A hybrid solution for extracting structured medical information from unstructured data in medical records via a double-reading/entry system

BMC Medical Informatics and Decision Making

Technical Advance

Lue Ping Zhao¹  Jiajia Hu²  Ligang Luo²  Boulin Hou²  Tianze Zhang²  Xiaozhe Wang²  Liping Li² 

[1]Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA, USA

[2]LinkDoc Inc, 8 Haidian Street, Block A, 8th Floor, Haidian District, Beijing, China

关键词: Big data;    Big medical data;    Clinical research;    Clinical decision support system;    Cloud-based system;    Double data entry;    Electronic medical record;    Health service research;    Structured data and unstructured data;   

DOI  :  10.1186/s12911-016-0357-5

 received in 2016-02-21, accepted in 2016-08-23,  发布年份 2016

来源: Springer

【 摘 要 】

BackgroundHealthcare providers generate a huge amount of biomedical data stored in either legacy system (paper-based) format or electronic medical records (EMR) around the world, which are collectively referred to as big biomedical data (BBD). To realize the promise of BBD for clinical use and research, it is an essential step to extract key data elements from unstructured medical records into patient-centered electronic health records with computable data elements. Our objective is to introduce a novel solution, known as a double-reading/entry system (DRESS), for extracting clinical data from unstructured medical records (MR) and creating a semi-structured electronic health record database, as well as to demonstrate its reproducibility empirically.MethodsUtilizing the modern cloud-based technologies, we have developed a comprehensive system that includes multiple subsystems, from capturing MRs in clinics, to securely transferring MRs, storing and managing cloud-based MRs, to facilitating both machine learning and manual reading, and to performing iterative quality control before committing the semi-structured data into the desired database. To evaluate the reproducibility of extracted medical data elements by DRESS, we conduct a blinded reproducibility study, with 100 MRs from patients who have undergone surgical treatment of lung cancer in China. The study uses Kappa statistic to measure concordance of discrete variables, and uses correlation coefficient to measure reproducibility of continuous variables.ResultsUsing the DRESS, we have demonstrated the feasibility of extracting clinical data from unstructured MRs to create semi-structured and patient-centered electronic health record database. The reproducibility study with 100 patient’s MRs has shown an overall high reproducibility of 98 %, and varies across six modules (pathology, Radio/chemo therapy, clinical examination, surgery information, medical image and general patient information).ConclusionsDRESS uses a double-reading, double-entry, and an independent adjudication, to manually curate structured data elements from unstructured clinical data. Further, through distributed computing strategies, DRESS protects data privacy by dividing MR data into de-identified modules. Finally, through internet-based computing cloud, DRESS enables many data specialists to work in a virtual environment to achieve the necessary scale of processing thousands MRs within days. This hybrid system represents probably a workable solution to solve the big medical data challenge.

【 授权许可】

CC BY   
© The Author(s). 2016