# Sharing EHR data of patients with rare diseases for research - The role of quality assessments in a national federated research data infrastructure

## TL;DR

This paper explores how data quality assessments can improve the use of electronic health records for rare disease research within a national data infrastructure.

## Contribution

The study introduces a novel data quality assessment tool and evaluates its effectiveness in identifying and resolving issues in FAIRified rare disease data.

## Key findings

- FHIR-DS had more data quality issues than EXP-DS, with 1988 distinct issues identified.
- 68% of the identified data quality issues in FHIR-DS were successfully resolved.
- The tool's runtime performance improved significantly with multi-core parallelization.

## Abstract

To obtain sufficient data for clinical research on Rare Diseases (RDs), the project “Collaboration on Rare Diseases” (CORD-MI) has connected 24 University Medical Centers (UMCs) in Germany. As part of the national Medical Informatics Initiative (MII), which aims at making the EHR data of all German UMCs FAIR and jointly usable for research, CORD-MI extends the methods and tools along the special requirements of RD data. Major challenges are insufficient coding of RDs and the high influence of individual data records on the results due to low cohort sizes. Hence, high data quality (DQ) of each data record is vital and exclusion due to quality issues may lead to insufficient sample sizes. DQ deficiencies induced by a FAIRification process may be corrected retrospectively; therefore their identification is of high relevance for RD research.

We used a self-developed DQ assessment tool to investigate the impact of the implemented FAIRification processes. Seven normalized DQ indicators – encompassing completeness, uniqueness, and plausibility – were assessed automatically across two datasets created from the same patient information system (PAS) but processed differently: claims data available in a FAIR format by MII standards (FHIR-DS), and a direct export (EXP-DS). The results were independently validated by domain experts, and measures to resolve the DQ issues were investigated. Furthermore, the tool’s FAIRness and computational performance were evaluated.

We found a higher number of DQ issues in FHIR-DS than in EXP-DS. Specifically, FHIR-DS contained 1988 distinct DQ issues: nine ambiguous RD cases, two missing data items, and 1977 completeness issues induced by the FAIRification processes. We could successfully resolve the majority (68%) of all identified DQ issues in FHIR-DS. The runtime performance could be reduced from 12.52 min to 0.12 min for FHIR-DS by multi-core parallelization. A comparison with current FAIR guidelines of the software showed compliance with 16 of 17 principles.

Completeness indicators are crucial for evaluating the quality of RD data, as most identified issues are related to the completeness dimension and largely stem from FAIRification processes. Our findings have shown the efficiency and effectiveness of our methodology in improving the quality and usefulness of FAIR data both retrospectively and prospectively.

Not applicable.

The online version contains supplementary material available at 10.1186/s12911-025-03153-x.