Generating Medical Logic Modules for Clinical Trial Eligibility

1
Generating Medical Logic Modules forClinical
Trial Eligibility

• Craig Parker
• Brigham Young University

2
Clinical Trials

• Important for modern medical research
• Efficacy of therapies
• Safety of therapies
• Sponsored by
• Government (www.clinicaltrials.gov)
• Academic institutions
• Industry (e.g. pharmaceutical companies)

3
Clinical Trial Enrollment

• Power in numbers
• Traditional enrollment methods have limitations

Methods Limitations
Regular office visits Limited pool of patients
Advertising Limited medical knowledge, Cost
Chart review Limited information, Requires medical training
4
What if . . .
we could automatically extract eligibility
criteria from electronic medical records?
5
Electronic Medical Records (EMRs)

• Good News
• Large amounts of data being collected
• Efforts to standardize representations are well
  supported
• Bad News
• Most EMRs are far from complete
• Not all representations are in a standard form
• Legacy data / Legacy data acquisition tools
• Unstandardized realms

6
Automating the Process

• Generate medical logic modules to represent
  eligibility criteria for clinical trials
• Start with criteria as first-order predicate
  logic
• Map concepts with medical vocabularies and
  ontologies
• Create medical logic modules
• Handle concepts that cant be mapped
• Measure results

7
Medical Logic Modules (MLMs)

• An abstract term describing the knowledge
  necessary for making a medical decision
• A specific type of medical program
• Represented using the Arden Syntax
• ANSI standard
• Broad vendor acceptance
• Compiles to an executable form

8
MLM in Arden Syntax
KNOWLEDGE TYPE . . . DATA
gestational_age read last SELECT
gest_age FROM Observation
EVOKE . . . LOGIC IF
gestational_age gt 73 days AND
gestational_age lt 97 days THEN
conclude true ENDIF
ACTION . . .
9
Beginning . . .
10
Beginning and End
KNOWLEDGE TYPE . . . DATA
gestational_age read last SELECT
gest_age FROM Observation
EVOKE . . . LOGIC IF
gestational_age gt 73 days AND
gestational_age lt 97 days THEN
conclude true ENDIF
ACTION . . .
11
Overview
Trial
12
Overview
Trial
13
Overview
Trial

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

14
Overview
Trial

• (P1) Gender Female
• (P2) Pregnant
• (P3) Gestational age gt 23.0 wks
• (P4) Gestational age lt 31.6 wks
• (P5) Chorioamnionitis
• (P6) Non-reassuring fetal testing

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

15
Overview
Trial

• (P1) Gender Female
• (P2) Pregnant
• (P3) Gestational age gt 23.0 wks
• (P4) Gestational age lt 31.6 wks
• (P5) Chorioamnionitis
• (P6) Non-reassuring fetal testing

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

P1 ? P2 ? (P3 ? P4) ? ?(P5 ? P6) Or in
conjunctive normal form P1 ? P2 ? P3 ? P4 ? ?P5
? ?P6
16
Overview
Trial

• (P1) Gender Female
• (P2) Pregnant
• (P3) Gestational age gt 23.0 wks
• (P4) Gestational age lt 31.6 wks
• (P5) Chorioamnionitis
• (P6) Non-reassuring fetal testing

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

P1 ? P2 ? (P3 ? P4) ? ?(P5 ? P6) Or in
conjunctive normal form P1 ? P2 ? P3 ? P4 ? ?P5
? ?P6
17
Overview
Trial
18
Overview
Trial
19
Overview
Trial
Additional Information
20
Steps to Create MLMs

• Classify predicates
• Map concepts from trial to database
• Translate logic
• Generate MLM

21
Classifying Predicates

• Numeric comparisons
• e.g. gestational age gt 23.0 wks
• time, lab values, physiologic measurements
• Single noun phrases
• e.g. pregnant
• diagnoses, observations
• Two noun phrases
• e.g. Gender Female
• name-value pairs

22
Mapping Example 1

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

23
Mapping Example 1

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

Single noun phrase likely to be a diagnosis or
observation
24
Mapping Example 1

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

From vocabulary of target database
pregnant pregnancy pregnancy appointment
type pregnant ambulatory status . . .
has-parent Diagnosis has-parent Appointment
type has-parent Ambulatory status
25
Mapping Example 1

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

pregnant pregnancy pregnancy appointment
type pregnant ambulatory status . . .
has-parent Diagnosis has-parent Appointment
type has-parent Ambulatory status
26
Mapping Example 1
pregnant pregnancy (83035)
has-parent diagnosis
Looking for pregnancy in schema of target
database.
27
Mapping Example 1
pregnant pregnancy (83035)
has-parent diagnosis
Looking for pregnancy in schema of target
database.
PregnancyObservation Is-Subtype-Of
DiagnosisAndFindingObservation value(codedTerm(
pregnancy, 83035)) negation(boolean) . . .
28
Mapping Example 1
pregnant pregnancy (83035)
has-parent diagnosis
PregnancyObservation Is-Subtype-Of
DiagnosisAndFindingObservation value(codedTerm(
pregnancy, 83035)) negation(boolean) . . .
Look for PregnancyObservation in vocabulary of
target database.
29
Mapping Example 1
pregnant pregnancy (83035)
has-parent diagnosis
PregnancyObservation Is-Subtype-Of
DiagnosisAndFindingObservation value(codedTerm(
pregnancy, 83035)) negation(boolean) . . .
Look for PregnancyObservation in vocabulary of
target database.
PregnancyObservation (59665)
has-parent observation
30
Mapping Example 1
pregnant pregnancy (83035)
has-parent diagnosis
PregnancyObservation Is-Subtype-Of
DiagnosisAndFindingObservation value(codedTerm(
pregnancy, 83035)) negation(boolean) . . .
PregnancyObservation (59665)
has-parent observation
SELECT FROM Observations WHERE ObsId 59665
AND value 83035
31
Mapping Example 2

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

32
Mapping Example 2

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

Two noun phrases likely to be a name-value pair
33
Mapping Example 2

• Inclusion
• Gender Female
• Pregnant
• Gestational age gt 23.0 wks and lt 31.6 wks
• Exclusion
• Chorioamnionitis
• Non-reassuring fetal testing

Demographics
Look in target database for a concept of Gender
with a value of Female.
Gender
Male
Female
34
Unmappable Concepts

• Concept or value not in target database
• Concept does not exist
• Delivery intended outside center
• Too Hard
• Inclusion Received full course of
  corticosteroids in the previous 7 days
• Exclusion Corticosteroid therapy, other than
  qualifying course

35
Unmappable Concepts

• Concept or value not in target database
• Concept does not exist
• Delivery intended outside center
• Too Hard
• Inclusion Received full course of
  corticosteroids in the previous 7 days
• Exclusion Corticosteroid therapy, other than
  qualifying course
• Solution
• Evaluate eligibility based on available data
• If eligibility is possible, present questionnaire
  to user for outstanding information needed

36
Evaluation

• Select 25 trials (200 predicates) from
  ClinicalTrials.gov
• Precision and recall of term mappings
• Precision and recall of predicate mappings
• Percentage of predicates that are mappable
• Correctness of logic in Arden Syntax modules