Grand Duchy of Luxembourg: a case study of a national master patient index in production since five years

Raffaella Vaccaroli, Frédéric Markus, Samuel Danhardt, Heiko Zimmermann, François Wisniewski, Pascale Lucas, Hervé BargeBMC Medical Informatics Decis. Mak 2020 · DOI

TL;DR

This paper details the successful implementation and operation of a national Master Patient Index (MPI) in Luxembourg, a multicultural country with a complex healthcare landscape. The MPI has significantly improved patient identification accuracy and safety by creating a unified view of patient data across various healthcare systems, leading to better interoperability and a more efficient national electronic health record.

What Problem Does This Paper Address?

In healthcare, accurately identifying patients is paramount for ensuring safe and effective treatment. Each interaction with a healthcare provider can generate multiple local identifiers across different information systems. This fragmentation poses a significant risk of misidentification, leading to potentially dangerous medical errors. Luxembourg, with its diverse population and cross-border workforce, faced the challenge of creating a unified, reliable patient identification system to support its national eHealth platform and electronic health record (EHR) system, known as DSP (Dossier de Soins Partagé). Before 2014, Luxembourg lacked a national MPI, necessitating a solution to consolidate patient identities and facilitate interoperability.

Methods

The authors describe a multi-step approach to establishing the national MPI. This involved:

  • Solution Selection: Evaluating available MPI solutions and selecting "Ideopass" (now IdeoIdentity) based on its ability to act as an identity server and manage patient identities centrally and locally.
  • Implementation: Integrating the MPI into the national eHealth platform and connecting it to the national healthcare system.
  • Policy Definition: Defining and implementing identity management policies in parallel with the technical implementation.

The MPI operates on a hierarchical federated identity management model. This model uses a central database of identity profiles, each assigned a unique Federated Identifier (FID). Local identifiers from various healthcare providers are reconciled and linked to these central FIDs. The system employs a deterministic matching algorithm that compares strict identity traits (like name, birth date, sex) and the national social security number (SSN) to match incoming data with existing profiles. The paper also details various workflows for identity reconciliation, including longitudinal (identity-producing domains) and transversal (identity-consuming domains) reconciliations, and how modifications are handled based on the source of the data. The system adheres to Integrating the Healthcare Enterprise (IHE) profiles (e.g., IHE-PAM, IHE-PDQ, IHE-XCPD) to ensure interoperability.

Datasets

The paper does not explicitly mention specific datasets used for training or evaluation in the traditional machine learning sense. Instead, the MPI is populated and updated using:

  • CCSS (Centre Commun de la Sécurité Sociale) data: This was the source for the initial injection of patient identity profiles.
  • RNPP (Registre National des Personnes Physiques) data: This national register provides and updates identification data.

The "dataset" in this context is the aggregated patient demographic and identification information managed by these national entities.

Key Insights and Ideas

  • Multiculturalism as a Challenge and Driver: Luxembourg's diverse population, with multiple languages and spelling habits, significantly increases the risk of identity anomalies. This heterogeneity necessitated a robust system that could handle variations in data entry.
  • The Importance of a Unified Identifier: The creation of the FID, a unique identifier specifically for healthcare, is crucial for reconciling disparate local identifiers and providing a single point of truth for patient identity.
  • Federated Model for Scalability: The hierarchical federated model allows for decentralized data input while maintaining central control and reconciliation, making the system scalable and adaptable to various healthcare providers.
  • Deterministic Matching is Sufficient (for now): While probabilistic matching is an area for future exploration, the paper demonstrates that a well-configured deterministic matching algorithm, combined with strict data quality checks, can achieve high accuracy in patient identification.
  • Continuous Improvement is Key: The MPI has undergone continuous updates and improvements since its inception in 2014, adapting to legal, technical, and operational requirements.

Key Results

  • High Coverage: As of December 31, 2018, the MPI contained 2,418,336 identity profiles, covering almost the entirety of Luxembourgish residents (95.2%) and affiliated cross-border workers.
  • Increased Usage: The usage rate of identities by external applications has steadily increased, representing 70% of identity reconciliations by the latest update.
  • Low Rejection Rate: The evaluation of data quality showed a low rejection rate of 0.2%, indicating high quality and completeness of identity profiles.
  • Reduced Anomalies: The system has significantly improved patient identification, leading to enhanced safety during medical journeys.
  • Decreased Duplicates: The analysis shows a trend of decreasing potential and merged duplicates over the years, with specific error types (e.g., spelling mistakes, use of marital names) identified as primary causes.

Limitations and Future Work

  • Deterministic Matching Limitations: The paper acknowledges that the current deterministic matching algorithm does not employ probabilistic matching, which could be a future area of development for handling more complex cases.
  • Handling of "Name" Trait: The MPI application currently rejects identities lacking specific strict traits, such as "Name," which is not universally defined or required in all countries. Cooperation with the software provider is ongoing to address this.
  • Pseudonymization for Research: The paper mentions the development of a pseudonymization service (SPS) for secondary use of patient data in research, indicating a future direction for enabling ethical data sharing.
  • Adaptation to Evolving Needs: The authors emphasize that the MPI is designed to be flexible and will continue to be adapted to meet future national and international standards and requirements.
  • Further Research on Errors: The paper highlights the need for continued statistical analysis of mistakes leading to identity issues to further optimize training and algorithms.

Summary updated

BibTeX
@article{vaccaroli2020grand,
  author = {Raffaella Vaccaroli and Frédéric Markus and Samuel Danhardt and Heiko Zimmermann and François Wisniewski and Pascale Lucas and Hervé Barge},
  title = {{Grand Duchy of Luxembourg: a case study of a national master patient index in production since five years}},
  journal = {BMC Medical Informatics Decis. Mak.},
  year = {2020},
  doi = {10.1186/s12911-020-01178-y},
}
Figure