Streamlining clinical data management: The design of a workflow for multi-modal heterogeneous real-world data to improve ventricular tachycardia research

Introduction
In hospitals an abundance of data is created but not always used to its full potential. Coupling different multimodal data and analysing their time-evolving trends might unravel new, useful insight. For post-myocardial infarction (MI) ventricular tachycardia (VT) patients, primary risk stratification for sudden death and life-threatening arrhytmias is based on left ventricular ejection fraction (LVEF) and symptoms. However, improved risk assessment is highly required as many patients remain undertreated, and many other patients are overtreated with expensive and invasive therapies. LVEF is a general measurement which barely encompasses the VT development complexity. This complexity might be better interpreted by visualizing and analysing the different routinely obtained data of MI patients.
QME design project
The availability of this data however is not straight forward. In this QME (Qualified Medical Engineer) project, a workflow was designed to gather, clean, filter and analyse multimodal data. The workflow was applied successfully in the COMBAT-VT project, in two subsequent retrospective studies comprising almost 3000 post-MI patients, with data covering the period between 1995-2022. The developed workflow consists of six components:
1. Patient selection: selection of patients and classification of VT occurrence;
2. Import data: obtain the data from different databases at the CZE;
3. Database: saving the data in a structured, cleaned and pseudonymized database;
4. Analysis: explore metadata and provide statistical tools and analysis;
5. Share: enable access to the data while considering privacy risks, laws and regulations;
6. Dashboard: visualize multimodal data while providing context, in an interactive user-friendly tool.
Implementation of the design
The first step of this QME project was the Pilot-VT study, in which the workflow was designed, developed and applied. The Pilot-VT included 11 data modalities and 46 MI patients, of which 10 patients had developed VT. Secondly, the follow-up study of COMBAT-VT23 included 6 data modalities of 2892 MI patients, among which 166 VT survivors. The two studies are both unique in their size; the Pilot-VT in the number of data modalities, the COMBAT-VT23 study in the number of patients included. A substantial number of (automated) methods and generalizable solutions were designed as a part of the workflow, which can be applied to other research projects, while the data from both the Pilot-VT and COMBAT-VT23