Who
The Leveraging AI in SARS COVID-2/COVID-19 Data in Rwanda (LAISDAR) project1 was launched in late 2020 in response to the COVID-19 pandemic, with the objective to leverage centralized COVID-19 test results and survey data to support the Rwandan government’s needs in monitoring and predicting the COVID-19 burden2. Today, the LAISDAR federated network consists of initial 11 hospitals, 2 health centres and 1 survey dataset, but is being widened with 9 additional hospitals, making it the largest ever created onAfrica continent.
The Challenge and Why It Matters
Health data in Rwanda, similar to other places in Africa and worldwide, is currently fragmented and often incomplete, as they are scattered across different institutions like hospitals, clinics and laboratory testing sites. Analysis of fragmented datasets results in poor evidence. In response, the network was designed to support federated data analyses to drive evidence-based COVID-19 patient care, and to provide insight (e.g., hospital admission, infection rates) on the burdens of COVID-19 in the Rwandan health community.
To support the Rwandan government’s objectives, the network aimed to fulfill the following:
- Data standardization into one data structure to facilitate efficient and accurate data analyses
- Original data owner to remain in complete control of dataset to respect ethical and privacy rules
- Easy to maintain infrastructure to ensure sustainable operation by local technical expertise
- Use of open-source standards and tools, whenever possible, to leverage ready-made resources such as data visualization and dashboarding tools
- Create a sustainable and scalable data ecosysteminfrastructure enabling pandemic and epidemic surveillance and intelligence forother areas such as infectious diseases and others
The process
edenceHealth led the design and implementation of the technical solution for the LAISDAR project, including the development of the ETL pipeline and the infrastructure set-up. Two different Electronic Health Record (EHR) systems, openClinic GA and openMRS, are in use by the participating institutions. Additionally, a national dataset of COVID-19 diagnostic test results and follow-up COVID-19 survey results were available.
Data Standardization – Structural and Semantic Mapping
The Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) was selected as the data standardization model as it is an open community data standard, with a wealth of existing documentation, training materials and tools. To begin the standardization of the various datasets, edenceHealth started by mapping the EHR data structures to the OMOP-CDM clinical data tables. Using Rabbit-In-A-Hat, a “scan report” of the existing data tables and fields within the EHR is generated. Upon thorough review and understanding of the scan report, our analysts document structural mapping logic to translate the EHR data into the target tables; this document become the instructions for our ETL developers.
A central component of the OMOP CDM is the OHDSI standardized vocabularies. This allows for organization and standardization of medical terms to be used across various clinical domains in the OMOP CDM, and ensures standardized analytics by OHDSI collaborators. In parallel with the structural mapping, our analysts review the responses available in the scan report and design a semantic mapping document of the information to the standardized OMOP vocabularies.
Extract-Transform-Load Development
In this project, we developed comprehensive and robust ETL (Extract, Transform, Load) solutions to manage the data transformation processes for the different data sets. At the core of the solutions is the creation of a Docker container, ensuring consistency across environments. The ETL process is controlled by environment variables set at the start of the workflow, which act as configurations for the process. A single ETL was created for the two EHR systems, which is also configurable by setting the environment variables to either run the openMRS or openClinicGA transformations.
The transformation for each table in the target database is managed by a dedicated Python script, designed for modularity and clarity. These scripts embed PostgreSQL queries customized to the specific requirements of each table. The main Python script orchestrates the entire workflow, managing the sequence in which the transformations are performed, handling dependencies, and ensuring efficient execution. This flexible design not only optimizes performance but also provides an easily adaptable process in case further updates are needed based on validation or feedback from the sites.
Infrastructure Design and Set-up
In the LAISDAR network, a central node was set-up at the Ministry of Health (MOH) and hosts the national COVID-19 datasets. Each hospital node was also equipped with a dedicated workstation machine (here we used MacMini), which serves as hospital data nodes. An API established at the central server hosted by the Rwanda Biomedical Center, RBC (an implementation body of Ministry of Health) facilitates encrypted datatransfers between the central COVID data repository and the core EHR data. The hospital data nodes were installed, configured and hand delivered to each participating institution. Docker was used to package and deploy all the tools, including an OMOP CDM database instance (PostgreSQL), OHDSI tooling (Atlas, R/RStudio (w/HADES), Achilles/DQD, and ARES Indexer) and other supporting services. With sustainability in mind, we also installed SimpleMDM, an Apple-focused mobile device management solution, to apply application updates and orchestrate processes.
Data Quality and Validation
The application of another OHDSI tool, ARES, played a key role in supporting the quality control and validation processes across the LAISDAR network. ARES is an open-source tool designed to monitor network data quality by providing a centralized view on real-time quality issues and mapping statuses across the data sites. edenceHealth deployed the ARES web application on the central server, and used SimpleMDM to launch automated scripts that executed and uploaded the ARES Indexer results from the data nodes following each ETL run. In addition, the processes were automated, such that each site’s Atlas instance refreshed as part of the ETL process. A copy of the aggregated results (data profile) could be uploaded to the central server as needed, to allow review of the different data node profiles in one central location.
Data Reporting
Although it was not a key component in the initial requirements, we learned through interaction with the participating sites that an efficient reporting tool was central to ensuring high usability of the OMOP CDM database. With this in mind, we designed and developed a proof-of-concept (POC) solution for creating monthly quality reports with OMOP CDM data, based on the EHDEN CdmInspection R package. The POC solution was popular among the institution as it streamlined an effort-intensive monthly reporting process that was performed entirely by hand.
Tangible outcomes
As of June 2023 (time of project wrap-up), the LAISDAR project has successfully integrated 14 hospital nodes and transformed their EHR data into OMOP CDM, with a total of ±3.5M patients represented. COVID-19 test results have also been converted into OMOP CDM for ±10 000 participants.
The network serves as the foundation for other potential disease areas, as new datasets can easily be integrated. The potential of a federated network like LAISDAR has not gone unnoticed by the research community, as expansion of the network is in progress. Among them are two exciting opportunities: DRIVE*, a project aiming to “drive” comprehensive care to tuberculosis patients in sub-Saharan Africa and the BRIDGE Training Program, which aims to equip African researchers with advanced skills in health informatics and data sciences.
Furthermore, LAISDAR has inspired analogous projects, both within Africa and in low-to-middle-income countries (LMICs) elsewhere, that will make use of the solutions and strategies edenceHealth developed and deployed.
"LAISDAR Project has been a landmark achievement in Africa's journey toward data-driven healthcare. As we built the largest federated health data network on the continent, through BRIDGE NETWORK Project and OHDSI Africa Chapter, we witnessed firsthand how local ownership, open standards, and ethical data governance can transform fragmented systems into powerful engines for clinical insight and public health action. This project taught us that African institutions, when empowered with the right tools and collaborations, can lead the global frontier in digital health innovation. Looking ahead, LAISDAR project is not an end but a foundation—its architecture is now seeding broader applications in tuberculosis care, health informatics training, and pandemic preparedness in Rwanda and across the region. The lessons from LAISDAR Project point to a bold future: one where Africa doesn’t just catch up but sets new global standards for equitable, federated, and intelligent health data ecosystems."
Professor Dr. Marc Twagirumukiza, MD, PhD
LAISDAR Project Promotor
1 LAISDAR was aproject managed by the University of Rwanda alongside Rwanda Biomedical Center(RBC), University of Ghent and the Regional Alliance for SustainableDevelopment (RASD) in Rwanda. The project was funded by Canada’s InternationalDevelopment Research Centre (IDRC) and the Swedish International DevelopmentCooperation Agency (Sida), under the Global South AI4COVID Program. https://rbc.gov.rw/laisdar/
2 Nishimwe, A.,Ruranga, C., Musanabaganwa, C. et al. Leveraging artificialintelligence and data science techniques in harmonizing, sharing, accessing andanalyzing SARS-COV-2/COVID-19 data in Rwanda (LAISDAR Project): study designand rationale. BMC Med Inform Decis Mak 22, 214 (2022).https://doi.org/10.1186/s12911-022-01965-9