Speakers
Description
Douala General Hospital is a first-class hospital in Cameroon where we meet a
multidisciplinary medical team treating several thousand patients each year. This hospital hosts
numerous patient records that may be useful for public health research. However, majority of
these records are paper-based, hence limiting their exploitation. For some cases, particularly
the pulmonology department, the data of hospitalized patients are recorded in heterogeneous
datasheets mainly collected for research purposes without standardization or uniform structure.
This data system also limits the exploitation for clinical research, care management and
decision-making. Furthermore, the lack of standardization limited the integration of data within
broader health systems, thus hindering its secure and reusable sharing.
To address this challenge, we undertook the implementation of a complete ETL pipeline
aligned with the OMOP Common Data Model (CDM) version 5.4, an internationally
recognized framework for standardizing health data. Our objective was to transform,
standardize, and integrate patient data from the pulmonology department of this hospital into a
database compliant with the FAIR (Findable, Accessible, Interoperable, Reusable) principles
to facilitate their reuse for research, clinical management, and patient monitoring. This
approach aimed to improve the quality of hospital data, strengthen its interoperability with
other systems, and lay the foundation for advanced use in data science for healthcare.
We made use of patient level dataset on Respiratory illness, which included more than 120
variables covering a wide range of clinical and administrative variables such as
sociodemographic data, medical history, clinical signs and symptoms, laboratory results, final
and secondary diagnoses, and medical observations, as well as information on hospital stays.
This data was extracted from datasheets, medical records, and papers, presenting varied forms,
heterogeneous levels of completeness, and missing observations.
To standardize this data, we made use of a number of OHDSI tools such as: WhiteRabbit for
data profiling, USAGI for vocabulary mapping from our source vocabularies to OMOP
standardized concepts/vocabularies and Rabbit in a Hat for data mapping of the source tables
to the standard OMOP CDM tables, including: Person, drug_exposure, Measurement,
visit_occurrence, condition_occurrence, and observation. The concepts used in the mappings
were derived from SNOMED, LOINC, and RxNORM vocabularies, while integrating
adaptations specific to the local context.
The ETL pipeline of this data was based on SQL skeleton files exported from Rabbit in Hat
after the data mapping. These scripts were then customized using pgAdmin interface for
PostgreSQL. After creating the OMOP tables using scripts from OHDSI's GitHub, we loaded
our transformed data into the OMOP PostgreSQL databases, structuring them in accordance
with the model.
To ensure the quality and compliance of our standardized database, we used the Achilles tool
(developed in R), which automatically checked the completeness, conformance, and
plausibility of the transformed data. This tool achieved an overall data quality score of 97%,
thus attesting to the reliability of the ETL pipeline and the robustness of the database.
Standardizing Respiratory illness data to OMOP CDM in the African context is a novel and
promising field that will set pace for collaboration, data sharing and interoperability across
health systems. This work made it possible to establish a standardized database that meets all
FAIR requirements and is ready to be used for analysis, clinical research, and data science, thus
opening up a perspective for decision-makers who can benefit from data-driven decisions to
improve pulmonology care practices in Cameroon.
