JMIR Publications

ABSTRACT

Background:

The rapid growth of clinical data, driven by digital technologies and high-resolution sensors, presents significant challenges for healthcare organisations aiming to support advanced AI research and improve patient care. Traditional data management approaches may struggle to handle the large, diverse and rapidly updating datasets prevalent in modern clinical environments.

Objective:

This study compares three clinical data management architectures - clinical data warehouses (cDWH), clinical data lakes (cDL) and clinical data lakehouses (cDLH) - by analysing their performance using the FAIR principles and the Big Data 5 Vs (Volume, Variety, Velocity, Veracity, Value). The aim is to provide guidance on selecting an architecture that balances robust data governance with the flexibility required for advanced analytics.

Methods:

We developed a comprehensive analysis framework that integrates aspects of data governance with technical performance criteria. A rapid literature review was conducted to synthesise evidence from multiple studies, focusing on how each architecture manages large, heterogeneous and dynamically updating clinical data. The review assessed key dimensions such as scalability, real-time processing capabilities, metadata consistency, and the technical expertise required for implementation and maintenance.

Results:

The results show that cDWHs offer strong data governance, stability and structured reporting, making them well suited for environments that require strict compliance and reliable analysis. However, they are limited in terms of real-time processing and scalability. In contrast, cDLs offer greater flexibility and cost-effective scalability for managing heterogeneous data types, although they may suffer from inconsistent metadata management and challenges in maintaining data quality. cDLHs combine the strengths of both approaches by supporting real-time data ingestion and structured querying; however, their hybrid nature requires high technical expertise and involves complex integration efforts.

Conclusions:

The optimal data management architecture for clinical applications depends on an organisation's specific needs, available resources, and strategic goals. Healthcare institutions need to weigh the trade-offs between robust data governance, operational flexibility and scalability to build future-proof infrastructures that support both clinical operations and AI research. Further research should focus on simplifying the complexity of hybrid models and improving the integration of clinical standards to improve overall system reliability and ease of implementation.