Can big data yield big
insights for depression?
www.gillanlab.com
AWARE Conference, Dublin 2019
Claire Gillan, PhD
Assistant Professor of Psychology and MQ Fellow
Trinity College Dublin
@clairegillanTCD
OCD
Bipolar
Schizophrenia
Depression
***
0
20
40
60
80
100
Resp
on
ses (
%)
Contr
ol
OC
D
Gillan et al., American Journal of Psychiatry, 2011
Habit
Same goes for schizophrenia!
And social anxiety…
And cocaine addiction…
…and binge-eating disorder, alcohol use disorder,
mixed results in anorexia…
OCD
OCD
Bipolar
Schizophrenia
Depression
OCD
BipolarDepression
Schizophrenia
*p<.05 ** p<.01 ***p<.001
Co
ntr
ol o
ver
hab
it
**** ***
p >.14
1,413 participants
(general population)
Eating D
isord
ers
Imp
uls
ivity
OC
D
Alc
ohol A
ddic
tion
Schiz
oty
py
Depre
ssio
n
Tra
it A
nxie
ty
Apath
y
Socia
l Anxie
ty
Gillan et al., eLife, 2016
Habits forming, trans-diagnostically
Anxious-Depression
Co
mp
uls
ivit
y
Social Withdrawal
0.5
0.0
0.5
0.0
-
0.5
0.0
0.5 -
0.5
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Inter-correlation of 209
individual self-report
questionnaire items
Gillan et al., eLife, 2016
Contr
ol over
habit
***
**
Dimension or Disorder?
Gillan et al., JAMA Psychiatry, 2019
Contr
ol over
habit
Contr
ol over
habit (
beta
)
N=285 diagnosed patients
GAD OCD GAD
+
OCD
This is interesting, but is this
valuable?
33% remission rate
47% response rateTrivedi et al., 2006
Antidepressants work!
They just don’t work for everyone.
Several candidate predictors of treatments response
But :
• Non-specific
• Weak, one-dimensional predictors
• Impractical (too expensive)
None are in clinical use
www.AntidepressantResearch.com
Can we use big data to make psychiatry’s first
objective tests?
Precision Treatment
Several candidate predictors of treatments response
But :
• Non-specific
• Weak, one-dimensional predictors
• Impractical (too expensive)
None are in clinical use
Our Objective:
• Specific – which treatment will work for you?
• Strong, multidimensional predictors
• Scalable, cheap and easy to access
Make this tool available
www.AntidepressantResearch.com
A novel method to achieve these goals
www.AntidepressantResearch.com
• Scalable: We want to recruit as we mean to implement
• Strong: We need a large sample (N=1000) to use machine
learning to capture multidimensional space
• Specific: We need to test comparative treatments
Early Identification
• Prevention is better than treatment.
• Can we detect an episode before it happens?
• Network theories suggest we might be able to pin it
down to a matter of weeks
anxietysleep motivation
self-esteem
moodappetite
Are tightly connected
networks are more
vulnerable?
Borsboom & Kramer, 2008
Weakly connected(‘resilient’)
Strongly connected(‘vulnerable’)
Baseline Network
Future diagnoses
Future treatments
freq
uenc
y
A BLow Disability OCD Patients (N=102)
High Disability OCD Patients (N=101)
Weakly connected(‘resilient’)
Strongly connected(‘vulnerable’)
Baseline Network
Future diagnoses
Future treatments
freq
uen
cy
A BLow Disability OCD Patients (N=102)
High Disability OCD Patients (N=101)
Kelley, …., & Gillan, in preparation
Early Identification
N=200 patients
Higher ImpairmentLower Impairment
episode
begins
weeks
co
nnectivity
@use_mybrain
Can we use signals like this to predict the future?
Want to help?
VISIT: www.AntidepressantResearch.com
Seeking:
• GPs
• Pharmacists
• Individuals +/- 2 days of starting antidepressants
@use_mybrain
Gillan LabTricia Seow
Andrew Pringle
Kevin Lynch
Eoghan Gallagher
Sean Kelley
Thank You
www.AntidepressantResearch.com
@use_mybrain