JMIR Publications

ABSTRACT

Background:

The increasing amount of open-access medical data provides new opportunities to gain clinically relevant information without recruiting new patients. Even in the absence of the full desired parameter control, this information can be used to support the development of on-site studies. Open-source computational pipelines, linked to specific data collections, facilitate access to the necessary data and ensure consistency in data processing and analysis, thus reducing the risk of errors and creating comparable and reproducible results. Understanding the link between neuroactive medications and physiological brain changes is one of the important goals which can be supported with open data and standardized computational approaches. Such changes can be correlated to the shifts in electroencephalography spectrum and understood better through standardized computations on open data sets.

Objective:

We present a computational pipeline, which explores the publicly available electroencephalographic (EEG) data of the Temple University Hospital (TUH) to identify EEG profiles associated with the usage of specific neuroactive medications. This pipeline allows to create a custom list of medications, access the necessary EEG records and compare the spectral content within the defined groups of interest.

Methods:

The open-source computational pipeline is constructed using easily controlled modules. It allows the user to define the medications of interest and their classes, as well as the comparison groups. Accordingly, the pipeline downloads and preprocesses the necessary data, extracts the spectral features and performs the statistical group comparison with final visualization through a topographic EEG map. The pipeline is easily adjustable to answer a variety of specific research questions. In order to illustrate the pipeline performance, two commonly used and well-researched drugs, Carbamazepine and Risperidone, were chosen to be compared with the control data and with other medications from the same classes (correspondingly anticonvulsants and antipsychotics).

Results:

The illustrative computations on Carbamazepine and Risperidone showed good agreement with the state-of-the-art research, however, several differences were observed and discussed. One of the major problems with the result comparison is the lack of uniform study designs and computational procedures in the literature. The computational times of the initial data downloading and preprocessing are relatively high, but subsequently multiple different hypotheses can be tested very quickly. The size of the database allows us to create arge groups for many different neuroactive medications, even if multi-drug records are eliminated. Additionally, TUH Corpus provides a collection of records labeled as “normal” after expert assessment, which is convenient for the creation of the control group.

Conclusions:

The pipeline allows fast testing of different hypotheses regarding the links between the medications and EEG spectrum through ecological usage of the readily available data. Despite the lack of full study parameter control, it can be utilized to make informed decisions about the design of new fully controlled clinical studies.