Skip to content
forked from vmenger/deduce

Deduce: de-identification method for Dutch medical text

License

Notifications You must be signed in to change notification settings

nedap/deduce-lydia

 
 

Repository files navigation

deduce

tests coverage build documentation pypi version pypi python versions pypi downloads license black

Installation - Versions - Getting Started - Documentation - Contributiong - Authors - License

Deduce 2.0.0 has been released! It includes a 10x speedup, and way more features for customizing and tailoring. Some small changes are needed to keep going from version 1, read more about it here: docs/migrating-to-v2

De-identify clinial text written in Dutch using deduce, a rule-based de-identification method for Dutch clinical text.

The development, principles and validation of deduce were initially described in Menger et al. (2017). De-identification of clinical text is needed for using text data for analysis, to comply with legal requirements and to protect the privacy of patients. By default, our rule-based method removes Protected Health Information (PHI) in the following categories:

  • Person names, including initials
  • Geographical locations smaller than a country
  • Names of institutions that are related to patient treatment
  • Dates (combinations of day, month and year)
  • Ages
  • BSN numbers
  • Identifiers (7+ digits without a specific format, e.g. patient identifiers, AGB, BIG)
  • Telephone numbers
  • E-mail addresses
  • URLs

If you use deduce, please cite the following paper:

Menger, V.J., Scheepers, F., van Wijk, L.M., Spruit, M. (2017). DEDUCE: A pattern matching method for automatic de-identification of Dutch medical text, Telematics and Informatics, 2017, ISSN 0736-5853

Installation

pip install deduce

Versions

For most cases the latest version is suitable, but some specific milestones are:

  • 2.0.0 - Major refactor, with speedups, many new options for customizing, functionally very similar to original
  • 1.0.8 - Small bugfixes compared to original release
  • 1.0.1 - Original release with Menger et al. (2017)

Detailed versioning information is accessible in the changelog.

Getting started

The basic way to use deduce, is to pass text to the deidentify method of a Deduce object:

from deduce import Deduce

deduce = Deduce()

text = (
    "betreft: Jan Jansen, bsn 111222333, patnr 000334433. De patient J. Jansen is 64 jaar oud en woonachtig in "
    "Utrecht. Hij werd op 10 oktober 2018 door arts Peter de Visser ontslagen van de kliniek van het UMCU. "
    "Voor nazorg kan hij worden bereikt via j.JNSEN.123@gmail.com of (06)12345678."
)

doc = deduce.deidentify(text)

The output is available in the Document object:

from pprint import pprint

pprint(doc.annotations)

AnnotationSet({
    Annotation(text="(06)12345678", start_char=272, end_char=284, tag="telefoonnummer"),
    Annotation(text="111222333", start_char=25, end_char=34, tag="bsn"),
    Annotation(text="Peter de Visser", start_char=153, end_char=168, tag="persoon"),
    Annotation(text="j.JNSEN.123@gmail.com", start_char=247, end_char=268, tag="email"),
    Annotation(text="patient J. Jansen", start_char=56, end_char=73, tag="patient"),
    Annotation(text="Jan Jansen", start_char=9, end_char=19, tag="patient"),
    Annotation(text="10 oktober 2018", start_char=127, end_char=142, tag="datum"),
    Annotation(text="64", start_char=77, end_char=79, tag="leeftijd"),
    Annotation(text="000334433", start_char=42, end_char=51, tag="id"),
    Annotation(text="Utrecht", start_char=106, end_char=113, tag="locatie"),
    Annotation(text="UMCU", start_char=202, end_char=206, tag="instelling"),
})

print(doc.deidentified_text)

"""betreft: <PERSOON-1>, bsn <BSN-1>, patnr <ID-1>. De <PERSOON-1> is <LEEFTIJD-1> jaar oud en woonachtig in 
<LOCATIE-1>. Hij werd op <DATUM-1> door arts <PERSOON-2> ontslagen van de kliniek van het <INSTELLING-1>. 
Voor nazorg kan hij worden bereikt via <EMAIL-1> of <TELEFOONNUMMER-1>."""

Aditionally, if the names of the patient are known, they may be added as metadata, where they will be picked up by deduce:

from deduce.person import Person

patient = Person(first_names=["Jan"], initials="JJ", surname="Jansen")
doc = deduce.deidentify(text, metadata={'patient': patient})

print (doc.deidentified_text)

"""betreft: <PATIENT>, bsn <BSN-1>, patnr <ID-1>. De <PATIENT> is <LEEFTIJD-1> jaar oud en woonachtig in 
<LOCATIE-1>. Hij werd op <DATUM-1> door arts <PERSOON-2> ontslagen van de kliniek van het <INSTELLING-1>. 
Voor nazorg kan hij worden bereikt via <EMAIL-1> of <TELEFOONNUMMER-1>."""

As you can see, adding known names keeps references to <PATIENT> in text. It also increases recall, as not all known names are contained in the lookup lists.

Documentation

A more extensive tutorial on using, configuring and modifying deduce is available at: docs/tutorial

Basic documentation and API are available at: docs

Contributing

For setting up the dev environment and contributing guidelines, see: docs/contributing

Authors

  • Vincent Menger - Initial work
  • Jonathan de Bruin - Code review
  • Pablo Mosteiro - Bug fixes, structured annotations

License

This project is licensed under the GNU LGPLv3 license - see the LICENSE.md file for details

About

Deduce: de-identification method for Dutch medical text

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.7%
  • Makefile 0.3%