SGPSM

Similarity measures for generalized propensity score matching (SGPSM) is a propensity score method for multiple treatments matching. It can reduce confounding effects and provide unbiased average treatment effect for the treated (ATT) for multiple treated groups, similar to other propensity score methods. This R code implements SGPSM for propensity score matching with (k >= 3) treated groups.

Usage

Example:

data <- read.csv("~/data.csv")

source("~/SGPSM.R")

matched_data <- SGPSM(
  df = data,   # file name
  treat_group = "trt",   # column name of treatment
  gpsm = c("X0","X1","X2"),    # input column name of propensity scores vector
  algo = "manhattan",    # cosine, euclidean, manhattan
  caliper = 0.1   # criteria for selecting subjects in treatment A whose propensity score is "close" to that of a subject in treatment B
)

• df:   File name.
  
• treat_group:   Variable for treatment groups.
  
• gpsm:   Variables for generalized propensity score (input column name of propensity scores vector).
  
• algo:   Algorithm of cosine, euclidean, manhattan.
  
• caliper:   Caliper.
  

Dependencies

R package: dplyr >= 1.0.8, proxyC >= 0.2.4

Papers

Yi-Chia Su, Chih-Chien Wu, Yi-Hau Chen, Pei-Ting Lee, Chien-Chou Su. Similarity Measures for Generalized Propensity Score Matching With 3 Target Therapies in Cancer Study: Drawing From Distance Metrics in Unsupervised Learning (2024; submission)

Correspondent: Chien-Chou Su