous = c("postbiom"),
              breaks = c(0, 0.7, 1))

# Permuation test
# Access significance of global and local association
las <- permtest(las)

# Global association between drug intake and recovery but not for all patients
# Being in the drug group is locally independent of having not recovered
print(las)

# Local association subgroup analysis
sub <- subgroups(las, trial, select = "resistance", alpha = 0.01)

# Variable 'resistance' explains differences between sensitive and resistance patients
print(sub)

# }