Transcript No Slide Title

VARIABLE SELECTION FOR
DECISION MAKING IN MENTAL HEALTH
1,2
Gunter ,
Lacey
STATISTICS
Departments of
and Institute for Social
Introduction
A=1
A=0
0.8
0.0
A=0
0.0
0.4
0.4
A=0
R
0.4
A=1
R
0.4
A=1
0.0
0.0
X4
0.4
0.8
0.0
0.4
X5
big interaction
big proportion
We have 3 components:
1. observations X = (X1, X2,…, Xp),
2. treatment action, A,
3. response, R
0.8
0.8
0.8
We propose ranking the variables in X based on potential for a qualitative
interaction with the treatment. We give a score for ranking the variables
based on 2 factors for evaluating qualitative interactions
1. The magnitude of the interaction between the variable and treatment
2. The proportion of patients whose optimal treatment changes given
knowledge of the variable
These 2 factors are illustrated in the plots below.
0.0
This poster discusses variable selection for medical
decision making; in particular, decisions regarding
when to provide treatment and which treatment to
provide patients with mental health disorders. Variable
selection is often needed in this setting to reduce costs
incurred by collecting unnecessary information and to
inform clinicians which variables are important for
individualizing treatment. We present a new technique
designed to find variables that aid in decision making.
We demonstrate the utility of this technique on data
from a randomized controlled trial which compared
three alternate treatments for chronic depression.
0.8
2
Research ,
University of Michigan, Ann Arbor
We demonstrate this method on data from a depression study to
determine which variables might help decipher the optimal depression
treatment for each patient.
Aim of the Nefazodone CBASP trial(1) – to compare efficacy of three
alternate treatments for chronic major depressive disorder (MDD):
1. Nefazodone,
2. Cognitive behavioral-analysis system of psychotherapy (CBASP)
3. Nefazodone + CBASP
For our analysis we used data from 440 patient with:
X
64 baseline variables listed in the table to the right
A
Nefazodone vs. Nefazodone + CBASP
R
Last observed Hamilton’s Rating Scale for Depression
score, post treatment
X6
small interaction
big proportion
big interaction
small proportion
We estimate the interaction factor by:
Dj = change in the effect of the optimal treatment over range of variable Xj
See plot below for illustration; a* is the overall optimal treatment, the blue
and red lines represent the fitted model, green ticks represent observations
Policy: guidelines for choosing treatment, A, given
observations, X
and Susan
1,2
Murphy
Depression Study Results
New Methods
R
Abstract
1
Statistics
Ji
1
Zhu ,
We used bootstrap sampling to minimize the variability of the results. On
each of 100 bootstrap samples, we performed the following analysis:
1. run new method U and the standard method
2. record the interaction variables selected
The plots below give the percentage of time each interaction was
selected for each method.
R
patient's condition and symptoms post treatment
R
A=0
0.0
0.4
X2
0.8
40
60
variable number
0.10
0.4
0.2
0.8
0.0
0.8
A=1
A=0
0.0
0.4
Algorithm
0.8
A=1
0.0
X1
0.8
0.0
2
Pj 
7
20
0
20
40
60
variable number
The green threshold lines in above plots were determined as follows:
1. Remove interaction effects from the data
2. Run methods on new data
3. Threshold : largest percentage of time a variable was selected
This helped assess the maximum selection percentage we expect to see
when no interaction effects exist. Only interaction variables with
selection percentages above these thresholds should be selected.
Results: The standard method selected 30 interaction variables. The
new method selected only 4 interaction variables: 2 indicators dealing
with alcohol, a somatic anxiety score and an indicator dealing with
specific phobia. For more details see (2).
1 Gender
2 Racial category
3-4 Marital status
5 Body mass index
6 Age in years at screening
7 Treated current depression
8 Medication current depression
9 Psychotherapy current depression
10 Treated past depression
11 Medication past depression
12 Psychotherapy past depression
13 Age of MDD onset
14-16 Number of depressive episodes
17 Length current episode
18-19 MDD type of current episode
20-21 MDD current severity
22-23 MDD chronic status
24 MDD threshold frequency
25 Dysthymic disorder current
26 Dysthymia initial onset
27 Length current dysthymia episode
28-29 Alcohol
30 Drug
31-32 Social phobia
33-34 Specific phobia
35 Obsessive compulsive
36-37 Post traumatic stress
38-39 Generalized anxiety
40 Anxiety disorder NOS
41-42 Panic disorder
43 Body dysmorphic current
44 Anorexia or Bulimia nervosa
45 Global assessment of function
46-47 Main study diagnosis
48 Severity of illness
49 Chronic or double depression
50 Total HAMA score
51 HAMA Sleep disturbance factor
52 HAMA Psychic Anxiety Score
53 HAMA Somatic Anxiety Score
54 Total HAMD-24 score
55 Total HAMD-17 score
56 HAMD Cognitive Disturbance
57 HAMD Retardation Score
58 HAMD Anxiety/Somatic symptom
59 IDSSR Total Score
60 IDSSR Anxious depression type
61 IDSSR General/Mood Cognition
62 IDSSR Anxiety/Arousal Score
63-64 IDSSR Sleep scores
Qualitative
Interaction
0.4
0.8
R
0.4
A=0
We combine Dj and Pj to make a score, Uj for each variable. The scores, U,
can be used to rank the variables.
Non-qualitative
Interaction
0.4
0.8
R
0.4
0.0
0.0
2 out of 7 subjects would
change choice of optimal
treatment given Xj
Xj
What is a qualitative interaction?
A=0
A=1=a*
0.4
R
Predictive selection techniques have been proposed,
but are only part of the puzzle. We need variables that
help determine the optimal treatment for each patient,
variables that qualitatively interact with the treatment.
0
We estimate the proportion factor by:
Pj = percentage of patients in the sample whose optimal treatment
changes when variable Xj is added to the fitted model
For example, see plot below.
0.8
Reasons for variable selection in decision making:
● limited resources
● better interpretability
● improved performance
A=1
0.8
Xj
Goal: discover optimal treatment for any future patient
No
Interaction
0.4
Dj = max effect –
min effect
0.00
assigned treatment
0.0
maximum
effect of
treatment
a* on R
New Method U
% of time chosen
A
A=0





0.0
X
baseline variables such as patient’s background,
medical history, current symptoms, etc.
0.4
Example:
clinical trial to test two alternative drug treatments
A=1=a*
0.0
minimum
effect of 

treatment 
a* on R
R
Goal: find the policy which results in the highest
average response
% of time chosen
Standard Method
Variable
% Chosen
Standard Method
Method
U
21
0
11
1
2,13
0,1
2
1
9
0
2
0
23
0
16
0
16
1
10
0
38
0
5
0
16,22,14
0,0,0
14
2
19,18
1,1
9,3
0,1
15,20
0,0
8
0
4
0
23
0
1
0
28,46
12,17
1
0
11,28
2,3
3,32
0,6
51
0
9,2
0,0
28,15
0,0
26
0
26,27
0,0
8
0
22
3
5
0
5,8
0,0
1
0
6
0
3
2
3
0
3
1
34
14
4
1
4
0
10
0
2
0
1
0
2
0
5
0
5
4
0
3
10,3
0,0
0.0
0.4
0.8
X3
X qualitatively interacts with the treatment if at least two
subsets of X values result in different optimal treatments.
For a complete variable selection method using this new ranking procedure,
we suggest the following algorithm which we call New Method U:
1. Select important predictors of R in X using a predictive variable
selection method
2. Rank the variables in X using score U; select the top k in rank
3. Use a predictive variable selection method to select from important
predictors chosen in step 1, A, and k interactions chosen in step 2
We compare this method versus a standard method: a Lasso of the main
effects of X, A and the interactions between X and A
Conclusion
In this poster, we presented a new technique
explicitly designed to select variables for decision
making. We demonstrated this method on a
depression data set. We found new method U did a
better job eliminating interaction variables that are not
important for prescribing treatment, which allow
clinicians to focus on important variables that can
help make treatment more individualized.
Acknowledgements: We wish to thank Martin Keller and the investigators of [2] for
use of their data, and gratefully acknowledge Bristol-Myers Squibb for helping fund the
study. We also acknowledge financial support from NIH grants R21 DA019800, K02
DA15674, P50 DA10075, and NSF grant DMS 0505432 and technical support from A.
John Rush, MD, University of Texas Southwestern.
References:
(1) Keller, M.B., McCullough, J.P., Klein, D.N.et al.: A Comparison of Nefazodone, the
Cognitive Behavioral-analysis System of Psychotherapy, and Their Combination
for Treatment of Chronic Depression. N. Engl. J. Med. 342 (2000) 331-366
(2) Gunter, L., Murphy, S.A., Zhu, J.: Variable Selection for Optimal Decision Making.
Technical Report 463, University of Michigan Statistics Department