Development and validation of method for defining conditions using Chinese electronic medical record

Yuan Xu, Ning Li, Ming-Shan Lu, Robert P. Myers, Elijah Dixon, Robin Walker, Libo Sun, Xiaofei Zhao, Hude QuanBMC Medical Informatics Decis. Mak 2016 · DOI

TL;DR

This paper presents a novel method for extracting clinical information from Chinese Electronic Medical Records (EMRs) to define specific conditions, particularly liver diseases. The developed method demonstrates high validity, offering a crucial step towards leveraging the vast, yet often unstructured, data within Chinese EMRs for research and clinical applications.

What Problem Does This Paper Address?

The widespread adoption of Electronic Medical Records (EMRs) in China presents a massive opportunity for medical research and policy. However, a significant challenge lies in the fact that Chinese EMR systems are not fully standardized and often contain free-text or unstructured data. Unlike in Western countries, standardized coding systems like ICD are not consistently used. This lack of standardization, coupled with language differences and the absence of widely validated Natural Language Processing (NLP) tools for Chinese EMRs, makes it difficult to extract and utilize clinical information effectively. This paper addresses this gap by developing and validating a robust method for defining clinical conditions, specifically liver diseases, from Chinese EMRs, paving the way for broader data utilization.

Methods

The authors developed a three-step EMR data extraction method.

  1. Keyword Development: They created keywords (including Chinese words and punctuation) corresponding to different free-text entries in the EMR, using "AND"/"OR" operators. These keywords were designed to capture various ways a clinical condition might be described by physicians.
  2. Field Selection: They identified relevant sections and fields within the EMR for keyword searching. This involved considering the disease diagnostic pathway, and the availability, completeness, and temporality of information. Examples include diagnosis fields, pathology/radiology reports, and laboratory results.
  3. Logical Combination: The search results were logically combined, incorporating inclusion and exclusion criteria informed by clinical expertise. For liver diseases, this involved referencing national and international practice guidelines. The method also incorporated a framework for constructing definitions for other conditions.

The authors specifically mention that they did not use NLP due to the lack of established and validated tools for Chinese EMRs. Instead, they focused on a keyword-based approach combined with clinical expertise.

Datasets

The study utilized EMR data from YouAn Hospital, a large teaching hospital affiliated with Capital Medical University in Beijing, China, spanning from 2010 to 2014. This dataset comprised 85,524 EMRs. For validation, a random sample of 450 EMRs was selected.

Key Insights and Ideas

  • Bridging the Unstructured Data Gap: The paper demonstrates that a well-designed keyword-based extraction method, combined with clinical expertise, can effectively overcome the challenges posed by unstructured free-text data in Chinese EMRs, even without advanced NLP.
  • Clinical Expertise is Paramount: The success of the method hinges on the deep integration of clinical knowledge to define keywords, select relevant fields, and logically combine search results, reflecting real-world diagnostic pathways.
  • Adaptability of the Framework: The proposed framework for developing extraction methods is designed to be adaptable to other Chinese EMR systems, acknowledging that while specific EMR implementations may vary, the core components and data fields are generally similar across national standards.
  • High Validity Achieved: The developed definitions show a high degree of accuracy, suggesting that Chinese EMRs, despite their limitations, are a valuable source of clinical information when properly extracted.

Key Results

The study validated 40 EMR definitions for 6 liver diseases, 5 disease severity conditions, and 29 comorbidities and treatments.

  • Liver Diseases: Sensitivity ranged from 78.9% to 100.0%, and Positive Predictive Value (PPV) ranged from 82.1% to 100.0%.
  • Disease Severity Conditions: Sensitivity ranged from 91.0% to 100.0%, and PPV ranged from 79.2% to 100.0%.
  • Comorbidities and Treatments: Out of 29 definitions, 23 had sensitivity over 90.0%, and 25 had PPV over 80.0%.
  • Overall Performance: Specificity and Negative Predictive Value (NPV) for all 40 EMR definitions were over 90.0%.

Limitations and Future Work

The authors acknowledge several limitations:

  • Physician Review as Gold Standard: While considered adequate, the validity of the physician review itself as a gold standard is not explicitly assessed.
  • Low Prevalence Conditions: The validity of definitions for certain conditions (e.g., intrahepatic cholangiocarcinoma, pulmonary circulation disorders) could not be assessed due to low prevalence in the dataset.
  • Single-Center Study: The study was conducted at a single hospital specializing in liver disease, which may influence the prevalence-based statistics.

Future work should focus on:

  • Modification and Validation: Modifying the extraction method for application to other Chinese EMR systems and further validating it.
  • Broader Condition Coverage: Expanding the method to extract definitions for a wider range of conditions.
  • NLP Integration (Future): While not used in this study, future research could explore integrating NLP once robust tools for Chinese EMRs become available.

Summary updated

BibTeX
@article{xu2016development,
  author = {Yuan Xu and Ning Li and Ming-Shan Lu and Robert P. Myers and Elijah Dixon and Robin Walker and Libo Sun and Xiaofei Zhao and Hude Quan},
  title = {{Development and validation of method for defining conditions using Chinese electronic medical record}},
  journal = {BMC Medical Informatics Decis. Mak.},
  year = {2016},
  doi = {10.1186/s12911-016-0348-6},
}
Figure