Endometriosis Phenotype Development, Validation, and ... · Endometriosis •Chronic disease in...
Transcript of Endometriosis Phenotype Development, Validation, and ... · Endometriosis •Chronic disease in...
Endometriosis Phenotype Development, Validation, and Characterization from
Observational Health Databases
Mollie McKillop1, Sharon Lipsky Gorman1, Shadi Safar Goli1, Chris D’ambrosia1, Christoper Knoll2,
Patrick Ryan2, Noémie Elhadad1
1Columbia University; 2Janssen Pharmaceuticals
Endometriosis Phenotype Development, Validation, and Characterization from
Observational Health Databases
Mollie McKillop1, Sharon Lipsky Gorman1, Shadi Safar Goli1, Chris D’ambrosia1, Christoper Knoll2,
Patrick Ryan2, Noémie Elhadad1
1Columbia University; 2Janssen Pharmaceuticals
Endometriosis
• Chronic disease in reproductive-age women
• Endometrial cells grow outside uterus
• Characterized through surgical findings
• Menstrual pain and infertility as the most common characterization
• Prevalent
• Estimated to ~10% of women in reproductive age
• Highly enigmatic
• No known biomarker, etiology, or treatment response
• Long lag to diagnosis (~10 years)
OHDSI Call – Oct 30, 2018
Research gaps
• Incomplete characterization • Systemic condition with symptoms beyond dysmenorrhea and infertility
• From onset of symptoms to after diagnosis
• Because misdiagnoses/under-diagnosis, need to identify the patterns of endo patients prior to diagnosis towards earlier/better detection
• Lack of validated phenotype• Epidemiological studies rely on single high-level ICD code
• Types of endometriosis have been proposed but focus on surgical findings and do not correlate with patient experience of disease
OHDSI Call – Oct 30, 2018
Research questions
1. What is an accurate phenotype for endometriosis to identify cohorts from EHR and claims databases?
2. What are the patterns of patients experiences before diagnosis? • Signs/symptoms, treatments, healthcare utilization patterns, etc.
• Additional desiderata: • Phenotype valid across claims and EHR databases to identify a wide range of
patients
• Open access to other OHDSI members
OHDSI Call – Oct 30, 2018
Methods
1. Define and validate an endometriosis phenotype for EHRs / claims
databases
2. Characterize cohorts pre-diagnosis across databases
OHDSI Call – Oct 30, 2018
Relevant concepts
• Endo diagnosis: endometriosis diagnosis codes (including adenomyosis)
• Endo-related procedures: guideline-based procedures for endometriosis
diagnosis and treatment (e.g., pelvic laparoscopy)
• Endo-prevalent procedures: procedures present >50% of patients with
≥1 endo diagnosis
• Endo-related imaging procedures: guidelines-based imaging procedures
(e.g. pelvic MRI)
OHDSI Call – Oct 30, 2018
Cohort definition experiments
OHDSI Call – Oct 30, 2018
Cohor
tInitial Event Inclusion Rules
A
endo-related procedures AND endo diagnosis
OR
endo-prevalent procedures AND endo diagnosis
Females ages 15-49
B
endo-related procedures AND endo diagnosis
OR
endo-prevalent procedures AND endo diagnosis
Females ages 15-49
AND
2 endo diagnosis
after index date
C
endo-related procedures AND endo diagnosis
AND
endo-related imaging procedures AND endo diagnosis before
index date
Females ages 15-49
D
endo-related procedures AND endo diagnosis
AND
endo-related imaging procedures AND endo diagnosis before
index date
Females ages 15-49
AND
2 endo diagnosis
after index date
Cohort definition validation
• Manual chart review from clinical experts on EHR database (Columbia)• Index data between 1/1/2016 to 6/1/2018
• Reviewed records of patients across all cohort definitions
• Confirmed endometriosis diagnosis through histological analysis post-laparoscopy
• Gold-standard annotations of 1,406 patient records• Two annotators
• Kappa on 38 records: .89*• Determined N=38 provides expected confidence limits between .6 and 1
OHDSI Call – Oct 30, 2018
* p-value <.05
Cohort definition validation
• Cohort B had highest precision/recall
• Negative predictive value 0.84
OHDSI Call – Oct 30, 2018
Cohort
name on
OHDSI.org
Precision Recall
Records in cohort
at Columbia EHR
2016-2018*
Total records in
cohort at Columbia
EHR 1999-2018
Cohort D 0.84 0.26 162 1,248
Cohort C 0.78 0.28 189 1,950
Cohort B 0.85 0.70 430 3,328
Cohort A 0.37 1** 1,406 5,666
* each of these cohorts was reviewed by clinical experts.
** all other cohorts are a subset of this cohort so recall=1.
Cohort definitions
•Available at:
• http://www.ohdsi.org/web/atlas/#/cohortdefinition/1769393
• http://www.ohdsi.org/web/atlas/#/cohortdefinition/1769395
• http://www.ohdsi.org/web/atlas/#/cohortdefinition/1769396
• http://www.ohdsi.org/web/atlas/#/cohortdefinition/1769397
OHDSI Call – Oct 30, 2018
Best performing cohort we
used to characterize
Methods
1. Define and validate an endometriosis phenotype for EHRs / claims
databases
2. Characterize cohorts pre-diagnosis across databases
OHDSI Call – Oct 30, 2018
Cohort characterization
• Select best-performing cohort, Atlas characterization
• Analysis carried out on four databases
• Columbia EHR (Columbia EHR)
• Optum® Clinformatics® Extended DataMart (Optum)
• IBM MarketScan® Commercial Database (MCD)
• IBM MarketScan® Multi-State Medicaid Database (MMMD)
• Report on prevalent (>10% of patients) conditions and drugs
• For comparison, prevalence in general cohort of women of reproductive
age
OHDSI Call – Oct 30, 2018
Cohort characterization
• Overall >480,000 patients in endometriosis cohort
OHDSI Call – Oct 30, 2018
Database Total records in cohort
Columbia EHR 3,328
Optum 24,725
MMMD 54,609
MCD 398,015
Conditions
OHDSI Call – Oct 30, 2018
Comparison (not control) Group Conditions
OHDSI Call – Oct 30, 2018
0
10
20
30
40
Ute
rine le
iom
yom
aP
ain
in p
elv
isC
yst of ova
ryA
bdom
inal p
ain
Fem
ale
genita
l org
an s
ympto
ms
Dys
menorr
hea
Urinary
tra
ct in
fect
ious
dis
ease
Exc
ess
ive a
nd fre
quent m
enst
ruatio
nH
eadach
eC
hest
pain
Nonin
flam
mato
ry d
isord
ers
of th
e o
vary
Ess
entia
l hyp
ertensi
on
Nonin
flam
mato
ry d
isord
er of th
e v
agin
aA
nxi
ety
dis
ord
er
Vagin
itis
and v
ulv
ova
gin
itis
Rig
ht lo
wer quadra
nt pain
Irre
gula
r periods
Cough
Dys
pnea
Low
back
pain
Const
ipatio
n
Conditions
Patie
nts
(%)
Columbia EHR
Endo Cohort
Comparison Group
Drugs
OHDSI Call – Oct 30, 2018
0
20
40
60
Anili
des
Natu
ral o
piu
m a
lkalo
ids
Pro
pio
nic
aci
d d
eriva
tives
Sero
tonin
(5H
T3) anta
gonis
tsS
ofteners
, em
olli
ents
Opiu
m a
lkalo
ids
and d
eriva
tives
Pro
ton p
um
p in
hib
itors
Benzo
dia
zepin
e r
ela
ted d
rugs
Benzo
dia
zepin
e d
eriva
tives
Pro
puls
ives
Pro
gest
ogens
and e
stro
gens
Penic
illin
s w
ith e
xtended s
pect
rum
Opio
id a
nest
hetic
s
Oth
er
dru
gs
for
gast
roin
test
inal d
isord
ers
Macr
olid
es
First
-genera
tion c
ephalo
sporins
Conta
ct la
xativ
es
Flu
oro
quin
olo
nes
Vita
min
D a
nd a
nalo
gues
H2-r
ece
pto
r anta
gonis
ts
ATC Class 5
Patien
ts(%
)
Comparison Group Drugs
OHDSI Call – Oct 30, 2018
0
20
40
60
Anili
des
Pro
pio
nic
aci
d d
eriva
tives
Natu
ral o
piu
m a
lkalo
ids
Softeners
, em
olli
ents
Opiu
m a
lkalo
ids
and d
eriva
tives
Sero
tonin
(5H
T3) anta
gonis
ts
Benzo
dia
zepin
e r
ela
ted d
rugs
Penic
illin
s w
ith e
xtended s
pect
rum
First
-genera
tion c
ephalo
sporins
Conta
ct la
xativ
es
Benzo
dia
zepin
e d
eriva
tives
Vita
min
D a
nd a
nalo
gues
Pro
ton p
um
p in
hib
itors
Pro
gest
ogens
Macr
olid
es
Pro
gest
ogens
and e
stro
gens
H2-r
ece
pto
r anta
gonis
tsP
ropuls
ives
Oth
er
pla
in v
itam
in p
repara
tions
Oth
er
dru
gs
for
gast
roin
test
inal d
isord
ers
Folic
aci
d a
nd d
eriva
tives
Solu
tions
for
pare
nte
ral n
utriti
on
Alu
min
ium
com
pounds
Asc
orb
ic a
cid (
vita
min
C), p
lain
Iron b
ivale
nt, o
ral p
repara
tions
Flu
oro
quin
olo
nes
Glu
coco
rtic
oid
s
ATC Class 5
Patien
ts(%
)
Discussion
• First study to develop well-validated endometriosis phenotype for
cohort selection
• Findings across databases consistent with ongoing endometriosis research
• Primary symptoms present prior to diagnosis related to pelvic pain and heavy pain
medication
• Results congruent with new knowledge about endometriosis
• Beyond dysmenorrhea and pelvic pain, systemic impact of disease (i.e. anxiety,
constipation)
• OHDSI essential for characterization
• Wouldn’t have started to get full picture w/o collaborators!
OHDSI Call – Oct 30, 2018
Next steps
• Characterize in larger, more diverse cohort
• Get in touch if you’d like to participate
• All definitions available on ohdsi.org and further queries available on demand
• Use phenotype definition for patient-level prediction
• In women in reproductive age presenting in ED with abdominal pain, who is likely
to be diagnosed with endometriosis 1year+ later?
• Using PLP modules
• Get in touch if you’d like to participate
OHDSI Call – Oct 30, 2018
Thanks, questions?
• Columbia OHDSI bootcamp participants
OHDSI Call – Oct 30, 2018
Comparison (not control) Group Conditions
0
20
40
60
Adult
health
exa
min
atio
nA
bdom
inal p
ain
Sin
gle
live
birth
Gyn
eco
logic
exa
min
atio
n
Patie
nt cu
rrently
pre
gnant
Neve
r sm
oke
d tobacc
oD
eliv
ery
norm
al
Headach
e
Urinary
tra
ct in
fect
ious
dis
ease
Fam
ily h
isto
ry o
f cl
inic
al f
indin
gC
hest
pain
Vagin
itis
and v
ulv
ova
gin
itis
His
tory
of cl
inic
al f
indin
g in
subje
ct
Fem
ale
genita
l org
an s
ympto
ms
Ess
entia
l hyp
ert
ensi
on
Scr
eenin
g for
malig
nant neopla
sm o
f bre
ast
Acu
te u
pper
resp
irato
ry in
fect
ion
Ante
nata
l ultr
aso
und s
can for poss
ible
abnorm
alit
yP
rim
igra
vida
Low
back
pain
Ante
nata
l scr
eenin
gP
ain
in li
mb
Acu
te p
hary
ngiti
sA
sthm
a
Com
plic
atio
n o
ccurr
ing d
uring p
regnancy
Cough
Subse
quent st
age o
f st
aged o
pera
tion
Pre
gnancy
test
negativ
eB
reast
lum
p
Conditions
Pa
tients
(%)
OHDSI Call – Oct 30, 2018