Software development
Transparency and reproducibility
The Effective Statistician Podcast: The FAIRification Of Research In Real-World Evidence: A Practical Introduction To Reproducible Analytic Workflows Using
https://podcastae8fac.podigee.io/370-new-episode
smdi - An R package to perform routine structural missing data investigations in real-world data
Partially observed covariates are a common challenge in the analysis of electronic health records. Backed by large-scale simulations, this package eases and streamlines the implementation of routine missing data checks to characterize the underlying missingness and make informed decisions about the appropriate analytical choice for your study.
- Available on CRAN
install.packages("smdi")
Link to package website
Package presentation at the New England Statistics Symposium 2023:
autoencoderPS - An autoencoder-based propensity score for causal inference
I’m fascinated by neural networks and deep learning and the following paper highlights some of my thoughts around utilizing these methods for causal inference:
Weberpals J, Becker T, Davies J, Schmich F, Rüttinger D, Theis FJ, Bauer-Mehren A. Deep learning-based propensity scores for confounding control in comparative effectiveness research: A large-scale, real-world data study. Epidemiology. 2021 May 1;32(3):378-88.
The development and analysis code published in the article “Deep learning-based propensity scores for confounding control in comparative effectiveness research: A large-scale, real-world data study” (Weberpals et al., Epidemiology, 2021) can be accessed on the following Github repository:
The computing code used in this study is available as Python Jupyter Markdown scripts (.html) as supplementary material. All of the analyses described in the article were performed in R version 3.2.2. The PCA and autoencoder training was performed using sckit-learn and Keras with Tensorflow backend in Python version 3.6.0, respectively. The code that was used for the simulation is available as Rmarkdown.