Public datasets.

Download the 8 public datasets to measure the effects of the site (EoS)-removal effectiveness of a MAREoS and the R scripts to conduct the calculations.

Download public datasets
Download R scripts

General instructions

For each fold of the 10-fold cross-validation, find the effects of site and the two clinical covariates in the training subset, remove them from both the training subset and the test sets, fit a lasso algorithm in the training subset, and predict the response in the test set. Afterward, find the balanced accuracy and estimate the EoS-removal effectiveness.

Example for fold 1

  1. Find the effects of the site and the two clinical covariates using individuals labeled to be in folds folds 2 to 10 (the "training subset"). To find the effects of the site, fit the MAREoS of interest. To find the effects of the covariates, fit linear regressions.
  2. Remove the effects of the site and the covariates from these individuals and fit a lasso algorithm. To remove the effects, use the models fitted in step 1. In the logistic regression, the dependent variable is the response to the treatment, and the independent variables are the MRI data (e.g., cortical thickness measures).
  3. Remove the effects of the site and the covariates from individuals labeled to be in fold 1 (the "test set") and predict their response to the treatment. To remove the effects, use the models fitted at step 1.
  4. Find the sensitivity, specificity, and balanced accuracy.
  5. Estimate the EoS-removal effectiveness.