Contents

Datasets of relevance




Recipe Overview
Reading Time
15 minutes
Executable Code
No
Difficulty
Datasets of relevance
FAIRPlus logo
Recipe Type
Guidance
Audience
Principal Investigator, Data Manager, Terminology Manager, Data Scientist
Maturity Level & Indicator
[F+MM-1.1C] [F+MM-1.2C]
hover me Tooltip text
Cite me with FCB0XX

Main Objectives

The FAIR cookbook aims to provide hands-on, practical advice on how to deliver FAIR data through interactions with Innovative Medicine Initiative projects. These research projects, by nature often involve patient-centric information but dealing with real-world data and human-centric information, clinical data, in particular, is challenging. It most often mandates interacting with DACs, i.e. Data Access Committees, and undergoing a vetting process, which can be lengthy and convoluted. This can become a hindrance if the focus of the work is to deliver training on the computational methods available to deal with such data rather than data custody-related tasks, however important these are.

This FAIR cookbook recipe aims to provide a list of relevant resources belonging to the realm of clinical data so readers can, with the minimal hassle :

  • familiarize with the data types (for instance, how do Electronic Health records look like).

  • familiarize with the procedures to gain access to sensitive data.

  • obtain datasets with which to work and hone computational skills.

The recipe will cover two types of datasets:

  • real datasets such as the MIMIC-III dataset 2, which corresponds to actual medical notes data for which data access requests must be made but which are made available to computational scientists for research purposes.

  • synthetic datasets, which are available without restrictions since produced by computational methods and independent of any real patient. While handy, this type of data may come with a number of limitations prospective users need to be aware of.

Electronic Health Records: The MIMIC-III Critical Care Database

Electronics Medical Notes: The EBM NLP

  • Data Type: Electronic Medical Notes

  • Nature of the data: Synthetic Data

  • Description: A corpus containing 4,993 abstracts annotated with (P)articipants, (I)nterventions, and (O)utcomes.

Training labels are sourced from AMT workers and aggregated to reduce noise. Test labels are collected from medical professionals. 1

  • Data Type: Electronic Medical Notes

  • Nature of the data: Synthetic Data

  • Description: This test data set, generated by Lee Evans while working at LTS Computing LLC is an OMOP-CDM version 5 formatted collection of 1000 patient sample of CMS 2008-2010 Data Entrepreneurs’ Synthetic Public Use File (DE-SynPUF). The information held in the dataset corresponds to Synthetic patients & medicare claims/prescription data. he dataset can be downloaded in a compressed (zipped) format from LTS Computing LLC website: http://www.ltscomputingllc

  • Format: OMOP CDM v5.2.2

  • Availability: http://www.ltscomputingllc.com/downloads/

  • Purpose:

    • Public data to demo the OHDSI Ontologies

    • Benchmark performance

    • Developing & testing OHDSI tools

    • OHDSI tools training

  • License: LTS Computing LLC license

Synthean Electronic Health Records

One of the main bottlenecks for data miners is the lack of dataset availability of electronic health records, due to, as we saw it to HIPAA concerns. To bypass these roadblocks, several tools have been developed to generate synthetic datasets, free of any restrictions. Below, we provide information about one such tool.

https://github.com/synthetichealth/synthea/wiki

  • Data Type: Electronic Health Records

  • Nature of the data: Synthetic EHR

  • Description: Synthean, Synthetic Electronic Health Records generating software 3

  • Format: HL7 FHIR

  • Availability: https://synthetichealth.github.io/synthea/ https://synthea.mitre.org/downloads

  • Purpose:

    • Generation of synthetic training dataset for data mining and algorithm development

  • License: CC0 - “Public Domain Dedication”

Exemplar datasets available for download
Exemplar Synthetic EHR datasets available for download in HL7 FHIR

Fig. 11 Exemplar Synthetic EHR datasets available for download in HL7 FHIR

Synthetic Electronic Medical Notes: the OMOP CDMv5 Test Data

  • Data Type: Electronic Medical Notes

  • Nature of the data: Synthetic Data

  • Description: This test data set, generated by Lee Evans while working at LTS Computing LLC is an OMOP-CDM version 5 formatted collection of 1000 patient sample of CMS 2008-2010 Data Entrepreneurs’ Synthetic Public Use File (DE-SynPUF). The information held in the dataset corresponds to Synthetic patients & medicare claims/prescription data. he dataset can be downloaded in a compressed (zipped) format from LTS Computing LLC website: http://www.ltscomputingllc

  • Format: OMOP CDM v5.2.2

  • Availability: http://www.ltscomputingllc.com/downloads/

  • Purpose:

    • Public data to demo the OHDSI Ontologies

    • Benchmark performance

    • Developing & testing OHDSI tools

    • OHDSI tools training

  • License: LTS Computing LLC license

SynPUF 1000 person dataset in OMOP CDM v5.2.2 format:

This synthetic dataset corresponds to 1000 person composite dataset:

http://www.ltscomputingllc.com/wp-content/uploads/2018/08/synpuf1k_omop_cdm_5.2.2.zip

For more information about the OMOP Common Data Model, refer to the following:

CDM 5.2.2 DDL for the OHDSI supported DBMSs is available health-related

https://github.com/OHDSI/CommonDataModel/tree/v5.2.2

Clinical Trial Data in CDISC SDTM format:

  • Data Type: Clinical Trial Data

  • Nature of the data: Dummy Trial Data

  • Description: This is a sample study dataset containing CDISC SDTM formatted data files created originally by CDISC for demo purposes. This dataset can be used by anyone who is interested in CDISC SDTM formatted dataset. This version is used to demonstrate loading standard clinical datasets into PlatformTM: a data custodianship platform for translational research data. Live demo here: https://platformtm.cloud/app/#/projects/15/main

  • Format: CDISC SDTM

  • Availability: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/51B6NK

  • Purpose:

    • Benchmark performance

    • Developing & testing CDISC tools

    • CDISC SDTM tools training

  • License: CC0 - “Public Domain Dedication”

Observational Data in OMOP CDM format:

  • Data Type: Observational Data

  • Nature of the data: Dummy Trial Data

  • Description:

  • Format: OMOP CDM v5

  • Availability:

  • Purpose: *

  • License: CC0 - “Public Domain Dedication”

Conclusions

This content provides you with a set of resources to kick start your exploration of unstructured text in clinical context. These are useful resources for gaining familiarity with these data types. Remember to understand the data stewardship requirements that go along with handling real clinical data but also the limitations associated with some synthetic datasets.

References

Reference

1

Cdmv5 test data. URL: https://www.ohdsi.org/wp-content/uploads/2015/04/Lee_Evans_CDMV5_Test_Data_Presentation.pdf.

2(1,2)

A. E. Johnson, T. J. Pollard, L. Shen, L. W. Lehman, M. Feng, M. Ghassemi, B. Moody, P. Szolovits, L. A. Celi, and R. G. Mark. MIMIC-III, a freely accessible critical care database. Sci Data, 3:160035, May 2016.

3

J. Walonoski, M. Kramer, J. Nichols, A. Quina, C. Moesel, D. Hall, C. Duffett, K. Dube, T. Gallagher, and S. McLachlan. Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record. J Am Med Inform Assoc, 25(3):230–238, Mar 2018.

Authors

Authors

Name

ORCID

Affiliation

Type

ELIXIR Node

Contribution

Philippe Rocca-Serra

University of Oxford

Writing - Original Draft

Susanna-Assunta Sansone

University of Oxford

Writing - Review & Editing

Yojana Gaidya

Fraunhofer Institute

Writing - Review & Editing
Public knowledge graphs resources for life sciences Prioritization of projects for FAIRification