Selection and Misclassification Bias Adjustment for Logistic Regression Models

Health research using data from electronic health records (EHR) has gained popularity, but misclassification of EHR-derived disease status and lack of representativeness of the study sample can result in substantial bias in effect estimates and can impact power and type I error for association tests. Here, the assumed target of inference is the relationship between binary disease status and predictors modeled using a logistic regression model. 'SAMBA' implements several methods for obtaining bias-corrected point estimates along with valid standard errors as proposed in Beesley and Mukherjee (2020) , Biometrics.

SAMBA

Health research using data from electronic health records (EHR) has gained popularity, but misclassification of EHR-derived disease status and lack of representativeness of the study sample can result in substantial bias in effect estimates and can impact power and type I error for association tests. Here, the assumed target of inference is the relationship between binary disease status and predictors modeled using a logistic regression model. SAMBA implements several methods for obtaining bias-corrected point estimates along with valid standard errors as proposed in Beesley and Mukherjee (2020), currently under review.

Installation

SAMBA can be downloaded from Github via the R Package devtools

devtools::install_github("umich-cphds/SAMBA", build_opts = c())

Vignette

Once you have SAMBA installed, you can type

vignette("UsingSAMBA")

in R to bring up a tutorial on SAMBA and how to use it.

Questions

For questions and comments about the implementation and methods, please contact Lauren Beesley ([email protected]).

Reference

Statistical inference for association studies using electronic health records: handling both selection bias and outcome misclassification Lauren J Beesley, Bhramar Mukherjee Biometrics 10.1111/biom.13400