R hands-on projects archive

R program for randomization in clinical trials, including simple, blocked, stratified, stratified-blocked, and dynamic randomization. Most R packages for randomization requires complete information about the all study samples (for example, covariate values). However, in real-world setting, patients enrollment is usually conducted during a period of time, so complete information is not available. One advantage of this R program is it allows patients information to be added incementally and assign corresponding assignment for newly enrolled study samples. The Full codes can be access here.

Conducted a simulation study to investigate which statistics(namely T test statistics with/without equal variance, Z test, and wilcox) should be used for two sample hypothesis testing under different scenarios(data distribution, sample size) by assessing type I and type II error. The results are documented here.

Metrics for risk prediction model: Risk prediction research papers require reporting performance metrics. Some R packages exist, however, you may have to use several functions to get the metrics you want to calculate. So I wrote a function where you can easily get most of the performance metrics() all at once. The output is a neat table, so you can save it to your local PC. All you need to provide is the predicted probabilities and actual labels.

In my honor’s thesis(thesis paper,dissertation slides), I proposed a new test statistics for ROC comparison when intersect exists: The area under the ROC curve (AUC) has been widely used for evaluating performance in diagnostic tests. In practical applications, the two ROC curves may intersect. This study aims to explore new statistical method for performance evaluation when ROC curves intersect and to use Monte-Carlo simulation to compare the proposed method with Delong’s method. Finally, the differences between the two methods are discussed with examples.