Supplementary Information for the paper "Identifying patients using antidepressants for the treatment of depression"

This R package contains similar (synthetic) data to the dataset used in the paper, and all R code used in the analysis.

Install this package

install.packages('devtools')
devtools::install_github("CBDRH/antidepressants")

Load synthetic data

library(antidepressants)
data("synthetic")

Bootstrap and fit glinternet models

bootglinternet(B = 10, nLambda = 20)

All coefficients of the model fitted to the real dataset

See the file all_coefs.md or all_coefs.csv

Reproduce analysis

View the README.Rmd file that produces the output below. Note that the output below is derived from the synthetic data and 10 Bootstrap samples only, hence the signal is weaker accordingly.

Calculating AUC

For each value of λ, we give the mean value of the AUC (Area under the ROC curve) and its standard deviation:

The best expected AUC is 0.7030967 and our choice within one standard deviation is 0.6929934.

Sensitivity and Specificity

threshold	gold	positives	sens	spec	ppv
0.1	1323	1115.8	1.000	0.000	0.438
0.2	1323	1111.8	0.999	0.006	0.439
0.3	1323	1002.2	0.972	0.159	0.474
0.4	1323	772.2	0.833	0.417	0.527
0.5	1323	285.7	0.378	0.839	0.655
0.6	1323	31.9	0.044	0.983	0.676
0.7	1323	0.0	0.000	1.000	NaN
0.8	1323	0.0	0.000	1.000	NaN
0.9	1323	0.0	0.000	1.000	NaN

ROC plot

Plot ROC curves of 5 bootstrap samples.

Bootstrap Percentiles for effects and interactions

Which effects were picked

effect_name	num_level
amitriptyline	4
mirtazapine	3
reboxetine	2
comed_iron	2
age	1

Which interactions were picked

effect_name_1	effect_name_2
amitriptyline	mirtazapine
amitriptyline	reboxetine
amitriptyline	comed_iron