MAPPING THE DIABETES HEALTH PROFILE (DHP-18) ONTO THE EQ-5D AND SF-6D GENERIC PREFERENCE BASED MEASURES OF HEALTH Brendan Mulhern 1, Keith Meadows 2, Donna.

MAPPING THE DIABETES HEALTH PROFILE (DHP-18)

ONTO THE EQ-5D AND SF-6D GENERIC PREFERENCE BASED

MEASURES OF HEALTH

Brendan Mulhern1, Keith Meadows2, Donna Rowen1 & John Brazier1

1 Health Economics and Decision Science, University of Sheffield2 DHP Research & Consultancy Ltd, London

Contact: [email protected]

Contents

• Introduction• Measurement of HRQL and cost utility in diabetes

• Methods• Data

• Mapping techniques

• Results• Mapping functions developed to estimate EQ-5D utility

scores

• Mapping functions developed to estimate SF-6D utility scores

• Discussion19/04/23 © The University of Sheffield

Introduction• Diabetes interventions place a significant burden on health resources

• Approx 10% of the NHS budget

• Resource allocation informed by economic evaluation of new treatments

• Quality Adjusted Life Year (QALY) a recommended measure of outcome (NICE, 2008).

• Value for ‘quality’ derived using generic preference based measures• EQ-5D and SF-6D

• Generic measures should be included in trials to facilitate economic evaluations but are often not used

19/04/23 © The University of Sheffield

Mapping

• Mapping used to:• Predict generic measure utility values from existing condition

specific measures of HRQL using regression modelling

• Mapping is possible when • The generic measure performs well in the disease area

• EQ-5D and SF-6D validated in diabetes populations

• There is a correlation between the generic and condition specific HRQL measures

• This study predicts EQ-5D and SF-6D utility values scores using the Diabetes Health Profile-18 (DHP-18)

19/04/23 © The University of Sheffield

Generic preference based measures

19/04/23 © The University of Sheffield

• SF-6D• Derived from SF-36/SF-12

• 6 dimensions• physical functioning• role limitations• social functioning• pain• mental health• vitality

• Generates 18,000 health states

• 249 states valued using Standard Gamble

• Utility score range 0.29 to 1

• EQ-5D• Recommended by NICE for use

in cost utility analysis

• 5 dimensions with 3 response levels

• Mobility/self-care• daily activities• pain/discomfort• anxiety/depression

• Generates 243 (35) health states

• Selection of states valued using Time Trade Off

• Results modelled to produce single figure utility score, range -0.594 to 1

Diabetes Health Profile-18• Measures HRQL in diabetes

• Psychological distress

• Barriers to activity

• Disinhibited eating

• Demonstrates reliability, validity and patient acceptability

• 26 translations

• Completed using a range of media

• Diabetes-specific measure selected for the UK Department of Health PROMs Pilot for Long Term Conditions in Primary Care

• Visit www.diabeteshealthprofile.com for more information

19/04/23 © The University of Sheffield

Mapping specifications

19/04/23 © The University of Sheffield

• Model type• OLS; RE GLS; Tobit; Two part models

• Model performance indicators:• R2, Walt chi squared

• Mean absolute error and mean squared error

• Plots of observed and predicted scores

Number Model specification

1 DHP dimension scores

2 DHP dimension scores, DHP dimension scores squared

3 DHP dimension scores, DHP dimension scores squared, DHP dimension score interactions

4 DHP dimension scores, DHP dimension scores squared, DHP dimension score interactions, Age, Gender

5 DHP item scores

6 DHP item scores, DHP item scores squared

7 DHP item scores, DHP item scores squared, Age, Gender

Sample

• UK longitudinal dataset of a community-based postal survey• ≥18 years of age

• Data collected at baseline and 1 year

• Pooled data used for mapping

• Type 1 n=286; Type 2 n=2755

19/04/23 © The University of Sheffield

Characteristic Type 1 Type 2

M (SD) Range M(SD) Range

Age 59.65 (15.62) 18-93 66.19 (11.32) 26-98

Male 38.46% 60.87%

Diabetes related health complications 59.44% 38.05%

Other health complications 78.32% 79.79%

EQ-5D index (pooled) 0.60 (0.37) -0.59 to 1 0.65 (0.32) -0.43 to 1

SF-6D index (pooled) 0.66 (0.16) 0.35 to 1 0.69 (0.16) 0.35 to 1

DHP-18 Psychological distress (pooled) 28.16 (24.7) 18.61 (20.7)

DHP-18 Barriers to activity (pooled) 35.58 (21.2) 21.90 (19.4)

DHP-18 Disinhibited eating (pooled) 36.43 (23.2) 35.78 (22.8)

Results: Measure distributions

19/04/23 © The University of Sheffield

05

10

15

20

25

Pe

rcen

t

-.6 -.4 -.2 0 .2 .4 .6 .8 1EQ5Dindex

05

10

15

20

25

Pe

rcen

t

-.6 -.4 -.2 0 .2 .4 .6 .8 1EQ5Dindex

05

10

15

20

25

Pe

rcen

t

-.6 -.4 -.2 0 .2 .4 .6 .8 1SF6D

05

10

15

20

25

Pe

rcen

t

-.6 -.4 -.2 0 .2 .4 .6 .8 1SF6D

EQ-5D type 2

SF-6D type 2

EQ-5D type 1

SF-6D type 1

Model specifications: EQ-5D

19/04/23 © The University of Sheffield

• RE GLS model 7 (DHP-18 items, items squared, age and gender) performed best

Actual vs. predicted utility values

Actual vs. predicted utility values

• Type 1 (R2: 0.50) Type 2 (R2: 0.29) Mean absolute error Mean absolute error

Model specifications: SF-6D

19/04/23 © The University of Sheffield

Actual vs. predicted utility values (type 1)

Actual vs. predicted utility values (type 2)

• RE GLS model 7 (DHP-18 items, items squared, age and gender) performed best• Type 1 (R2: 0.65) Type 2 (R2:

0.40)

Discussion (1)

• Mapping an increasingly popular method for deriving utility scores

• EQ-5D and SF-6D utility scores can now be estimated for type 1 and type 2 diabetes

• GLS model 7 (DHP-18 items, squared item scores and demographics) performed best

• Type 1 predictions perform better than type 2 predictions

• SF-6D predictions perform better than EQ-5D predictions

• Over predicts utility for severe states and under predicts utility for mild states

19/04/23 © The University of Sheffield

Discussion (2)

• Using mapped values is second best to the direct inclusion of generic measures

• Lack external validity if not validated on external sample

• Using mapped values increases the error of the estimates

• How can mapping function precision be improved?

• Analyses using predicted values should consider the precision of the estimates.

19/04/23 © The University of Sheffield

Any questions?

Mapping algorithms available at: www.diabeteshealthprofile.com

19/04/23 © The University of Sheffield