MedXN: Medication Extraction and Normalization for Clinical Text

Introduction

Medication Extraction and Normalization (MedXN, pronounced [med-eks-en]) is a Apache UIMA-based medication information extraction system that focuses on assigning the most specific RxNorm RxCUI to medication description. MedXN finds medication and its complete attributes and normalize them to the most specific RxNorm RxCUI using flexible matching, abbreviation expansion, inference, etc. MedXN uses externalized resources (ie, medication dictionary, attribute definitions, and regular expression attribute patterns) to allow a simple customization process for the needs of end users.

Prerequisites

• Java 1.8
• Apache Maven
• Apache Ant

If you do not have the prerequisites installed, you may need to download the binary files from their official download sites and add the unzipped bin directories to PATH. These steps are dependent on your OS.

Installation

MedXN is runnable out-of-the-box: simply download MedXN.zip corresponding to the latest release, extract, and follow the instructions for running

Compiling From Source Code

To make changes and re-compile medxn from source code, first clone this repo and enter the project directory as root:

git clone https://github.com/OHNLP/MedXN.git
cd MedXN
git checkout dist

To download MedTagger and Backbone dependencies from github packages, you will have to add the following to your maven settings.xml (typically located in ~/.m2/settings.xml)

 <servers>
        <server>
            <id>medtagger</id>
            <username>your_github_username</username>
            <password>your_github_access_token</password>
        </server>
        <server>
            <id>backbone-maven</id>
            <username>your_github_username</username>
            <password>your_github_access_token</password>
        </server>
    </servers>

where your github access token corresponds to an appropriate github token with the read:packages permissions

To build MedXN, build the jar file using:

mvn clean install

If it goes smoothly, you will see MedXN-{$version}-SNAPSHOT-shaded.jar under target.

To further build a distributable directory, use the ant script:

ant dist

Once finished, the dist directory should contain the required resources, scripts and the MedXN.jar to be distributed.

Run

Run in batch from CLI

The org.ohnlp.medxn.Main class provides a simple command line interface (CLI) to process a directory of input files (e.g. testdata) and write into a single output file (e.g. out.txt).

To run the org.ohnlp.medxn.Main class through CLI (under dist):

In Windows:

java -cp resources;MedXN.jar org.ohnlp.medxn.Main $YOUR_INPUT_DIR $YOUR_OUTPUT_FILE

In Unix:

java -cp resources:MedXN.jar org.ohnlp.medxn.Main $YOUR_INPUT_DIR $YOUR_OUTPUT_FILE

Note: the delimiter of classpath is different in Windows (;) from Unix (:)

Run via UIMA GUIs (CVD and CPE)

To execute MedXN for a collection of documents, make sure you are in the dist directory built previously and run runMedXNCVD.bat (runMedXNCVD.sh) or runMedXNCPE.bat (runMedXNCPE.sh) which will test processable analysis engines and collection processing engines.

In Windows:

java -Xms512M -Xmx2000M -cp resources;MedXN.jar org.apache.uima.tools.cvd.CVD
java -Xms512M -Xmx2000M -cp resources;MedXN.jar org.apache.uima.tools.cpm.CpmFrame

In Unix/Linux:

java -Xms512M -Xmx2000M -cp resources:MedXN.jar org.apache.uima.tools.cvd.CVD
java -Xms512M -Xmx2000M -cp resources:MedXN.jar org.apache.uima.tools.cpm.CpmFrame

It will fire up UIMA Cas Visual Debugger (CVD) or the collection processing engine (CPE) GUI.

To visualize a specific aggregate engine through CVD, go to load AE under the Run menu, choose $MedXNHOME/dist/resources/desc/medxndesc/aggregate_analysis_engine/MedXNAggregateTAE.xml

To process a collection of documents, go to the FILE menu and open the corresponding CPE descriptor file available in $MedXNHOME/dist/resources/desc/collection_processing_engine/MedXN_CPE.xml

Packaged files under dist

• resources/desc: example descriptors for Aggregate Analysis Engines and Collection Process Engine (CPE)
• testdata: test input data
• testdata_output: expected output in xmi format
• runMedXNCVD.bat (runMedXNCVD.sh): scripts for Cas Visualize Debugger (CVD)
• runMedXNCPE.bat (runMedXNCPE.sh): scripts for CPE

High-level Algorithm

Text: "Sulfasalazine [AZULFIDINE] 500-mg 2 tabs by mouth two times a day"

1. Medication Extraction

   Eg) Sulfasalazine [AZULFIDINE] RxCUI="9524::IN::202770::BN"

2. Attribute Extraction

   Eg) 500-mg (strength), 2 (dose), tabs (form), mouth (route), two times a day (frequency)

3. Medication & Attribute Association

   Eg) <Sulfasalazine [AZULFIDINE]> + <500-mg, 2, tabs, mouth, two times a day>

4. Convert to RxNorm Standard

   Eg) sulfasalazine <in>500 mg<st> oral tablet<df>azulfidine<bn>

5. Convert to RxCUI Representation

   Eg) 9524<in>500 mg<st>317541<df>202770<bn>

6. Normalize to Specific RxCUI

   Eg) Sulfasalazine 500 MG Oral Tablet [AZULFIDINE] RxCUI=208437::SBD

MedXN components

MedXN descriptors

Under dist/resources/desc/medxndesc/:

• Aggregate TAE: aggregate_analysis_engine/MedXNAggregateTAE.xml

• Collection Processing Engine: collection_processing_engine/MedXN_CPE.xml

• Primary Annotators

  • ACLookupDrugAE.xml: extracts medication name
  • MedAttrAE.xml: extract medication attributes
  • MedExtAE.xml: associates medication name and its attributes
  • MedNormAE.xml: normalizes medication information to RxNorm standard
  • ACLookupDrugNormAE.xml: maps medication information to a specific RxNorm name
  • MedNormRxCUIAE.xml: convert medication information to RxCUI representation
  • ACLookupRxCUIDrugNormAE.xml: maps RxCUI-represented medication information to a specific RxNorm name

• Cas Consumer

  • MedXNCC.xml: prints out results.
    • Parameters: OutputFile – output file path and name Delimiter – a delimiter of medication information in the output
    • Output format:

       filename|medication::b::e|medication Rxcui|strength::b::e|dose::b::e|form::b::e|route::b::e|frequency::b::e|duration::b::e|specific RxNorm name|specific RxCUI|sentence (b: begin offset, e: end offset) 
      

MedXN dictionary resources

Under dist/resources/medxnresources/lookup

• RxNorm_BNIN.alphanum.BnInPinMinSyn.txt: a dictionary for medication names compiled from RxNorm ingredient and brand name (ie, IN, PIM, MIN, BN, and manually compiled abbreviations). Also, includes any other medication variations that have the same RxCUI as the above medications.
  • Format: medication name (lower-cased, non-alphanumeric replaced with space, tokens are separated by tap)|RxCUI|RxNorm term type|RxNorm name
  • Example: aspirin|1191|IN|Aspirin
• RxNorm_Name.norm.txt: a dictionary of full medication descriptions complied from RxNorm SCDC, SCDF, SCD, SBDC, SBDF, SBD, and SY
  • Format: full medication description (lower-cased, [] removed, tokens are separated by tap)|RxCUI|RxNorm term type|RxNorm name
  • Example: aspirin 81 mg oral tablet|243670|SCD|Aspirin 81 MG Oral Tablet
• RxCUI.norm.txt: RxCUI representation of RxName.norm.txt – ie, medication name and dose form are replaced with RxCUI
  • Format: RxCUI representation|RxCUI|RxNorm term type|RxNorm name
  • Example: 1191 81 mg 317541|243670|SCD|Aspirin 81 MG Oral Tablet
• doseDict.txt: list of RxNorm dose forms and its RxCUI
  • Format: dose form (lower-cased)|RxCUI|RxNorm name
  • Example: oral tablet|317541|Oral Tablet
• falseMedDic.txt: list of potential false medication – ie, these are in RxNorm but potentially false drugs in clinical notes
  • Format: lower-cased medication
  • Example: today

Regular expressions file

Under dist/resources/medxnresources

regExPatterns.txt: contains medication attribute patterns written in Java regular expression (includes usage descriptions).