adepedia 2.0-integration of normalized adverse drug events (ades)

7/28/2019 ADEpedia 2.0-Integration of Normalized Adverse Drug Events (ADEs)

1/23

2013 MFMER | slide-1

ADEpedia 2.0: Integration of

Normalized Adverse Drug Events(ADEs) Knowledge from the UMLS

Guoqian Jiang, MD, PhD

Mayo Clinic2013 AMIA Summit on Clinical Research InformaticsMarch 22, 2013


2/23


Acknowledgements

Co-authorsHongfang Liu, PhDHarold R. SolbrigChristopher G. Chute, MD. Dr. PH

This work was supported in part by the SHARP Area 4: Secondary Use of EHR

Data (90TR000201)


3/23


Introduction

Adverse drug events (ADEs) are a well-recognizedcause of patient morbidity and increased healthcare costs in the United States.

Traditionally, spontaneous reporting is used as themain source for drug safety surveillance.

Notably, the US Food and Drug Administration(FDA) uses an adverse event reporting system(AERS) to monitor for new adverse events and

medication errors that might occur with allapproved drug and therapeutic biologicproducts


4/23


Introduction

To facilitate the ADE reporting and detection,there is emerging interest in secondary use ofclinical data from the electronic medical records(EMRs).

A fundamental challenge is that the communitylacks a publicly-available, standardized ADEknowledge base that encodes known ADE

information for drug surveillance.


5/23


Introduction

The known ADE information is very useful for avoidingover-alerting of ADE signals detected by data miningalgorithms, and reducing pharmacovigilance study noiselevels, so that only novel signals are considered for

further exploration.ADEpedia: a standardized knowledge base of ADEs for

drug safety surveillance

FDA Structured Product Labeling (SPL), FDA Adverse Event Reporting System (AERS) and the Unified Medical Language System (UMLS)

http://adepedia.org


6/23


Introduction

The Unified Medical Language System (UMLS),developed by the National Library of Medicine(NLM), intends to promote creation of moreeffective and interoperable biomedicalinformation system and services [12].

A systematic review and organization of knownADE knowledge from the UMLS would be a

good starting point to facilitate the integration ofknown ADE knowledge and ultimately form acomprehensive ADE knowledge base.


7/232013 MFMER | slide-7

Co-occurrence

Co-occurrence is one of essential statistics that hasbeen widely used in ADE text mining and knowledgeacquisition from biomedical and clinical documents.

Wright et al. used co-occurrence-based associationrule mining technique to identify a large number ofclinical accurate associations that may be useful for

identifying probable gaps in the problem list. Liu Y, et al demonstrated an approach that

statistically significant co-occurrence of drug-diseasementions in the clinical notes can be used to detect

ADE signals .



The objectives of this study

To extract and normalize ADE knowledge fromthe UMLS and integrate the ADE knowledgewith the ADEpedia; and

To enrich the knowledge base with the drug-disorder co-occurrence data from a large-scaleEMR system.



Knowledge Resources for ADE Integration

1) UMLS 2011AB: We downloaded and installed the version of UMLS

2011AB.

2) Semantic groups: We used 2 UMLS semantic groups Chemicals &

Drugs (CHEM) and Disorders (DISO), and theircorresponding semantic types.

3) ADE data from the ADEpedia. In the knowledge base, the medication data are

represented by RxNorm codes and the ADE data arerepresented by SNOMED CT and MedDRA codes.


10/232013 MFMER | slide-10

Electronic Medical Records (EMRs) DataSources

We used the UMLS CUI co-occurrence datathat were extracted from a 51-million-documentcorpus of clinical text covering Mayo Clinicclinical notes between 1/1/2001-12/31/2010 in a

previous study.

The clinical notes were originally retrieved fromthe Mayos Enterprise Data Trust (EDT), which

is a comprehensive snapshot of Mayo Clinicsservice areas and includes structured data,unstructured text, and Clinical Notes Indexing(CNI)-produced annotations.


11/232013 MFMER | slide-11

Methods

Profiled the Drug-Disorder Pairs andRelationships in the UMLS

Chemicals & Drugs Disorders

Enriched the UMLS ADE data with the EMR co-occurrence statistics

Evaluated the Utility of the UMLS ADE Data


12/232013 MFMER | slide-12

Profiling Results of ADE Knowledge in theUMLS

We extracted 266,832 concept pairs betweenthe two concept sets, covering 14,256 (1.69%)concepts from the group Chemicals & Drugsand 19,006 (3.53%) concepts from the group

Disorders.

We profiled the contribution of sourcevocabularies to the concept pairs. In total, there

are 47 source vocabularies having thecontribution.


13/232013 MFMER | slide-13

Distribution of drug-disorder concept pairscontributed by source vocabularies (top 20)

0.3%

0.3%

0.4%

0.4%

0.5%

0.6%

0.6%

0.6%

0.6%

0.6%

0.6%

0.6%

0.7%

0.7%

0.7%

2.1%

3.1%

7.3%

7.3%

70.4%

0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0%

MSHSPA

CCPSS

MSH

MSHSCR

MEDCIN

MSHITA

MSHPOL

MSHRUS

MSHCZE

MSHGER

MSHFRE

MSHFIN

NCI

MSHSWE

MEDLINEPLUS

SNMI

MTH

SCTSPA

SNOMEDCT

NDFRT


14/232013 MFMER | slide-14

Profiling results of relationships betweendrugs and disorders


15/232013 MFMER | slide-15

Co-occurrence Enrichment Using theEMRs

Drug-disorder concept pairs identified fromUMLS

Out of 266,832 concept pairs, 71,626concept pairs co-occurred in the EMRs,accounting for 26.8%.

Drug-disorder concept pairs from the ADEpediaOf 299,476 drug-disorder concept pairs,

100,577 concept pairs co-occurred in theEMRs, accounting for 33.6%.


16/232013 MFMER | slide-16

Utility Evaluation and Case Study

We identified 2,164 CUIs overlapping across the twodata sets (ADEpedia vs. UMLS), accounting for 56.5%of RxNorm Drugs in the ADEpedia.

The 2,164 CUIs linked with 66,058 drug-disorder pairsfrom the UMLS.

The result indicated that the UMLS is a useful resourcefrom the perspective of its integration with the

ADEpedia.

As a case study, we selected a drug called Digoxin,which is indicated for the treatment of congestive heartfailure.


17/232013 MFMER | slide-17

ADE knowledge extracted from theUMLS for the drug Digoxin|C0012265


18/232013 MFMER | slide-18

Discussion

The profiling results of the UMLS clearlydemonstrated that the drug-disorder associationsin the UMLS are largely underspecified.

Only 1.69% (14,256) concepts from the groupChemicals & Drugs and 3.53% (19,006)concepts from the group Disorders in theUMLS had drug-disorder associationsasserted.


19/232013 MFMER | slide-19

Discussion

From the perspective of ADE detectionapplication, we classified the assertedrelationships between the drug-disorder pairsinto 4 categories.

This kind of categorization would provideaggregation capability for the knowledge sourceand improve its utility for the purpose of drug

surveillance.


20/232013 MFMER | slide-20

Discussion

We enriched the knowledge base with thestatistical co-occurrence information extractedfrom the EMRs.

The co-occurrence information extracted from a51-million-document EMR system will be usefulfor validating ADE detection algorithm acrossclinical institutions and across text corpora.

In future, we will explore the characteristics ofthe UMLS term occurrences, focusing on theADE detection use case.


21/232013 MFMER | slide-21

Summary

The UMLS is a useful source for providing standardizedADE knowledge relevant to indications, contraindicationsand adverse effects, and complementary to the ADE datafrom drug product labels.

The statistics enrichment from EMRs would potentiallyenable the meaningful use of ADE data for drug safetysurveillance.

The EMR-enriched ADE dataset is available for downloadat the ADEpedia website (http://adepedia.org).


22/232013 MFMER | slide-22

ADEKnowledge

From EMRs

Standardized ADE Knowledge Base

(RDF Store)

ADE

Data

From

SPLs

ADEKnowledge

From

UMLS

Semantic Web RESTful Services(JSON, XML, RDF, TTL, etc.)

ADE Detection / Drug SurveillanceApplications

Data Layer

KnowledgeBase Layer

ServiceLayer

ApplicationLayer

Other ADEResources

(AERS)


23/23

Questions & Discussion

adepedia 2.0-integration of normalized adverse drug events (ades)

Documents