The sequential randomization tool is a web application that can be used to support research trials in which the subjects are enrolled sequentially into the trial. The subjects are assigned to treatment arms as they are recruited, using the Pocock and Simon minimization algorithm (Biometrics 31, 1975). Assignment to treatment groups is random, but the assignment probabilities are adjusted to promote balance with respect to confounding factors.

Example: Suppose we are conducting a trial aiming to compare three approaches to math tutoring. The subjects take a pre-test before beginning the study, where the possible scores are low, medium, and high. The following treatment group assignments would be considered unbalanced with respect to the pre-test scores:

In the situation illustrated by the table sown above, a simple comparison of post-test scores could be biased. We might expect treatment C to do better than treatment A, since stronger students were enrolled in treatment C compared to treatment A. This apparent treatment effect of method C may not have anything to do with the treatment itself.

Since the treatment assignments are partially random, we do not obtain perfect balance when using the minimization algorithm. However the results will be approximately balanced with high probability. For example, we might obtain the following:

The randomization tool will ensure that within each pre-test score band, the relative numbers of people in the treatment arms are similar.

Some features of the randomization tool are:

• Runs on Google Cloud Platform/Google AppEngine, typical usage on a standard Google user account does not incur any charges

• Cloud-based data storage in the Google Cloud Datastore, data are replicated on multiple servers

• Login and authentication using Google accounts

• Unlimited projects per user

• Role-based access, e.g. project leaders may delete a project but study managers cannot to this

• Supports multi-center trials

• Option to disable online storage of disaggregated data

• Customization of post-randomization data editing

Users are authenticated through their Google account. This is the only form of authentication that is currently supported.

The randomization tool must store aggregated data to apply the minimization algorithm ("aggregated data" are the marginal totals per treatment arm within each level of each covariate). Disaggregated data are not stored by default, but the project creator can opt-in to have disaggregated data stored.

The data are stored in a cloud database on Google servers. If the Google AppEngine account on which the randomization tool was installed is compromised, data could be lost or viewed by others. Google supports two factor authentication for greater security. If a project owner's account is compromised, the project could be altered or viewed by others. If a study coordinator's account is compromised, the data could be viewed by others, but any changes could be reverted (reverting is only possible with opt-in for storing disaggregated data).

The project is written in the Go programming language, but no knowledge of Go is needed to install or use the application.

1. Download and unpack the latest version of the application.

2. Download the appropriate version of the Go AppEngine Software development kit for your platform here. Unpack and install as appropriate for your system.

3. Navigate to https://console.cloud.google.com to create a project. More detailed instructions are here. At this page you will choose an application id that will become part of the URL of your web application.

4. Edit the "app.yaml" file in the src directory of your local copy of the randomization source code. Replace "randomization" in the first line of "app.yaml" with the application id that you selected in step 3 when creating your AppEngine project on-line.

5. Deploy your application using the command goapp deploy or using appcfg.py. Both of these commands must be issued from the command line. You must be in the src directory of the randomization project when deploying the application. When you first run goapp or appcfg.py, your web browser will launch an authentication page where you must agree to link the app to your Google account.

6. If your application id is "my_randomization", then after deployment it will be available at the URL "my_randomization.appspot.com". Note that the first time you install the application it will take a few minutes for AppEngine to build the database indices. While this is happening, the application will not be usable.

You can perform any of these simple customizations:

• Replace the file stylesheets/logo.png with a PNG containing your organization's logo, it will then be displayed on the landing page.

• Add a stylesheets/favicon.ico icon file which will be displayed on the browser tab.

• Edit the text in the information_page.html file however you like.

The project is hosted at https://github.com/kshedden/randomization. Bug reports, feature requests, and pull requests can be made through the Github site.