quint (version 2.1.0)

quint: Qualitative Interaction Trees

Description

This is the core function of the package. It performs a subgroup analysis by QUalitative INteraction Trees (QUINT; Dusseldorp & Van Mechelen, 2014) and is suitable for data from a two-arm randomized controlled trial. Ingredients of the analysis are: one continuous outcome variable \(Y\) (the effect variable), one dichotomous treatment variable \(T\) (indicating two treatment conditions, e.g., A and B), and several background characteristics \(X1,\dots,XJ\). These background characteristics are measured at baseline and may have a numeric or ordinal measurement level (i.e., in R a numeric or integer variable) or a nominal measurement level (i.e., in R a factor). They are used to identify the following subgroups (i.e., partition classes): Subgroup 1: Those patients for whom Treatment A is better than Treatment B (P1); Subgroup 2: Those for whom Treatment B is better than Treatment A (P2), and Subgroup 3: Those for whom it does not make any difference (P3).

Usage

quint(formula, data, control = NULL)

Arguments

formula

a description of the model to be fit. The format is Y ~ T | X1 + … + XJ, where the variable before the | represents the dichotomous treatment variable \(T\) and the variables after the | are the baseline characteristics used for partitioning. If the data are in the order Y, T, X1,…, XJ, no formula is needed. The lay-out of this formula is based on Zeileis & Croissant (2010).

data

a dataframe containing the variables in the model. The treatment variable can be a numeric or a factor variable with two values (or levels).

control

a list with control parameters as returned by quint.control.

Value

Returns an object of class quint with components:

call

the call that created the object.

crit

the partitioning criterion used to grow the tree. The default is the Effect size criterion. Use crit="dm" for the Difference in means criterion.

control

the control parameters used in the analysis.

fi

the fit information of the final tree.

si

the split information of the final tree.

li

the leaf information of the final tree. Treatment A is denoted with \(T\)=1, and treatment B is denoted with \(T\)=2. Can display either the output for Difference in Means (crit='dm') or Cohen's d effect size (crit='es').

data

the data used to grow the tree.

orig_data

the original data used as input.

nind

an \(N\) x \(L\) matrix indicating leaf membership.

siboot

an \(L\) x 9 x \(B\) array with split information for each bootstrap sample: C_boot = value of \(C\); C_compdif = value of Difference in treatment outcome component; checkdif = indicates if pooled Difference in treatment outcome component in test set (i.e., original sample) is positive, with values: 0 = yes,1 = negative in P1, 2 = negative in P2, 3 = negative in P1 and P2; C_compcard = value of Cardinality component;checkcard = indicates if value of pooled cardinality in test set is zero, with values: 0 =no,1 = zero in P1, 2 = zero in P2, 3 = zero in P1 and P2; opt = value of optimism (C_boot-C_orig).

indexboot

an \(N\) x \(B\) matrix indicating bootstrap sample membership.

formula

a description of the model to be fit.

pruned

a boolean indicating whether the tree has been already pruned or not.

Details

The method QUINT uses a sequential partitioning algorithm. The algorithm starts with a tree consisting of a single node, that is, the root node containing all patients. Next, it follows a stepwise binary splitting procedure. This procedure implies that in each step a node, a baseline characteristic, a split of that characteristic, and an assignment of the leaves of the current tree to partition classes 1, 2, and 3 (P1 to P3) are chosen that maximize the partitioning criterion. Note that this means that after each split, all leaves of the tree are re-assigned afresh to the partition classes P1, P2, and P3.

References

Dusseldorp, E., Doove, L., & Van Mechelen, I. (2016). Quint: An R package for the identification of subgroups of clients who differ in which treatment alternative is best for them. Behavior Research Methods, 48(2), 650-663. DOI 10.3758/s13428-015-0594-z

Dusseldorp E. and Van Mechelen I. (2014). Qualitative interaction trees: a tool to identify qualitative treatment-subgroup interactions. Statistics in Medicine, 33(2), 219-237. DOI: 10.1002/sim.5933.

Zeileis A. and Croissant Y. (2010). Extended model formulas in R: Multiple parts and multiple responses. Journal of Statistical Software, 34(1), 1-13.

van der Geest M. (2018). Decision Trees: Amelioration, Simulation, Application. Can be found in: https://openaccess.leidenuniv.nl/handle/1887/65935

See Also

summary.quint, quint.control, prune.quint, bcrp, quint.bootstrapCI

Examples

# NOT RUN {
#EXAMPLE with data from the Breast Cancer Recovery Project
data(bcrp)
#Start with expliciting the model for quint
#The outcome Y is a change score between timepoint 3 and timepoint 1
#A positive Y value indicates an improvement in depression (i.e., a decrease)

formula1<- I(cesdt1-cesdt3)~cond | nationality+marital+wcht1+age+
  trext+comorbid+disopt1+uncomt1+negsoct1

#Perform a quint analysis
#The BCRP data contain 3 conditions. Quint only works now for 2 conditions.
#For the example, we disregard the control condition
#To save computation time, we also adjust the control parameters

set.seed(2)
control1<-quint.control(maxl=5,B=2) #The recommended number of bootstraps is 25.
quint1<-quint(formula1, data= subset(bcrp,cond<3),control=control1)
quint1pr<-prune(quint1)

#Inspect the main results of the analysis:
summary(quint1pr)

#plot the tree
plot(quint1pr)

# }