Decision Support Tool for Adaptive Interdisciplinary Chronic Pain Management

Aera K. LeBoulluec, Victoria C. P. Chen, Li Zeng , Aisha Gillan

Industrial & Manufacturing Systems Engineering, The University of Texas at Arlington, Arlington, Texas, 76019

Robert J. Gatchel
Psychology, The University of Texas at Arlington, Arlington, Texas, 76019
Constant variance predicted vs. residual values with

IMPUTATIONS

PROJECT PURPOSE
To develop a tool that would, taking into account the patients

Identified variables in the database that had missing data

past and present medical history, aid doctors in prescribing a

as candidates for imputation. Using SPSS software, ran

treatment plan to adult patients with chronic pain. The Eugene

stepwise regressions to find the algorithm to fill in

McDermott Center for Pain Management at UT Southwestern

missing data in each individual variable. Used generated

Medical Center in Dallas utilizes a two-step treatment plan, and

constant variance plots to identify and remove outliers,

conducts a survey of pain at the midpoint (end of stage 1) and at

then ran regressions again.

the end of treatment (end of stage 2). This tool would be designed

EX. DEPENDENT VARIABLE: MID_OSW

using collected data from a database from the Pain Management

Remove patients with missing data from data set. Run

center at UTSW of about 300 patients who underwent the two-

stepwise linear regressions to determine contributing

step treatment plan.

variables and use B values to determine the algorithm for

TREATMENT PLAN SUMMARY

updated dataset (assures data set is complete enough to

cover a broad spectrum)

mid_OSW

Stages:

Coefficientsa

Stage 1: First round of treatment is administered based on the

patients history and other characteristics.

Unstandardized Coefficients
Model

(Constant)

Stage 2: An evaluation is conducted halfway through the

treatment plan. Based on outcome, plan is adjusted or kept the
same

Std. Error

Pain is evaluated based on outcome questionnaires including:

1.214

pre_ProcGr3

3.514

1.032

litigat

3.515

phydx6
post_RxGr5

-Beck Depression Inventory (BDI)

Sig.
11.154

.000

.188

3.404

.001

1.288

.151

2.729

.007

3.047

.906

.189

3.362

.001

1.252

.504

.139

2.482

.014

.030

.010

.161

2.876

.004

post_ProcGr5

2.853

.938

.177

3.042

.003

mid_ProcGr4

-3.067

1.100

-.164

-2.788

.006

-11.220

5.360

-.116

-2.093

.037

SghxGr4
a.

Beta

13.536

duration

Monitoring patients pain: pre, mid, post, 1-year.

Standardized Coefficients

Check dataset for normality, outliers (Observed vs. Expected Cum Prob)

Dependent Variable: mid_OSW

Mid_OSW=13.536+3.514(Pre_ProcGr3)+3.515(litigat)+3.047(phydx6)+1.252(post_RxGr5)+.03(duration)+2.853(post_ProcGr5)

-Oswestry Pain Disability Questionnaire (OSW)

Run stepwise again using only the variables pulled from

-Pain Drawing Analogue (PDA)

the first regression to remove any unnecessary variables

Coefficientsa

Types of treatment:

Medications 21 types.
Procedures for pain 21 types.

Unstandardized Coefficients

Model
6

PERSONAL CONTRIBUTIONS
Researched all drugs and procedures used in the treatment plan to
find possible harmful interactions if administered concurrently.

B
(Constant)

pre_ProcGr3

3.233

1.051

phydx6

2.660

post_RxGr5

litigat

RxGr2

RxGr1

RxGr3

RxGr4

RxGr5

RxGr6

RxGr7

RxGr8

1+5

1+21

1+14

1+5

0+7

1+1

16

3+37

2+14

0+4

38

RxGr2

ProcGr1

ProcGr2

ProcGr3

ProcGr4

ProcGr5

ProcGr6

ProcGr7

ProcGr8

ProcGr9

88

82

8+23

0+14

2+16

0+4

80

72

RxGr4

1+21

4+25

4+35

4+16

1+12

0+5

60

60

RxGr5

1+14

3+37

5+16

3+17

1+7

78

72

RxGr6

1+5

2+14

0+8

1+6

22

22

RxGr7

0+7

0+3

18

20

RxGr8

1+1

ProcGr1

0+4
16

ProcGr2
ProcGr3

5+16

2+16

1+12

3+17

0+8

0+4

0+5

1+7

1+6

0+3

38

20

VIF

10.748

.000

.172

3.075

.002

.979

1.021

.912

.165

2.918

.004

.965

1.036

1.205

.514

.133

2.346

.020

.956

1.046

3.296

1.309

.141

2.518

.012

.981

1.019

.026

.010

.141

2.521

.012

.985

1.015

1.946

.909

.120

2.142

.033

.975

1.025

drug interaction matrix will later be utilized to create an

algorithm for the decision support tool. This algorithm

0
56

ProcGr5

will identify the key interactions between drugs and

procedures that result in better pain management, i.e.

1
1

ProcGr7

ProcGr8

ProcGr9

88

80

60

78

22

18

ProcGr10

82

72

60

72

22

20

Key:

Yellow

Mild-monitor closely

Green

Essential cousage

Orange

Moderate-find alternative

Teal

Recommended cousage

Red

Severe-contraindicated

Blue

Optional cousage

Purple

Only severe w/ Lithium

Pink

Applies to all but Bedrest

0
0

1
0

better scores on the outcome questionnaires. In the

future, this tool will aid doctors in prescribing a

212
56

X+Y

CONCLUSION
The dataset that I have helped to complete along with the

ProcGr4

ProcGr6

View individual partial plots, check for constant variance.

4+25

4+16

Tolerance

ProcGr10

0+14

Sig.

between 1 and 4, preferable if closer to 1

1+5

20

Also run collinearity diagnostics. Confirm VIF values are

RxGr3

8+23 4+35

Collinearity Statistics

a. Dependent Variable: mid_OSW

promote prescription of drugs with positive cousages.

RxGr1

Beta
1.139

post_ProcGr5

treatments to prevent harmful interactions among treatments or to

Std. Error
12.247

duration

Created a matrix that the tool would utilize when prescribing

Standardized Coefficients

212

X=Only those 2 prescribed together

Y=Prescribed with a mix of other drugs

Center on Stochastic Modeling, Optimization, & Statistics

Generate a constant variance plot to check for outliers. If

personalized treatment plan consisting of a combination

they exist, find and remove them from the dataset. Run

of medication and procedures for adult patients

regression again with updated dataset

experiencing chronic pain.