Establishing Medical Intelligence—Leveraging Fast Healthcare Interoperability Resources to Improve Clinical Management: Retrospective Cohort and Clinical Implementation Study

Alexander Brehmer, Christopher Martin Sauer, J Rodriguez, K Herrmann, Moon Kim, Julius Keyl, Fin Hendrik Bahnsen, Benedikt Frank, Martin Köhrmann, Tienush Rassaf, Amir A. Mahabadi, Boris Hadaschik, Christopher Darr, Susanne Tan, Jan Buer, Thorsten Brenner, Hans Christian Reinhardt, Felix Nensa, Michael Gertz, Jan Egger, Jens KleesiekJournal of Medical Internet Research 2024 · DOI

TL;DR

This paper introduces "medical intelligence," a framework for leveraging Fast Healthcare Interoperability Resources (FHIR) to extract actionable insights from real-world clinical data. By developing a novel Python package (FHIRPACK) and creating intuitive dashboards, the researchers demonstrate how this approach can enhance clinical decision-making and improve patient care across multiple medical disciplines.

What Problem Does This Paper Address?

The healthcare industry is awash in data, but effectively using it for clinical decision-making remains a significant challenge. Electronic Health Records (EHRs) are often heterogeneous, multimodal, and uncurated, making it difficult to extract meaningful information. While standards like FHIR aim to improve interoperability, their complexity can hinder widespread adoption and application, especially for extracting and analyzing granular data. This gap between data availability and actionable insights limits the potential for improving healthcare quality and patient outcomes.

Methods

The core of this work lies in the development of FHIRPACK, an open-source Python package designed to streamline the process of accessing, cleaning, linking, and transforming FHIR data. This toolkit enables the creation of custom workflows and the generation of various data representations, including those suitable for powerful data analysis libraries like pandas.

The researchers then applied FHIRPACK to a large FHIR database from the University Hospital Essen. They developed a standardized data pipeline to extract relevant information for five common, high-burden clinical use cases: myocardial infarction (MI), stroke, diabetes, sepsis, and prostate cancer (PC). For each use case, they identified patient cohorts using ICD-10 codes and extracted specific patient information from FHIR resources, including laboratory values, procedures, diagnostic reports, and medication administrations. The insights derived from this analysis were then presented through browser-based dashboards, designed to be intuitive for healthcare professionals at the point of care (PoC).

Datasets

The primary dataset used in this study is the University Hospital Essen's FHIR database, which contains over 2 billion resources and represents data from over 1.4 million unique patients. The analysis focused on data from the calendar year 2022 for most use cases, with an extended period for prostate cancer due to its specific treatment and monitoring requirements. No new benchmarks were introduced; instead, the study leveraged existing real-world clinical data.

Key Insights and Ideas

  • "Medical Intelligence" as a Paradigm Shift: The paper proposes "medical intelligence" as a concept analogous to business intelligence, emphasizing the use of data analytics to support clinical decisions and improve healthcare operations.
  • FHIRPACK as a Crucial Enabler: The development of FHIRPACK is highlighted as a key innovation, addressing the technical challenges of accessing and processing FHIR data, making it more accessible for clinicians and data scientists.
  • Dashboard-Driven Clinical Insights: The creation of intuitive, browser-based dashboards is crucial for translating complex data analysis into actionable information for healthcare providers at the point of care.
  • Uncovering Gaps in Standard Coding: The study revealed significant discrepancies between biochemically identified conditions (e.g., dysglycemia) and those captured by ICD-10 coding, suggesting that relying solely on coding can lead to missed diagnoses and opportunities for intervention.
  • Identifying Opportunities for Guideline Adherence: The analysis pinpointed specific patient groups who were not receiving guideline-recommended therapies, such as statins for MI patients or appropriate antibiotic regimens for sepsis, indicating areas for quality improvement.

Key Results

  • Large-Scale Analysis: Over 1.3 million patient encounters were analyzed in 2022.
  • Myocardial Infarction (MI): 72.7% of STEMI patients received guideline-recommended therapy (statin and/or antiplatelet).
  • Stroke: Out of 1277 patients, 165 received thrombolysis and 108 received thrombectomy. The study also analyzed time-to-intervention metrics against early neurological improvement.
  • Diabetes: Of 35,494 unique patients, 39% had dysglycemic findings. Crucially, 55.8% of these patients with biochemically proven dysglycemia were not identified by ICD-10 coding alone.
  • Sepsis: Staphylococcus epidermidis was the most frequently isolated pathogen (28.9%), and piperacillin/tazobactam was the most prescribed antibiotic (37.2%). The study also provided patient-level insights into inflammatory markers and antibiotic use.
  • Prostate Cancer (PC): The study identified 3 patients with PSA persistence or biochemical recurrence after treatment, highlighting the utility of longitudinal PSA monitoring.

Limitations and Future Work

  • Descriptive Nature of Results: The authors acknowledge that the current results are primarily descriptive. They suggest that integrating predictive and artificial intelligence algorithms could further enhance the framework.
  • Challenges in FHIR Implementation: Despite FHIR's standardization, challenges remain regarding the precise content and structure of data, the complexity of the standard itself, and the need for joint technical and medical expertise for implementation.
  • Data Quality: The accuracy and completeness of data exchanged via FHIR are dependent on the quality of data entered into primary systems, which can be a limitation.
  • Prospective Impact: While the study demonstrates the technical feasibility and potential benefits, a prospective impact on clinical decision-making remains to be demonstrated.
  • Data Availability: The analyzed datasets are not publicly available due to German legal codes but can be provided upon reasonable request.

Future work could focus on developing more sophisticated AI models, addressing the remaining implementation challenges, and conducting prospective studies to quantify the impact of medical intelligence on patient outcomes.

Summary updated

Abstract

Leveraging FHIR data through large-scale analytics can enhance health care quality and improve patient outcomes across 5 clinical specialties. We identified (1) patients with sepsis requiring less broad antibiotic therapy, (2) patients with myocardial infarction who could benefit from statin and antiplatelet therapy, (3) patients who had a stroke with longer than recommended times to intervention, (4) patients with hyperglycemia who could benefit from specialist referral, and (5) patients with PC with early increases in cancer markers.
BibTeX
@article{brehmer2024establishing,
  author = {Alexander Brehmer and Christopher Martin Sauer and J Rodriguez and K Herrmann and Moon Kim and Julius Keyl and Fin Hendrik Bahnsen and Benedikt Frank and Martin Köhrmann and Tienush Rassaf and Amir A. Mahabadi and Boris Hadaschik and Christopher Darr and Ken Herrmann and Susanne Tan and Jan Buer and Thorsten Brenner and Hans Christian Reinhardt and Felix Nensa and Michael Gertz and Jan Egger and Jens Kleesiek},
  title = {{Establishing Medical Intelligence—Leveraging Fast Healthcare Interoperability Resources to Improve Clinical Management: Retrospective Cohort and Clinical Implementation Study}},
  journal = {Journal of Medical Internet Research},
  year = {2024},
  doi = {10.2196/55148},
}
Figure