gms | German Medical Science

Introduction: The MIMIC-III database (Medical Information Mart for Intensive Care, v1.4) is a public intensive care database, which documents in detail the course and treatment of more than 40,000 patients and has already proven to be a useful resource for visualization in the medical field [1], [2], [3], [4]. In the present work we used Tableau Desktop, a commercially available visual analytic software, to visualize MIMIC-III data by building dashboards. It can be linked and serve as a visualization tool for clinical data and might be useful for non-primarily technical users through its easy-to-use drag- and drop functionality [5].

State of the Art: A recent study used the KNIME analytics platform to implement a workflow for the MIMIC-III data and the related MIMIC-III Waveform Database particularly for time-series data [4]. Another approach for the previous version database (MIMIC-II) established a data visualization tool based on web-based technologies such as HTML, CSS, and JavaScript libraries on the client side and tools like PHP and PostgreSQL on the server side [3]. Furthermore, a visualization tool which is only usable in the context of MIMIC-III data is provided by the PhysioNet group, the publisher of the MIMIC-III Dataset [6].

Concept: We focus on the following limited potential use cases, which may be useful for health researchers and clinicians.

1.

A user selects a patient cohort based on available data categories and check their distributions (Cohort Selection Dashboard).

2.

A user receives a brief summary of the data on the patient level, such as age, gender, diagnosis, laboratory tests and more (Patient Summary Dashboard).

3.

A user checks all the prescribed medication at a specific date and time as well as the medication duration (Prescription Timeline Dashboard).

4.

A user checks the course of time-dependent variables such as glucose levels on the patient level and dynamically adjust them (Laboratory View Dashboard).

Implementation: The basis for our implementation is the commercially available software Tableau Desktop (Version 2020.3) provided by Tableau (Seattle, Washington, USA) [5]. We used the comma-separated values (CSV) files of the demo MIMIC-III data and built the corresponding data model based on the information about the data types provided by the PhysioNet website and the data schema provided by the SchemaSpy analysis of MIMIC-III [6], [7]. The dashboards were published for demonstration purpose on the Tableau Public server [8].

Lessons Learned: The created Tableau dashboards represent a basic data visualization tool for clinical data and allow researchers to perform exploration and basic analysis of their data. The tool provides cohort selection options and displays key patient parameters and charts of time-series variables. The presented approach can be expanded or adapted to specific questions and thus supplement already established visualization methods. Nonetheless, it must be considered that performance difficulties can arise with generic approaches, in which the complete data structure needs to be connected in advance for free data exploration. These approaches are of particular interest to the medical field. Whether and how generic approaches can be implemented with Tableau dashboards should be investigated in future work.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.