Propensity score profiling - a software package for comparing among multiple groups using propensity score with shrinkage estimation.
 

Huang , IC, Frangakis, CE, Dominici , F, Diette, G, and Wu, AW (2004). Application of a propensity score approach for risk adjustment in profiling multiple physician groups on asthma care. Forthcoming in Health Services Research.

The authors have no responsibility of any consequences resulting from the use of the software

 

A. PURPOSE OF THIS SOFTWARE

The software is written for the above paper. Provider profiles are used increasingly to enhance accountability and help consumers choose providers or health plans. However, comparisons of provider performance can be biased when patients cared for by different providers differ in background characteristics. In this study, we propose a method that employs propensity scores to achieve better balance of covariates. The method uses the propensity score to risk-adjust for multiple groups while addressing regression to the mean that can result from such multiple comparisons, based on Bayesian principles.

B. SOFTWARE DOWNLOADS

  1. Download software R from http://www.r-project.org/

 

  1. Download propensity score routine (routine.r) from

http://www.biostat.jhsph.edu/~cfrangak/papers/proscore_profiling/routine.r

 

C. TECHNICAL NOTES FOR USING PROPENSITY SCORE ROUTINE

 

1.      The routine are to be used for

·        Data without missing values

·        Binary outcomes, such as satisfaction vs. dissatisfaction with health care

·        Covariates entered by their names in the routine’s input. If a categorical covariate named “x2” is to be used as a factor, modify the input (myregressors="x1+x2"…) to be (myregressors="x1+factor(x2)"…).

·        Assigning group “g” as the reference group

 

  1. Steps of estimation

·        Step 1:  Estimates propensity score of each subject enrolling in each of the groups.

·        Step 2:  Estimates quantities of subjects within each groups based on propensity scores of group enrollment.

·        Step 3:  Obtains overall outcome probabilities adjusting for the propensity scores, and variance estimates of the probabilities and logit(probabilities).

·        Step 4:  Estimates shrinkage logit(probabilities) using the method of Everson, PJ and Morris, CN (1993).

·        Step 5:  Tests for equality of outcome across groups, using simulation.

·        Step 6:  Estimates odds ratios of comparing each group with the reference group given in the input, and provides confidence intervals based on variance estimates obtained by simulation.

D.    SUGGESTION

Users are encouraged to read the above-referenced paper by I-Chan Huang et al. (2005) to understand the assumptions involved in this method.
 

Return to Dr. Frangakis's  Home Page