The Application of Rasch Models to HealthRelated Quality of Life HRQoL Assessment

1
The Application of Rasch Models to Health-Related
Quality of Life (HRQoL) Assessment
2
Outline

• Rasch models
• What are Rasch Models?
• Application in HRQoL
• Current work
• Development of an item-bank using Rasch Models
• Future research

3
Rasch Models (1)

• Latent trait models
• Family of statistical models
• Rasch models (1960) /
• Item-response theory models (IRT, Lord Novick,
  1968)
• Educational assessment

4
Rasch Models (2)

• Probabilistic models
• Probability or log-odds (logits)
• Person with given level (ability) of latent trait
  (e.g. anxiety) responding to a question (level of
  difficulty) in a given way
• Estimates (parameters) of
• Item difficulty or item location
• Person ability or person measure
• Continuum

5
Rasch Models (3)

• Independence (Thurstone, 1928)
• Item location estimates independent of sample
• Person measures independent of items
• Joint Maximum / Conditional Likelihood estimation
• Item and person locations along latent trait
• Probability of patient selecting response
  category
• Log(P/1-P) B D (dichotomous)

6
Student 1
Student 2
Student 3
Student 4
Student 5
Item 1
Item 2
Item 3
Item 4
Item 5
Increasing difficulty (Items) Ability (Persons)
7
Rasch Models (4)

• Constraints
• Unidimensionality
• One latent trait
• Principal components analysis of residuals
• Item independence
• Differential item functioning / item bias
• Item fit
• Do items (and persons) fit the Rasch Model?
• Is variance accounted for by a single latent
  trait?
• Interval scales
• Changes in score of equivalent dimensions are
  equal across all points of latent trait continuum

8
Current research

• Development of an item-bank for screening
  assessing psychological distress in cancer
  patients
• National Institute for Clinical Excellence
  (NICE). Guidance on Cancer Services (2004)Key
  recommendation 9
• Commissioners and providers of cancer services,
  working through Cancer Networks, should ensure
  that all cancer patients undergo systematic
  psychological assessment at key points and have
  access to appropriate psychological support.

9
Development of an Item Bank Background

• Psychological distress in cancer patients
• Prevalence
• Anxiety disorders 7 - 23 (Stark House, 2000)
• Depression 7 - 47 (Sellick Crooks, 1999)
• Existing measures
• Disappointing efficacy (Cull et al., 2001)
• Large number of questions
• Lack of relevance
• Floor ceiling effects
• No consensus about which measures are optimal for
  which patient group (Holland et al., 2001)
• Difficulty comparing interventions assessed with
  different outcome measures

10
Initial Development of an Item Bank Data

• Data collated from published (Velikova et al.,
  1999 Smith et al., 2002 Sharpe et al., 2004)
  unpublished studies conducted at
• CR UK Psychosocial Clinical Practice Group,
  St. Jamess University Hospital, Leeds
• CR UK Psychological Medicine Group, Western
  General Hospital, Edinburgh.
• 4910 cancer patients
• 83 items from 8 (touchscreen) questionnaires

11
Development of an Item Bank Items

• Hospital Anxiety Depression Scale (HADS,
  Zigmond Snaith, 1983)
• State-Trait Anxiety Inventory (STAI, Spielberger,
  1983)
• Beck Depression Inventory (BDI, Beck et al.,
  1996)
• Mental Health Inventory (MHI, Veit Ware, 1983)
• Patient Health Questionnaire (PHQ9, Kroenke
  Spitzer, 2002)
• General Health Questionnaire-12 (GHQ12, Goldberg
  Williams, 1991)
• Emotional Well-being scale of FACT (EWB, Cell et
  al., 1993)
• Emotional Functioning Scale of EORTC QLQ-c30
  (Aaronson et al., 1993)

12
(No Transcript)
13
Development of an Item Bank Items

• Rating scale model (Andrich, 1978)
• Common item equating
• Common items shared by each respondent
• Initial calibration of common item parameters
• Common item parameters used as anchors (held
  constant)
• Additional items added iteratively
• Item fit determined after each iteration
• Misfitting items removed
• Process repeated until all items have been added

14
Development of an Item Bank Results (1)

• Item fit
• 20 items removed due to misfit
• Remaining (63) items
• Cronbachs alpha 0.84
• Principal Components Analysis of residuals
• Unidimensional scale (70 variance)
• Good item fit (fit statistics lt 1.3)
• Items indicate non-specific distress

15
Development of an Item Bank Results (2)

• Item person locations
• Items
• Face validity
• I.e. questions concerning suicidal ideation
  (BDI/PHQ9) or worthlessness are harder to
  endorse than inability to work or fatigue
• Mean item location -0.50 logits (s.e. 0.07,
  range -5 to 3)
• Persons
• Mean person location -2.21 logits (s.e. 1.01,
  range -5 to 1)
• Insufficient overlap between between items and
  persons
• Mean item locations targeting higher levels of
  distress

16
Final Item and Person Locations
17
Threshold Map / Cutoffs

• Thresholds
• Differential Item Functioning
• Gender
• Age (3 groups)

• Clinical cutoff points
• SCAN / SCID

18
Further developments

• Positive and Negative Affect Scale (PANAS)
  Watson, Clark Tellegen (1988)
• 20 items
• Feelings emotions
• Distressed Excited Upset Nervous
• Last 7 days
• 5-point scale
• Very slightly or not at all (1) -gt Extremely
  (5)

19
Sample

• Patients treated with curative intent
• Recruited 67 centres in England
• Total 1315 patients
• breast (801, 61), prostate (330, 25),
  colorectal (17, 1), gynaecological (90, 7),
  Non-Hodgkins Lymphoma (65, 5) and unknown (12,
  1).
• Chemotherapy / radiotherapy

20
Sample

• Completed questionnaires after initial treatment
  6 months
• The Hospital Anxiety Depression scale
• Positive Negative Affect Scale (PANAS)
• Supportive Care Needs Survey (SCNS)
• Auxiliary SCNS module 'Access to Services'
• Health Concerns Questionnaire

21
Method

• Differential Item Functioning two samples
  (existing item bank vs KCL data)
• Difference item locations gt 0.5 logit
• Common item equating
• HADS anchor
• PANAS items added
• Fit Unidimensionality Internal reliability

22
DIF - HADS
23
Item Bank - Results

• 3 items PANAS removed (misfit)
• Guilty (6) Proud (10) Ashamed (13)
• Item bank
• 80 items
• Good fit (Infit MNSQ lt 1.3)
• Unidimensional
• 71 variance explained
• Cronbachs alpha (internal reliability) 0.85

24
Item / Person Map
Mean item -0.71
Mean ability -2.01
25
Development of an Item Bank Future Work

• Further analysis
• DIF
• Longitudinal
• Development of Computer-adaptive Test (CAT)
• Field work
• Comparison of CAT with
• Static questionnaire
• Psychiatric interview (e.g. PSE / SCAN)

26
Collaborations

• CR-UK Psycho-social Clinical Practice Group,
  Leeds
• University of Leeds
• CR-UK Psychological Medicine Group, Edinburgh
• Kings College, London