Sentinel | Video | A Practical Framework for Data Quality Excellence in Healthcare Datasets
Under the evolving European Health Data Space (EHDS) guidelines, the quality and reliability of electronic health record (EHR) data are critical for advancing AI-driven decision tools and clinical research. LynxCare introduces Sentinel — a robust, OMOP Common Data Model-based system designed to continuously measure, benchmark, and improve EHR-derived datasets for regulatory compliance, collaborative studies, and translational research.
1. Introduction
Ensuring high-quality data is fundamental to successfully develop machine learning and AI-driven healthcare innovations. Recent policy developments, such as the proposed European Health Data Space (EHDS) [1], underscore the need for reliable data exchange and usage across EU member states. However, EHRs often contain missing values, erroneous measurements, and non-standard data types, impeding effective research and clinical decision-making.
The OMOP Common Data Model (CDM) [2] provides a standardized structure to harmonize disparate data sources for large-scale analytics. Yet, ensuring that data conform to the OMOP standards and remain consistent over time requires a systematic approach to quality control. LynxCare presents Sentinel, a continuous data quality monitoring framework, which evaluates datasets on key metrics—completeness, plausibility, conformance, and cross-hospital benchmarking—by building on Kahn’s data quality framework [3].
2. Methods & Results
2.1 Overall Architecture
LynxCare applies five quality control steps that span the entire data value chain, from source to database to insights. Sentinel evaluates database quality continuously using KPIs based on the Kahn data quality framework, including Completeness, Plausibility and Conformance. A cross-hospital benchmarking score and a weighted overall database quality score have also been added.
2.2 Quality Metrics & Examples
1. Completeness
Reflects the presence of data where it is expected.
Example: Value availability check: The percentage of patients with a recorded TTR-gene mutation status.
2. Plausibility
Examines if data is logically consistent with itself and with common knowledge or trusted external sources.
Example: Value plausibility check: Flagging impossible or highly unlikely entries (possibly not relevant to the patient) (e.g., weight = 5000 kg).
3. Conformance
Measures data compliance with formatting, relational, or computational definitions as prescribed by the OMOP CDM guidelines.
Example: Data type check: Ensuring columns match expected data types (numeric, string, date, etc.).
4. Benchmarking
Evaluates the plausibility of Real-World Data by examining deviations from the multi-center average.
Example: Cross-hospital match: Identifies positive deviations (e.g., a higher capture rate for chronic kidney disease in one hospital vs. in the median hospital database within the LynxCare network).
5. Overall Quality Score
Aggregates weighted KPIs into a single metric indicating the dataset’s readiness for AI research and regulatory compliance.
3. Discussion
The Sentinel framework demonstrates how a continuous, multi-faceted data quality approach can uphold the integrity of clinical datasets, meeting both regulatory requirements and enabling advanced analytics. By systematically applying Kahn’s data quality framework and harnessing the OMOP CDM, Sentinel ensures datasets are research-ready—supporting real-world evidence (RWE) studies, AI model training, and improved patient outcomes.
Given the heightened need for trustworthy data under EHDS, hospitals adopting Sentinel can readily validate data quality, benchmark performance across institutions, and refine data capture processes at the source. This cyclical improvement underpins the success of large-scale AI and predictive modeling efforts—tools vital for personalized medicine, resource allocation, and population health management.
4. Conclusion
Sentinel is poised to accelerate AI-driven healthcare by providing a transparent, continuous, and standardized data quality assessment mechanism. Researchers, clinicians, and healthcare administrators can rely on the insights from data quality-checked datasets to make informed decisions, optimize patient care pathways, and advance collaborative research within and beyond the European Health Data Space.
Key Takeaways
• High-quality data is crucial for AI development and clinical research excellence.
• Continuous monitoring using a standardized framework ensures integrity and trustworthiness.
• Cross-hospital benchmarking fosters collaboration and promotes best practices.
• Aligning with EHDS requirements facilitates compliance and paves the way for innovative multi-site research.
References
[1] https://health.ec.europa.eu/ehealth-digital-health-and-care/european-health-data-space_en
[2] https://www.ohdsi.org/data-standardization/
[3] Kahn MG, Raebel MA, Glanz JM, Raghavan R, Jackson KL, et al. (2016). A Pragmatic Framework for Single-site and Multisite Data Quality Assessment in Electronic Health Record-based Clinical Research. eGEMs (Generating Evidence & Methods to improve patient outcomes), 4(1): 1277.
1. Introduction
Ensuring high-quality data is fundamental to successfully develop machine learning and AI-driven healthcare innovations. Recent policy developments, such as the proposed European Health Data Space (EHDS) [1], underscore the need for reliable data exchange and usage across EU member states. However, EHRs often contain missing values, erroneous measurements, and non-standard data types, impeding effective research and clinical decision-making.
The OMOP Common Data Model (CDM) [2] provides a standardized structure to harmonize disparate data sources for large-scale analytics. Yet, ensuring that data conform to the OMOP standards and remain consistent over time requires a systematic approach to quality control. LynxCare presents Sentinel, a continuous data quality monitoring framework, which evaluates datasets on key metrics—completeness, plausibility, conformance, and cross-hospital benchmarking—by building on Kahn’s data quality framework [3].
2. Methods & Results
2.1 Overall Architecture
LynxCare applies five quality control steps that span the entire data value chain, from source to database to insights. Sentinel evaluates database quality continuously using KPIs based on the Kahn data quality framework, including Completeness, Plausibility and Conformance. A cross-hospital benchmarking score and a weighted overall database quality score have also been added.
2.2 Quality Metrics & Examples
1. Completeness
Reflects the presence of data where it is expected.
Example: Value availability check: The percentage of patients with a recorded TTR-gene mutation status.
2. Plausibility
Examines if data is logically consistent with itself and with common knowledge or trusted external sources.
Example: Value plausibility check: Flagging impossible or highly unlikely entries (possibly not relevant to the patient) (e.g., weight = 5000 kg).
3. Conformance
Measures data compliance with formatting, relational, or computational definitions as prescribed by the OMOP CDM guidelines.
Example: Data type check: Ensuring columns match expected data types (numeric, string, date, etc.).
4. Benchmarking
Evaluates the plausibility of Real-World Data by examining deviations from the multi-center average.
Example: Cross-hospital match: Identifies positive deviations (e.g., a higher capture rate for chronic kidney disease in one hospital vs. in the median hospital database within the LynxCare network).
5. Overall Quality Score
Aggregates weighted KPIs into a single metric indicating the dataset’s readiness for AI research and regulatory compliance.
3. Discussion
The Sentinel framework demonstrates how a continuous, multi-faceted data quality approach can uphold the integrity of clinical datasets, meeting both regulatory requirements and enabling advanced analytics. By systematically applying Kahn’s data quality framework and harnessing the OMOP CDM, Sentinel ensures datasets are research-ready—supporting real-world evidence (RWE) studies, AI model training, and improved patient outcomes.
Given the heightened need for trustworthy data under EHDS, hospitals adopting Sentinel can readily validate data quality, benchmark performance across institutions, and refine data capture processes at the source. This cyclical improvement underpins the success of large-scale AI and predictive modeling efforts—tools vital for personalized medicine, resource allocation, and population health management.
4. Conclusion
Sentinel is poised to accelerate AI-driven healthcare by providing a transparent, continuous, and standardized data quality assessment mechanism. Researchers, clinicians, and healthcare administrators can rely on the insights from data quality-checked datasets to make informed decisions, optimize patient care pathways, and advance collaborative research within and beyond the European Health Data Space.
Key Takeaways
• High-quality data is crucial for AI development and clinical research excellence.
• Continuous monitoring using a standardized framework ensures integrity and trustworthiness.
• Cross-hospital benchmarking fosters collaboration and promotes best practices.
• Aligning with EHDS requirements facilitates compliance and paves the way for innovative multi-site research.
References
[1] https://health.ec.europa.eu/ehealth-digital-health-and-care/european-health-data-space_en
[2] https://www.ohdsi.org/data-standardization/
[3] Kahn MG, Raebel MA, Glanz JM, Raghavan R, Jackson KL, et al. (2016). A Pragmatic Framework for Single-site and Multisite Data Quality Assessment in Electronic Health Record-based Clinical Research. eGEMs (Generating Evidence & Methods to improve patient outcomes), 4(1): 1277.
