Healthcare Models and Domains overview

 

This page gives you an overview of every healthcare problem and domain that can be solved with NLU for healthcare models, together with concrete examples. See this notebook and the accompanying video below for an introduction to every healthcare domain.

Open In Colab

Medical Named Entity Recognition (NER)

Named entities are sub-strings in a text that can be classified into catogires of a domain. For example, in the String
"Tesla is a great stock to invest in " , the sub-string "Tesla" is a named entity, it can be classified with the label company by an ML algorithm.
Named entities can easily be extracted by the various pre-trained Deep Learning based NER algorithms provided by NLU. NER models can be trained for many different domains and aquire expert domain knowledge in each of them. JSL provides a wide array of experts for various Medical, Helathcare and Clinical domains

This algorithm is provided by Spark NLP for Healthcare’s MedicalNerModel

Domain Description Sample NLU Spells Sample Entities Sample Predicted Labels Reference Links
ADE
(Adverse Drug Events)
Find adverse drug event (ADE)
related entities
med_ner.ade_biobert Aspirin , vomiting DRUG, ADE CADEC, Twimed
Anatomy Find body parts, anatomical sites a
nd reference
related entities
med_ner.anatomy tubules, nasopharyngeal aspirates,
embryoid bodies, NK cells,
Mitochondrial, tracheoesophageal fistulas,
heart, colon cancer, cervical,
central nervous system
Tissue_structure, Organism_substance,
Developing_anatomical_structure, Cell,
Cellular_component, Immaterial_anatomical_entity, organ,
Pathological_formation,
Organism_subdivision, Anatomical_system
AnEM
Cellular/
Molecular Biology
Find Genes, Molecules, Cell or
general Biology
related entities
med_ner.cellular.biobert human T-cell leukemia virus type 1 Tax-responsive ,
primary T lymphocytes,
E1A-immortalized,
Spi-B mRNA, zeta-globin
DNA, Cell_type,
Cell_line, RNA, Protein
JNLPBA
Chemical/Genes/
Proteins
Find Chemical, Gene and Protein
related entities
med_ner.chemprot.clinical nitrogen , β-amyloid , NF-kappaB CHEMICAL,
GENE-Y, GENE-N
ChemProt
Chemical Compounds Find general chemical compound
related entities
med_ner.chemicals resveratrol , β-polyphenol CHEM Dataset by John Snow Labs
Drug Chemicals Find chemical and drug
related entities
med_ner.drugs potassium , anthracyclines, taxanes DrugChem.
DrugChem.DrugChem
i2b2 + FDA
Posology/Drugs Find posology and drug
related entities
med_ner.posology.biobert 5000 units, Aspirin, 14 days, tablets,
daily, topically, 30 mg
DOSAGE, DRUG, DURATION, FORM,
FREQUENCY, ROUTE, STRENGTH.
i2b2 + FDA
Risk Factors Find risk factor of patient
related entities
med_ner.risk_factors.biobert coronary artery disease, hypertension,
Smokes 2 packs of cigarettes per day, morbid obesity,
Actos, Works in School, diabetic,
diabetic
CAD, HYPERTENSION,
SMOKER, OBESE,
FAMILY_HIST,
MEDICATION, PHI,
HYPERLIPIDEMIA, DIABETES
De-identification and Heart Disease Risk Factors Challenge datasets
cancer Genetics Find cancer and genetics
related entities
med_ner.cancer human, Kir 3.3, GIRK3, potassium,
GIRK, chromosome 1q21-23, pancreas, tissues,
fat andskeletal muscle, KCNJ9, Type II,
breast cancer,
patients, anthracyclines,
taxanes,
vinorelbine, patients,
breast,
vinorelbine inpatients, anthracyclines
Amino_acid, Anatomical_system,
cancer,
Cell, Cellular_component, Developing_anatomical_Structure
, Gene_or_gene_product, Immaterial_anatomical_entity,
Multi-tissue_structure, Organ, Organism ,
Organism_subdivision, Simple_chemical, Tissue
CG TASK of BioNLP 2013
Diseases Find disease related entities med_ner.diseases.biobert the cyst, a large Prolene suture,
a very small incisional hernia, the hernia cavity,
omentum, the hernia, the wound lesion,
The lesion,
the existing scar, the cyst,
the wound,
this cyst down to its base,
a small incisional hernia, The cyst
Disease CG TASK of BioNLP 2013
Bacterial Species Find bacterial species
related entities
med_ner.bacterial_species Neisseria wadsworthii,
N. bacilliformis,
Spirochaeta litoralis
SPECIES Dataset by John Snow Labs
Medical
Problem/Test/Treatment
Find medical problem,test
and treatment
related entities
med_ner.healthcare respiratory tract infection ,
Ourexpression studies,
atorvastatin
PROBLEM, TEST, TREATMENT i2b2
Clinical
Admission Events
Find clinical admission event
related entities
med_ner.admission_events 2007, 12 AM,
Headache,
blood sample,
presented,
emergency room,
daily
DATE, TIME, PROBLEM,
TEST, TREATMENT, OCCURENCE,
CLINICAL_DEPT, EVIDENTIAL, DURATION,
FREQUENCY, ADMISSION, DISCHARGE
Custom i2b2, enriched with Events
Genetic Variants Find genetic variant
related entities
en.med_ner.genetic_variants rs1061170, p.S45P, T13046C DNAMutation, ProteinMutation, SNP TMVAR
PHI (Protected Healthcare
Information)
Find PHI(Protected Healthcare)
related entities
en.med_ner.deid 2093-01-13, David Hale, Hendrickson,<br> Ora, 7194334,
01/13/93, Oliveira,
25-year-old, 1-11-2000, Cocke County Baptist Hospital,
0295 Keats Street., (302) 786-5227, Brothers Coal-Mine
MEDICALRECORD,
ORGANIZATION, DOCTOR,
USERNAME, PROFESSION,
HEALTHPLAN, URL, CITY,
DATE, LOCATION-OTHER, STATE,
PATIENT,
DEVICE, COUNTRY,
ZIP, PHONE,
HOSPITAL, EMAIL, IDNUM,
SREET, BIOID, FAX, AGE
n2c2 i2b2-PHI
Social Determinants /
Demographic Data
Find Social Determinants and
Demographic Data Related Entities
med_ner.jsl.enriched 21-day-old, male, congestion,
mom, suctioning yellow discharge,
she, problems with his breathing,
perioral cyanosis, retractions, mom,
Tylenol, His, his, respiratory congestion,
He, tired, fussy, albuterol
Age, Diagnosis, Dosage,
Drug_Name, Frequency, Gender,
Lab_Name, Lab_Result, Symptom_Name
Dataset by John Snow Labs
General Clinical Find General Clinical Entities med_ner.jsl.wip.clinical.modifier 28-year-old, female,
gestational, diabetes,
mellitus, eight, years,
prior, type,
two, diabetes, mellitus, T2DM,
HTG-induced, pancreatitis,
three, years,
prior, acute,
hepatitis, obesity,
body, mass, index,
BMI,
kg/m2, polyuria,
polydipsia,
poor,
appetite, vomiting,
Two,
weeks, prior,
she, five-day, course
Injury_or_Poisoning, Direction,
Test, Admission_Discharge, Death_Entity,
Relationship_Status, Duration, Respiration,
Hyperlipidemia, Birth_Entity, Age, Labour_Delivery,
Family_History_Header, BMI, Temperature,
Alcohol,
Kidney_Disease, Oncological,
Medical_History_Header, Cerebrovascular_Disease, Oxygen_Therapy,
O2_Saturation, Psychological_Condition,
Heart_Disease, Employment, Obesity,
Disease_Syndrome_Disorder, Pregnancy,
ImagingFindings, Procedure,
Medical_Device, Race_Ethnicity,
Section_Header, Symptom,
Treatment, Substance,
Route, Drug_Ingredient,
Blood_Pressure, Diet,
External_body_part_or_region,
LDL, VS_Finding, Allergen,
EKG_Findings, Imaging_Technique, Triglycerides,
RelativeTime, Gender, Pulse,
Social_History_Header, Substance_Quantity,
Diabetes, Modifier,
Internal_organ_or_component,
Clinical_Dept, Form, Drug_BrandName,
Strength, Fetus_NewBorn,
RelativeDate, Height, Test_Result,
Sexually_Active_or_Sexual_Orientation, Frequency, Time,
Weight, Vaccine,
Vital_Signs_Header,
Communicable_Disease, Dosage,
Overweight, Hypertension,
HDL, Total_Cholesterol, Smoking, `
Dataset by John Snow Labs
Radiology Find Radiology
related entities
med_ner.radiology.wip_clinical Bilateral, breast, ultrasound,
ovoid mass,
0.5 x 0.5 x 0.4, cm, anteromedial aspect,
left, shoulder, mass,
isoechoic echotexture, muscle,
internal color flow,
benign fibrous tissue, lipoma
ImagingTest, Imaging_Technique, ImagingFindings, OtherFindings, BodyPart, Direction, Test,
Symptom, Disease_Syndrome_Disorder,
Medical_Device, Procedure, Measurements, Units
Dataset by John Snow Labs, MIMIC-CXR and MT Radiology texts
Radiology Clinical
JSL-V1
Find radiology
related entities in clinical setting
med_ner.radiology.wip_greedy_biobert Bilateral, breast, ultrasound,
ovoid mass,
0.5 x 0.5 x 0.4, cm, anteromedial aspect,
left, shoulder, mass,
isoechoic echotexture, muscle,
internal color flow,
benign fibrous tissue, lipoma
Test_Result, OtherFindings, BodyPart, ImagingFindings,
Disease_Syndrome_Disorder, ImagingTest,
Measurements, Procedure,
Score, Test, Medical_Device, Direction,
Symptom, Imaging_Technique, ManualFix, Units
Dataset by John Snow Labs,
Genes
and Phenotypes
Find Genes and Phenotypes
(the observable physical
properties of an organism) related entities
med_ner.human_phenotype.gene_biobert APOC4 , polyhydramnios GENE, PHENOTYPE PGR_1, PGR_2
Normalized Genes
and Phenotypes
Find Normalized Genes and Phenotypes
(the observable physical
properties of an organism)
related entities
med_ner.human_phenotype.go_biobert protein complex oligomerization , defective platelet aggregation GO, HP PGR_1, PGR_2
Radiology Clinical
JSL-V2
Find radiology
related entities in clinical setting
med_ner.jsl.wip.clinical.rd   Kidney_Disease, HDL, Diet, Test, Imaging_Technique,
Triglycerides, Obesity, Duration, Weight,
Social_History_Header, ImagingTest, Labour_Delivery,
Disease_Syndrome_Disorder,
Communicable_Disease, Overweight,
Units, Smoking,
Score, Substance_Quantity,
Form, Race_Ethnicity,
Modifier, Hyperlipidemia, ImagingFindings,
Psychological_Condition, OtherFindings,
Cerebrovascular_Disease, Date, Test_Result,
VS_Finding, Employment,
Death_Entity, Gender, Oncological,
Heart_Disease, Medical_Device,
Total_Cholesterol, ManualFix,
Time, Route, Pulse,
Admission_Discharge, RelativeDate
, O2_Saturation, Frequency,
RelativeTime, Hypertension, Alcohol,
Allergen, Fetus_NewBorn,
Birth_Entity, Age,
Respiration, Medical_History_Header,
Oxygen_Therapy, Section_Header, LDL,
Treatment, Vital_Signs_Header, Direction,
BMI, Pregnancy,
Sexually_Active_or_Sexual_Orientation, Symptom,
Clinical_Dept, Measurements,
Height, Family_History_Header,
Substance,
Strength,
Injury_or_Poisoning,
Relationship_Status,
Blood_Pressure, Drug, Temperature, ,
EKG_Findings, Diabetes, BodyPart,
Vaccine, Procedure, Dosage
Dataset by John Snow Labs,
General
Medical Terms
Find general medical terms
and medical entities.
med_ner.medmentions   Qualitative_Concept, Organization, Manufactured_Object,
Amino_Acid, Peptide_or_Protein,
Pharmacologic_Substance, Professional_or_Occupational_Group,
Cell_Component, Neoplastic_Process, Substance, Laboratory_Procedure,
Nucleic_Acid_Nucleoside_or_Nucleotide,
Research_Activity, Gene_or_Genome, Indicator_Reagent_or_Diagnostic_Aid,
Biologic_Function, Chemical, Mammal,
Molecular_Function, Quantitative_Concept,
Prokaryote, Mental_or_Behavioral_Dysfunction,
Injury_or_Poisoning, Body_Location_or_Region,
Spatial_Concept, Nucleotide_Sequence,
Tissue, Pathologic_Function,
Body_Substance, Fungus, Mental_Process,
Medical_Device, Plant, Health_Care_Activity,
Clinical_Attribute, Genetic_Function,
Food, Therapeutic_or_Preventive_Procedure,
Body_Part_Organ,
Organ_Component, Geographic_Area, Virus,
Biomedical_or_Dental_Material, Diagnostic_Procedure, Eukaryote,
Anatomical_Structure, Organism_Attribute,
Molecular_Biology_Research_Technique, Organic_Chemical, Cell,
Daily_or_Recreational_Activity,
Population_Group, Disease_or_Syndrome,
Group, Sign_or_Symptom, Body_System
MedMentions

Entity Status Assertion

Named Entities extracted by an NER model can be further classified into sub-classes or statuses, depending on the context of the sentence. See the following two examples :

  1. Billy hates having a headache
  2. Billy has a headache
  3. Billy said his father has regular headaches

All sentences have the entity headache which is of class disease. But there is a semantic difference on what the actual status of the disease mentioned in text is. In the first and third sentence, Billy has no headache, but in the second sentence Billy actually has a sentence.
The Entity Assertion Algorithms provided by JSL solve this problem. The disease entity can be classified into ABSENT for the first case and into PRESENT for the second case. The third case can be classified into PRESENT IN FAMILY.
This has immense implications for various data analytical approaches in the helathcare domain.

I.e. imagine you want you want to make a study about hearth attacks and survival rate of potential procedures. You can process all your digital patient notes with an Medical NER model and filter for documents that have the Hearth Attack entity.
But your collected data will have wrong data entries because of the above mentioned Entity status problem. You cannot deduct that a document is talking about a patient having a hearth attack, unless you assert that the problem is actually there which is what the Resolutions algorithms do for you.

Keep in mind: This is a simplified example, entities should actually be mapped to their according Terminology (ICD-10-CM/ICD-10-PCS, etc..) to solve disambiguity problems and based on their codes all analysis should be performed

This algorithm is provided by Spark NLP for Healthcare’s AssertionDLModel

Domain Description Spell Predicted Entities Examples Reference Dataset
Radiology Predict status of Radiology related entities assert.radiology Confirmed, Negative, Suspected - Confirmed: X-Ray scan shows cancer in lung.
- Negative : X-Ray scan shows no sign of cancer in lung.
- Suspected :X-Ray raises suspicion of cancer in lung but does not confirm it.
Internal Dataset by Annotated by John Snow Labs
Healthcare/Clinical extended and Family JSL powerd Predict status of, Healthcare/Clinical/Family related entities.
Additional training with JSL Dataset
assert.jsl Present, Absent, Possible, Planned, Someoneelse, Past, Family, Hypotetical - Present: Patient diagnosed with cancer in 1999
- Absent: No sign of cancer was shown by the scans
- Possible: Tests indicate patient might have cancer
- Planned: CT-Scan is scheduled for 23.03.1999
- Someoneelse: The patient gave Aspirin to daugther.
- Past: The patient has no more headaches since the operation
- Family: The patients father has cancer.
- Hypotetical:Death could be possible.
2010 i2b2 + Data provided by JSL
Healthcare/Clinical JSL powerd Predict status of Healthcare/Clinical related entities.
Additional training with JSL Dataset
assert.jsl_large present, absent, possible, planned, someoneelse, past - present: Patient diagnosed with cancer in 1999
- absent: No sign of cancer was shown by the scans
- possible: Tests indicate patient might have cancer
- planned: CT-Scan is scheduled for 23.03.1999
- someoneelse: The patient gave Aspirin to daugther
- past: The patient has no more headaches since the operation
2010 i2b2 + Data provided by JSL
Healthcare/Clinical classic Predict status of Healthcare/Clinical related entities assert.biobert present , absent, possible, conditional, associated_with_someone_else ,hypothetical - present: Patient diagnosed with cancer in 1999
- absent: No sign of cancer was shown by the scans
- possible: Tests indicate patient might have cancer
- conditional If the test is positive, patient has AIDS
- associated_with_someone_else: The patients father has cancer.
-hypothetical :Death could be possible.
2010 i2b2

Entity Resolution

Named entities are sub-strings in a text that can be classified into catogires of a domain. For example, in the String
"Tesla is a great stock to invest in " , the sub-string "Tesla" is a named entity, it can be classified with the label company by an ML algorithm.
Named entities can easily be extracted by the various pre-trained Deep Learning based NER algorithms provided by NLU.

After extracting named entities an entity resolution algorithm can be applied to the extracted named entities. The resolution algorithm classifies each extracted entitiy into a class, which reduces dimensionality of the data and has many useful applications. For example :

  • Tesla is a great stock to invest in “
  • TSLA is a great stock to invest in “
  • Tesla, Inc is a great company to invest in” The sub-strings Tesla , TSLA and Tesla, Inc are all named entities, that are classified with the labeld company by the NER algorithm. It tells us, all these 3 sub-strings are of type company, but we cannot yet infer that these 3 strings are actually referring to literally the same company.

This exact problem is solved by the resolver algorithms, it would resolve all these 3 entities to a common name, like a company ID. This maps every reference of Tesla, regardless of how the string is represented, to the same ID.

This example can analogusly be expanded to healthcare any any other text problems. In medical documents, the same disease can be referenced in many different ways.

With NLU Healthcare you can leverage state of the art pre-trained NER models to extract Medical Named Entities (Diseases, Treatments, Posology, etc..) and resolve these to common healthcare disease codes.

This algorithm is provided by Spark NLP for Healthcare’s SentenceEntitiyResolver

Domain/Terminology Description Sample NLU Spells Sample Entities Sample Predicted Codes Reference Links
ICD-10 / ICD-10-CM
(International Classification of Diseases -
Clinical Modification)
Get ICD-10-CM codes of Medical and Clinical Entities.
The ICD-10 Clinical Modification (ICD-10-CM) is a modification of the ICD-10,
authorized by the World Health Organization,
used as a source for diagnosis codes in the U.S.
Be aware, ICD10-CM is often referred to as ICD10
resolve.icd10cm.augmented hypertension , gastritis I10, K2970 ICD-10-CM , WHO ICD-10-CM
ICD-10-PCS
(International Classification of Diseases -
Procedure Coding System)
Get ICD-10-PCS codes of Medical and Clinical Entities.
The International Classification of Diseases, Procedure Coding System (ICD-10-PCS),
is a U.S. cataloging system for procedural code
It is maintaining by Centers for Medicare & Medicaid Services
resolve.icd10pcs hypertension , gastritis DWY18ZZ, 04723Z6 ICD10-PCS, CMS ICD-10-PCS
ICD-O
(International Classification of Diseases, Oncollogy)
Topography & Morphology codes
Get ICD-0 codes of Medical and Clinical Entities.
The International Classification of Diseases for Oncology (ICD-O),
is a domain-specific extension of
the International Statistical Classification of Diseases
and Related Health Problems for tumor diseases.
resolve.icdo.base metastatic lung cancer 9050/3+C38.3, 8001/3+C39.8 ICD-O Histology Behaviour dataset
HCC
(Hierachical Conditional Categories)
Get HCC codes of Medical and Clinical Entities.
Hierarchical condition category (HCC) relies on ICD-10 coding to assign risk scores to patients.
Along with demographic factors (such as age and gender),
insurance companies use HCC coding to assign patients a risk adjustment factor (RAF) score.
resolve.hcc hypertension , gastritis 139, 188 HCC
ICD-10-CM + HCC Billable Get ICD-10-CM and HCC codes of Medical and Clinical Entities. resolve.icd10cm.augmented_billable metastatic lung cancer C7800 + ['1', '1', '8'] ICD10-CM HCC
CPT
(Current Procedural Terminology)
Get CPT codes of Medical and Clinical Entities.
The Current Procedural Terminology(CPT) is developed by the American Medical Association (AMA)
and used to assign codes to medical procedures/services/diagonstics.
The codes are used to derive the amount of payment a healthcare provider
may receives from insurance companies for the provided service.receives
resolve.cpt.procedures_measurements calcium score, heart surgery 82310, 33257 CPT
LOINC
(Logical Observation Identifiers Names and Codes)
Get LOINC codes of Medical and Clinical Entities.
Logical Observation Identifiers Names and Codes (LOINC)
developed by theU.S. organization Regenstrief Institute
LOINC is a code system for identifying test observations.
resolve.loinc acute hepatitis ,obesity 28083-4,50227-8 LOINC
HPO
(Human Phenotype Ontology)
Get HPO codes of Medical and Clinical Entities.
resolve.HPO cancer, bipolar disorder 0002664, 0007302, 0100753 HPO
UMLS
(Unified Medical Language System) CUI
Get UMLS codes of Medical and Clinical Entities.
resolve.umls.findings vomiting, polydipsia, hepatitis C1963281, C3278316, C1963279 UMLS
SNOMED International
(Systematized Nomenclature of Medicine)
Get SNOMED (INT) codes of Medical and Clinical Entities.
Defines sets of codes for entities in medical reports.
resolve.snomed.findings_int hypertension 148439002 SNOMED
SNOMED CT
(Clinical Terms)
Get SNOMED (CT) codes of Medical and Clinical Entities.
resolve.snomed.findings hypertension 73578008 SNOMED
SNOMED Conditions Get SNOMED Conditions codes of Medical and Clinical Entities.
resolve.snomed_conditions schizophrenia 58214004 SNOMED
RxNorm
and RxCUI
(Concept Uinque Indentifier)
Get Normalized RxNorm and RxCUI codes of Medical, Clinical and Drug Entities.
resolve.rxnorm 50 mg of eltrombopag oral 825427 [RxNorm Overview] [November 2020 RxNorm Clinical Drugs ontology graph]

Entity Relationship Extraction

Most sentences and documents have a lof of entities which can be extracted with NER. These entities alone already provide a lot of insight and information about your data, but there is even more information extractable…

Each entity in a sentence always has some kind of relationship to every other entity in the sentence. In other words, each entity pair has a relationship ! If a sentence has N entities, there are NxN potential binary relationships and NxNxK for k-ary relationships.
The RelationExtraction algortihms provided by JSL classify for each pair of entities what the type of relationship between is, based on some domain.

A concrete use-case example:
Lets say you want to analyze the survival rate of amputation procedures performed on the left hand.
Using just NER, we could find all documents that mention the entity amputation , left and hand.
The collected data will have wrong entries, imagine the following clinical note :

  • The patients left foot and his right hand were amputated

This record would be part of our analysis, if we just use NER with the above mentioned filtering.
The RelationExtraction Algorithms provided by JSL solves this problem. The relation.bodypart.directions model can classify for each entity pair, wether they are related or not.
In our example, it can classify that left and foot are related and that right and hand are related. Based on these classified relationships, we can easily enhance our filters and make sure no wrong records are used for our surival rate analysis.

But what about the following sentence?

  • The patients left hand was saved but his foot was amputated

This would pass all the NER and Relationship filters defined sofar. But we can easily cover this case by using the relation.bodypart.procedures model, which can predict wether a procedure entity was peformed on some bodypart or not. In the last example, it can predict foot and amputated are related, buthand and amputated are not in relationship, aswell as left and amputated (since every entity pair gets a prediction).

In conclusion, we can adjust our filters to additionaly verify that the amputation procedure is peformed on a hand and that this hand is in relationship with a direction entity with the value left.

Keep in mind: This is a simplified example, entities should actually be mapped to their according Terminology (ICD-10-CM/ICD-10-PCS, etc..) to solve disambiguity problems and based on their codes all analysis should be performed

These algorithms are provided by Spark NLP for Healthcare’s RelationExtraction and RelationExtractionDL

Entity Relationship Extraction - Overview

Domain Description Sample NLU Spells Predictable Relationships and Explanation
Dates and Clinical Entities Predict binary temporal relationship between
Date Entities and Clinical Entities
relation.date - 1 for Date Entity and Clinical Entity are related.
- 0 for Date Entity and Clinical Entity are not related
Body Parts and Directions Predict binary direction relationship between
Bodypart Entities and Direction Entities
relation.bodypart.direction - 1 for Body Part and Direction are related
- 0 for Body Part and Direction are not related
Body Parts and Problems Predict binary location relationship between
Bodypart Entities and Problem Entities
relation.bodypart.problem - 1 for Body Part and Problem are related
- 0 for Body Part and Problem are not related
Body Parts and Procedures Predict binary application relationship between
Bodypart Entities and Procedure Entities
relation.bodypart.procedure - 1 for Body Part and Test/Procedure are related
- 0 for Body Part and Test/Procedure are not related
Adverse Effects between drugs (ADE) Predict binary effect relationship between
Drugs Entities and Adverse Effects/Problem Entities
relation.ade - 1 for Adverse Event Entity and Drug are related
- 0 for Adverse Event Entity and Drug are not related
Phenotype abnormalities,Genes and Diseases Predict binary caused by relationship between
Phenotype Abnormality Entities,
Gene Entities and Disease Entities
relation.humen_phenotype_gene - 1 for Gene Entity and Phenotype Entity are related
- 0 for Gene Entity and Phenotype Entity are not related
Temporal events Predict multi-class temporal relationship between
Time Entities and Event Entities
relation.temporal_events - AFTER if Any Entity occured after Another Entity
- BEFORE if Any Entity occured before Another Entity
- OVERLAP if Any Entity during Another Entity
Dates and Tests/Results Predict multi-class temporal cause,reasoning
and conclusion relationship between
Date Entities, Test Entities and Result Entities
relation.test_result_date - relation.test_result_date
- is_finding_of for Medical Entity is found because of Test Entity
- is_result_of for Medical Entity reason for doing Test Entity
- is_date_of for Date Entity relates to time of Test/Result
- 0 : No relationship
Clinical Problem, Treatment and Tests Predict multi-class cause,reasoning
and effect relationship between
Treatment Entities, Problem Entities and Test Entities
relation.clinical - TrIP: A certain treatment has improved/cured a medical problem
- TrWP: A patient’s medical problem has deteriorated or worsened because of treatment
- TrCP: A treatment caused a medical problem
- TrAP: A treatment administered for a medical problem
- TrNAP: The administration of a treatment was avoided because of a medical problem
- TeRP: A test has revealed some medical problem
- TeCP: A test was performed to investigate a medical problem
- PIP: Two problems are related to each other
DDI Effects of using Multiple Drugs
(Drug Drug Interaction)
Predict multi-class effects, mechanisms and reasoning
for DDI effects(Drug Drug Interaction) relationships
between Drug Entities
relation.drug_drug_interaction - DDI-advise when an advice/recommendation regarding aDrug Entity and Drug Entity is given
- DDI-effect when Drug Entity and Drug Entity have an effect on the human body (pharmacodynamic mechanism). Including a clinical finding, signs or symptoms, an increased toxicity or therapeutic failure.
- DDI-int when effect between Drug Entity and Drug Entity is already known and thus provides no additional information.
- DDI-mechanism when ** Drug Entity** and Drug Entity are affected by an organism (pharmacokinetic). Such as the changes in levels or concentration in a drug. Used for DDIs that are described by their PK mechanism
- DDI-false when a Drug Entity and Drug Entity have no interaction mentioned in the text
Posology
(Drugs, Dosage,
Duration, Frequency,
Strength)
Predict multi-class posology relationships between
Drug Entities,Dosage Entities,
Strength Entities,Route Entities,
Form Entities, Duration Entities and Frequency Entities
relation.posology - DRUG-ADE if Problem Entity Adverse effect of Drug Entity
- DRUG-DOSAGE if Dosage Entity refers to a Drug Entity
- DRUG-DURATION if Duration Entity refers to a Drug Entity
- DRUG-FORM if Mode/Form Entity refers to intake form of Drug Entity
- DRUG-FREQUENCY if Frequency Entity refers to usage of Drug Entity
- DRUG-REASON if Problem Entity is reason for taking Drug Entity
- DRUG-ROUTE if Route Entity refer to administration method of Drug Entity
- DRUG-STRENGTH if Strength Entity refers to Drug Entity
Chemicals and Proteins Predict Regulator, Upregulator, Downregulator, Agonist,
Antagonist, Modulator, Cofactor,
Substrate relationships between
Chemical Entities and Protein Entities
relation.chemprot - CPR:1 if One ChemProt Entity is Part of of Another ChemProt Entity
- CPR:2 if One ChemProt Entity is Regulator (Direct or Indirect) of Another ChemProt Entity
- CPR:3 if One ChemProt Entity is Upregulator/Activator/Indirect Upregulator of Another ChemProt Entity
- CPR:4 if One ChemProt Entity is Downregulator/Inhibitor/Indirect Downregulator of Another ChemProt Entity
- CPR:5 if One ChemProt Entity is Agonist of Another ChemProt Entity
- CPR:6 if One ChemProt Entity is Antagonist of Another ChemProt Entity
- CPR:7 if One ChemProt Entity is Modulator (Activator/Inhibitor) of Another ChemProt Entity
- CPR:8 if One ChemProt Entity is Cofactor of Another ChemProt Entity
- CPR:9 if One ChemProt Entity is Substrate and product of of Another ChemProt Entity
- CPR:10 if One ChemProt Entity is Not Related to Another ChemProt Entity

Entity Relationship Extraction - Examples

Domain Sentence With Relationships Predicted Relationships for Sample Sentence Reference Links
Dates and Clinical Entities This 73 y/o patient had CT on 1/12/95, with cognitive decline since 8/11/94. - 1 for CT and1/12/95
- 0 for ** cognitive decline** and 1/12/95
- 1 for cognitive decline and 8/11/94
Internal Dataset by Annotated by John Snow Labs
Body Parts and Directions MRI demonstrated infarction in the upper - brain stem , left cerebellum and right basil ganglia - 1 for uppper and brain stem
- 0 for upper and cerebellum
- 1 for left and cerebellum
Internal Dataset by Annotated by John Snow Labs
Body Parts and Problems Patient reported numbness in his left hand and bleeding from ear. - 1 for numbness and hand
- 0 for numbness and ear
- 1 for bleeding and ear
Internal Dataset by Annotated by John Snow Labs
Body Parts and Procedures The chest was scanned with portable ultrasound and amputation was performed on foot - 1 for chest and portable ultrasound
- 0 for chest and amputation
- 1 for foot and amputation
Internal Dataset by Annotated by John Snow Labs
Adverse Effects between drugs (ADE) Taking Lipitor for 15 years, experienced much sever fatigue! Doctor moved me to voltaren 2 months ago , so far only experienced cramps - 1 for sever fatigue and Liptor
- 0 for sever fatigue and voltaren
- 0 for cramps and Liptor
- 1 for cramps and voltaren
Internal Dataset by Annotated by John Snow Labs
Phenotype abnormalities,Genes and Diseases She has a retinal degeneration, hearing loss and renal failure,
short stature, Mutations in the SH3PXD2B gene
coding for the Tks4 protein are responsible for the autosomal recessive.
- 1 for ** hearing loss** and SH3PXD2B
- 0 for retinal degeneration and hearing loss
- 1 for retinal degeneration and autosomal recessive
PGR aclAntology
Temporal events She is diagnosed with cancer in 1991.
Then she was admitted to Mayo Clinic in May 2000 and discharged in October 2001
- OVERLAP for cancer and 1991
- AFTER for additted and Mayo Clinic
- BEFORE for admitted and discharged
Temporal JSL Dataset and n2c2
Dates and Tests/Results On 23 March 1995 a X-Ray applied to patient because of headache, found tumor in brain - is_finding_of for tumor ** and **X-Ray
- is_result_of for headache ** and **X-Ray
- is_date_of for 23 March 1995 ** and **X-Ray
Internal Dataset by Annotated by John Snow Labs
Clinical Problem, Treatment and Tests - TrIP : infection resolved with antibiotic course
- TrWP : the tumor was growing despite the drain
- TrCP: penicillin causes a rash
- TrAP:Dexamphetamine for narcolepsy
- TrNAP: Ralafen was not given because of ulcers
- TeRP: an echocardiogram revealed a pericardial effusion
- TeCP: chest x-ray for pneumonia
- PIP: Azotemia presumed secondary to sepsis
- TrIP for infection and antibiotic course
- TrWP for tumor and drain
- TrCP for penicillin andrash
- TrAP for Dexamphetamine and narcolepsy
- TrNAP for Ralafen and ulcers
- TeRP for echocardiogram and pericardial effusion
- TeCP for chest x-ray and pneumonia
- PIP for Azotemia and sepsis
2010 i2b2 relation challenge
DDI Effects of using Multiple Drugs
(Drug Drug Interaction)
- DDI-advise: UROXATRALshould not be used in combination with other alpha-blockers
- DDI-effect: Chlorthalidone may potentiate the action of other antihypertensive drugs
- DDI-int : The interaction of omeprazole and ketoconazole has been established
- DDI-mechanism : Grepafloxacin may inhibit the metabolism of theobromine
- DDI-false : Aspirin does not interact with Chlorthalidone
- DDI-advise for UROXATRAL and alpha-blockers
- DDI-effect for Chlorthalidone and antihypertensive drugs
- DDI-int for omeprazole and ketoconazole
- DDI-mechanism for Grepafloxacin and theobromine
- DDI-false for Aspirin and Chlorthalidone
DDI Extraction corpus
Posology
(Drugs, Dosage,
Duration, Frequency,
Strength)
- DRUG-ADE: had a headache after taking Paracetamol
- DRUG-DOSAGE: took 0.5ML of** Celstone**
- DRUG-DURATION: took Aspirin daily for two weeks
- DRUG-FORM: took Aspirin as tablets
- DRUG-FREQUENCY : Aspirin usage is weekly
- DRUG-REASON : Took Aspirin because of headache
- DRUG-ROUTE: Aspirin taken orally
- DRUG-STRENGTH: 2mg of Aspirin
- DRUG-ADE for headache and Paracetamol
- DRUG-DOSAGE for 0.5ML and ** Celstone**
- DRUG-DURATION for Aspirin and for two weeks
- DRUG-FORM for Aspirin and tablets
- DRUG-FREQUENCY for Aspirin and weekly
- DRUG-REASON for Aspirin and headache
- DRUG-ROUTE for Aspirin and orally
- DRUG-STRENGTH for 2mg and Aspirin
Magge, Scotch, Gonzalez-Hernandez (2018)
Chemicals and Proteins - CPR:1 (Part of) : The amino acid sequence of the rabbit alpha(2A)-adrenoceptor has many interesting properties.
- CPR:2 (Regulator) : Triacsin inhibited ACS activity
- CPR:3 (Upregulator) : Ibandronate increases the expression of the FAS gene
- CPR:4 (Downregulator) : Vitamin C treatment resulted in reduced C-Rel nuclear translocation
- CPR:5 (Agonist) : Reports show tricyclic antidepressants act as agnonists at distinct opioid receptors
- CPR:6 (Antagonist) : GDC-0152 is a drug triggers tumor cell apoptosis by selectively antagonizing LAPs
- CPR:7 (Modulator) : Hydrogen sulfide is a allosteric modulator of ATP-sensitive potassium channels
- CPR:8 (Cofactor) : polyinosinic:polycytidylic acid ** and the **IFNα/β demonstrate capability of endogenous IFN.
- CPR:9 (Substrate) : ZIP9 plays an important role in the transport and toxicity of Cd(2+) cells
- CPR:10 (Not Related) **: Studies indicate that **GSK-3β inhibition by palinurin cannot be competed out by ATP
- CPR:1 (Part of) for amino acid and rabbit alpha(2A)-adrenoceptor
- CPR:2 (Regulator) for Triacsin and ACS
- CPR:3 (Upregulator) for Ibandronate and FAS gene
- CPR:4 (Downregulator) for Vitamin C and C-Rel
- CPR:5 (Agonist) for tricyclic antidepressants and opioid receptors
- CPR:6 (Antagonist) (Antagonist) for GDC-0152 and LAPs
- CPR:7 (Modulator) for Hydrogen sulfide and ATP-sensitive potassium channels
- CPR:8 (Cofactor) for polyinosinic:polycytidylic acid ** and **IFNα/β
- CPR:9 (Substrate) for ZIP9 and Cd(2+) cells
- CPR:10 (Not Related) ** for **GSK-3β and ATP
ChemProt Paper
Last updated