Rheumatology 2011;50:500505

RHEUMATOLOGY doi:10.1093/rheumatology/keq357
Advance Access publication 11 November 2010

Concise report
Measurement properties of the osteoarthritis of
knee and hip quality of life OAKHQOL questionnaire:
an item response theory analysis
Christophe Goetz1,2, Emmanuel Ecosse1,3, Anne-Christine Rat1,2,4,
Jacques Pouchot1,5, Joel Coste1,3 and Francis Guillemin1,2

Abstract
Objective. To further document the measurement properties of each domain of the OA of knee and hip
quality of life (OAKHQOL) questionnaire by a Rasch analysis.
Methods. The OAKHQOL self-administered questionnaire has been developed to assess health-related
quality of life in lower limb OA. Patients with various degrees of severity of knee or hip OA answered the
questionnaire. For each domain, their responses to the items were analysed with a Rasch family model,
the partial credit model. We examined the fit of data to model expectations, as well as assumptions of
unidimensionality and local independence. Invariance was assessed by analysis of differential item func-
tioning (DIF) by sex, age and joint. Analyses used the RUMM2020 software (Rumm Laboratory, Perth,
Western Australia).
Results. Responses for 544 patients were analysed: 297 had medically managed OA and 247 were
waiting for arthroplasty surgery. For the 40 items of the OAKHQOL, data analysis showed 5 with dis-
ordered thresholds and 9 with DIF (5 for joint, 3 for sex and 1 for age). Ten pairs of items showed local
dependence and four domains showed unidimensionality. Full-item domains and domains without the
misfitted items did not differ in patient-estimates data; therefore any bias at the item level is negligible
when considering the domain scores.
CLINICAL
SCIENCE

Conclusion. The five domains of the OAKHQOL questionnaire show good measurement properties by
Rasch analysis and provide valid scales.
Key words: Osteoarthritis, Quality of life, Item response theory, Rasch, Questionnaire.

Introduction measurement properties [1]. Because lower limb OA

Quality of life (QoL) instruments are widely used to meas-
ure the impact of diseases in terms of pain and symptoms
as well as physical functioning, mental health and social
functioning. Their usefulness is recognized once they have
undergone a rigorous validation process and show good
1
EA 4360 Apemac, Nancy University, Paris Descartes University, Paul
Verlaine University, Nancy, 2Department of Clinical Epidemiology and
Evaluation, INSERM CIC-EC CIE6, Nancy University Hospital, Nancy,
3
Biostatistics and Epidemiology Unit, Hopital Cochin, AP-HP, Paris,
4
Department of Rheumatology, Nancy University Hospital, Nancy and
5
Department of Internal Medicine, Hopital Europeen Georges
Pompidou, Assistance Publique, Hopitaux de Paris, Paris, France.
Submitted 8 April 2010; revised version accepted 23 September 2010.
Correspondence to: Francis Guillemin, Ecole de Sante Publique,
Faculte de Medecine BP 184, 54505 Vandoeuvre-les-Nancy, France.
E-mail: francis.guillemin@medecine.uhp-nancy.fr
affects specific aspects of QoL [2], specific instruments
may be useful [3].
The OA knee and hip QoL (OAKHQOL) is a self-
administered questionnaire specifically developed for
knee and hip OA [4]. Its factorial structure was assessed
by exploratory principal components analysis (PCA). Its
psychometric properties were evaluated according to
the classical test theory (CTT): construct validity, reprodu-
cibility and sensitivity to change [5]. Results were satisfac-
tory but the CTT did not consider important aspects of
measurement such as item properties, ordering of re-
sponse categories and invariance across conditions and
populations.
The item response theory (IRT) provides supplementary
methods to investigate the properties of health outcome
instruments [6, 7]. This theory describes the level of a

! The Author 2010. Published by Oxford University Press on behalf of the British Society for Rheumatology. All rights reserved. For Permissions, please email: journals.permissions@oup.com

subjects characteristics on the measured outcome
(underlying trait) as a function of responses to the items
as well as item properties. The simplest IRT model is the
Rasch model [8] and has proved useful to evaluate the
properties of health outcome measures [9]. We aimed to
further document and validate the measurement proper-
ties of each item of the OAKHQOL, using Rasch analysis.
Materials and methods
This non-interventional study was approved by the French
Consultative Committee for data processing in health re-
search (CCTIRS) and authorized by the National
Committee on Informatics and Freedom (CNIL). Written
consent was obtained from each included patient.
OAKHQOL scales
The OAKHQOL questionnaire contains 43 items and de-
scribes QoL in five domains: physical activities (16 items),
mental health (13 items), pain (4 items), social support
(4 items), social functioning (3 items); and 3 independent
items [5]. Each item is scored on a scale from 0 to 10.
When at least half the item scores in a domain are miss-
ing, the score for that domain is dropped. Scores are ob-
tained by computing the mean of the item scores for each
domain and normalized to a scale from 0 (worst) to 100
(best possible QoL).
Patients
We used data from a multicentre cohort constituted for
the first validation study of the OAKHQOL questionnaire
[4, 5] and for the AMISAT study [10]. Patients were
recruited between 2002 and 2005 from six rheumatology
and orthopaedic surgery outpatient clinics in Paris and in
the Lorraine region (France). Inclusion criteria were hip or
knee OA according to Altmans criteria [11], age >18
years and no other disabling disorder. Exclusion criteria
were an indication of total hip or knee replacement sur-
gery for a reason other than OA and another total hip or
knee replacement within the previous year.
For the analyses, patients at the post-operative stage
were excluded because their number was low, while pa-
tients medically treated and waiting for surgery were con-
sidered to represent a wide enough variety of OA severity.
Data for patients with both hip and knee OA or with miss-
ing data on the involved joint were excluded to avoid a
bias in analysis of the measurement invariance by the
involved joint. Data for questionnaires with >50% missing
data in a domain were excluded.
Rasch analysis
Analyses were conducted separately for each OAKHQOL
domain. We used a partial credit model [12], which
extends the Rasch model for polytomous items.
Distance between response categories was constrained
to be equal within each item, as an 11-category rating
scale shows near-interval-scale properties [13].
Fit of the data to the models was assessed according
to recent recommendations [7]. Details regarding the
www.rheumatology.oxfordjournals.org
OA of knee and hip quality of life
analyses are presented in supplementary data available
at Rheumatology Online. Item fit was explored with stan-
dardized residuals (expected to be between 2.5 and
+2.5, with a mean (S.D.) value of 0 (1), chi-square statis-
tics and examination of the item characteristic curves
(ICCs). The internal consistency of the domains was
examined with a person separation index (PSI) as defined
by Andrich and Douglas [14] A PSI > 0.85 is required to
use the scores at the individual level.
The invariance of the scales was assessed by looking
for differential item functioning (DIF) [15] across several
factors: sex, age (<60, 6070 and >70 years) and involved
joint (hip or knee). Items with DIF were split by group to
allow item parameters to be estimated separately in each
group.
Residual correlation matrices were examined to deter-
mine local independence of the scales. Local dependence
between two items is detected when the correlation of
their residuals is higher than that of the residuals between
most pairs of items. Unidimensionality was determined
through the PCA of the residuals.
Finally, to check whether the misfitting items of a
domain create a significant bias in the domain score, we
compared the person-estimates with and without these
items. A non-significant difference in person-estimates
(i.e. levels of QoL of patients) means that the misfitting
items do not create significant bias at the domain level.
The final sample size was 544 and met literature recom-
mendations [16, 17]. Rasch analysis was conducted using
RUMM2020 v4.1 software (Rumm Laboratory, Perth,
Western Australia). All P-values were adjusted according
to the Bonferroni method, and overall significance level
was set to 0.05.
Results
A sample of 648 patients was available. Data for
50 patients with >50% missing responses for a domain,
28 patients at a post-operative stage and 26 patients with
both hip and knee OA or missing data regarding the
involved joint were excluded from the Rasch analysis. Of
the remaining 544, 62% were women; 56 and 44% had
hip and knee OA, respectively; 55% were receiving med-
ical care and 46% were waiting for surgery. The item dif-
ficulties [level of QoL needed to answer positively to an
item (see supplementary data available at Rheumatology
Online)] demonstrated good coverage of the sample for
all domains except social functioning (see supple-
mentary figure 1, available as supplementary data at
Rheumatology Online).
Physical activities domain (16 items)
The domain showed a lack of fit to the model (Table 1).
Items 4, 5, 7, 8 and 10 showed a DIF for joint; Item 28 a
DIF for sex. The residuals and ICC suggested misfit for
Items 3, 6, 9, 14 and 28 (Table 2). Local dependence was
found between Items 4 and 5 (r = 0.42), Items 7 and
8 (r = 0.49) and Items 13 and 14 (r = 0.48). The PCA of
the residuals showed a significant deviation from unidi-
mensionality.
501
Christophe Goetz et al.

TABLE 1 Fit statistics for the models used for each OAKHQOL domain

Mean (S.D.) of fit residualsa Item trait interaction

Domain/model Items Persons Chi square Dfb P-value PSIc

Physical activities
Full-item domain 0.79 (2.402) 0.35 (1.59) 235.7 64 <0.0001 0.94
Eight-item set 0.07 (1.380) 0.62 (1.54) 53.2 32 0.05 0.89
Mental health
Full-item domain 1.01 (3.425) 0.16 (1.36) 267.3 52 <0.0001 0.93
Six-item set 0.18 (1.095) 0.30 (1.06) 42.3 24 0.002 0.88
Pain
Full-item domain 0.32 (1.702) 0.71 (1.33) 30.3 16 0.08 0.89
Social support
Full-item domain 0.39 (0.582) 0.47 (1.21) 76.0 36 0.006 0.91
Social functioning
Full-item domain 1.02 (1.540) 0.41 (1.18) 60.7 12 <0.0001 0.80

a
Expected mean and S.D. are 0 and 1, respectively. bDf: degrees of freedom. cPerson separation index.

Removal of items showing misfit, DIF or local depend-
ence one by one resulted in a set of eight items: Items 1,
2, 3, 6, 9, 11, 13 and 24 (fit of Items 3, 6 and 9 improved
during the iterative process and were thus kept in the
core set). This set of items showed good fit to the model
(Table 1), no local dependence was left and unidimension-
ality was respected. The comparison between person-
estimates of the core set and person-estimates of the
full-item domain showed that the difference was
non-significant, meaning that the removed items do not
create significant bias in the full-item domain.
Mental health domain (13 items)
The domain showed lack of fit to the model (Table 1).
Items 16 and 36 showed a DIF for sex and Item 29 a
DIF for age. The residuals and ICC suggested misfit for
Items 18, 20, 29 and 38 (Table 2). Local dependence was
found between Items 16 and 17 (r = 0.51), Items 19 and 20
(r = 0.32) and Items 36 and 37 (r = 0.41). Unidimensionality
was respected according to the PCA of residuals.
Removal of items showing misfit, DIF or local depend-
ence one by one resulted in a set of six items (Items 15,
17, 19, 21, 35 and 37) showing good fit to the model,
despite a significant chi-square statistic (Table 1). The dif-
ference between person-estimates of the reduced domain
and the full-item domain was non-significant, meaning
that the removed items do not create significant bias in
the full-item domain.
Pain domain (four items)
The domain demonstrated good fit to the model (Table 1).
No DIF was found for joint, sex or age. No item was mis-
fitting according to the residuals and the ICC (Table 2).
Residual correlations were found between Items 26
and 27 (r = 0.22) and Items 33 and 34 (r = 0.30).
Unidimensionality was respected according to the PCA
of residuals. Removing Items 27 and 34 left no local
dependence but had no impact on person-estimates.
Social support domain (four items)
The domain showed moderate fit to the model (Table 1).
No item was misfitting according to the residuals and the
ICC (Table 2). No DIF was found for joint, sex or age. Local
independence was respected; only a weak residual cor-
relation was found between Items 42 and 43 (r = 0.15).
Unidimensionality was respected according to the PCA
of residuals. Removing Item 43 left no local dependence
but had no impact on person-estimates.
Social functioning domain (three items)
The domain showed a lack of fit to the model (Table 1). No
DIF was found and no item was misfitting according to
the residuals and the ICC (Table 2). Local dependence
was found between Items 31 and 32 (r = 0.10). Removing
Item 32 left no local dependence but had no impact
on person-estimates. Unidimensionality was respected
according to the PCA of residuals.
Discussion
The Rasch analysis of the OAKHQOL confirms the validity
of the five scales, despite some misfitting items, that did
not compromise the measurement properties of the five
domains. Of 40 items, 9 showed DIF, 2 showed misfit and
10 pairs showed local dependence, but these findings do
not create a significant bias at the domain level. The target
population covered most of the severity levels of OA. For
all domains except social support (only three items), the
range of QoL for the sample was well covered by the item
difficulties. The questionnaire is thus well suited to dis-
criminate small variations in QoL for our target population.
This corroborates the good results obtained when map-
ping the OAKHQOL items with the OA core set of the

502 www.rheumatology.oxfordjournals.org
 OA of knee and hip quality of life

TABLE 2 Difficulties and fit statistics of the OAKHQOL items after splitting items showing DIF

Domain/item Difficulty Fit residual P-value

Physical activities
Q1. Walking 0.15 0.31 0.56
Q2. Bending or straightening 0.29 0.24 0.53
Q3. Carrying heavy things 0.30 3.19 0.03
Q4. Going down stairs (hip) 0.18 1.10 0.67
Q4. Going down stairs (knee) 0.41 0.50 0.71
Q5. Climbing stairs (hip) 0.16 1.67 0.76
Q5. Climbing stairs (knee) 0.40 0.93 0.18
Q6. Taking a bath 0.18 5.64 0.001*
Q7. Dressing (hip) 0.13 0.47 0.95
Q7. Dressing (knee) 0.25 0.29 0.63
Q8. Cutting toenails (hip) 0.33 1.85 0.51
Q8. Cutting toenails (knee) 0.03 0.02 0.39
Q9. Getting moving after staying in the same position 0.29 3.91 0.01
Q10. Getting in and out a car (hip) 0.22 1.06 0.41
Q10. Getting in and out a car (knee) 0.11 1.24 0.25
Q11. Using public transport 0.16 2.31 0.01
Q13. Need to spare oneself 0.11 0.42 0.80
Q14. Take longer doing things 0.15 2.61 0.06
Q24. Staying for a long time in the same position 0.32 0.66 0.84
Q25. Need a stick to walk 0.88 1.64 0.01
Q28. Need help (women) 0.37 0.13 0.13
Q28. Need help (men) 1.09 2.00 0.0002*
Mental health
Q15. Feel depressed because of pain 0.03 0.09 0.04
Q16. Been afraid of being dependent on others (women) 0.36 0.47 0.54
Q16. Been afraid of being dependent on others (men) 0.23 0.16 0.26
Q17. Been afraid of becoming an invalid 0.31 0.03 0.24
Q18. Embarrassed when people see me 0.23 2.02 <0.0001*
Q19. Worry 0.06 0.89 0.03
Q20. Feel depressed 0.26 1.45 <0.0001*
Q21. Hindered in family life 0.20 3.36 0.12
Q29. Feel older than my years (<60-years old) 0.12 3.25 0.56
Q29. Feel older than my years (>60-years old) 0.06 1.96 0.42
Q35. Wonder what is going to happen 0.06 1.54 0.005
Q36. Feel aggressive and irritable (women) 0.31 1.52 0.79
Q36. Feel aggressive and irritable (men) 0.10 0.12 0.34
Q37. Feel being a burden to close relatives 0.24 0.03 0.72
Q38. Worried about the side effects of treatment 0.14 6.08 <0.0001*
Q41. Feel embarrassed to ask for help 0.48 1.53 0.004
Pain
Q26. Frequency of pain 0.35 1.40 0.38
Q27. Intensity of pain 0.21 0.88 0.99
Q33. Having difficulties getting to sleep because of pain 0.26 1.97 0.002
Q34. Wake up at night because of pain 0.30 1.57 0.06
Social support
Q39. Talking about arthritis problems 0.01 0.32 0.04
Q40. Feel others understand arthritis problems 0.03 0.14 0.01
Q42. Feel support from people close to me 0.12 0.18 0.001*
Q43. Feel support from people around 0.10 1.22 0.39
Social functioning
Q30. Able to plan for the future 0.12 2.67 0.37
Q31. Going out whenever would like 0.07 0.38 <0.0001*
Q32. Have friends in whenever would like 0.05 0.76 0.0002*

When the fit residual was not between 2.5 and +2.5 or the chi-square test was significant, the misfit was confirmed by
examining the ICC. Items considered misfitting are shown in italics in the table. *Significant at 0.05 level after Bonferroni
adjustment.

www.rheumatology.oxfordjournals.org 503
Christophe Goetz et al.
International Classification of Functioning, Disability and
Health [18].
We used a conservative approach and retained in the
scales any misfitting item as long as the domain score
remained unbiased. Indeed, the Rasch model was origin-
ally considered a confirmatory tool [8] and may not be
appropriate to select items to be removed. Removing
items to obtain fit is not a guarantee that the scale im-
proved [19]. Moreover, removing items should imply a
consideration of the content validity and a validation
of the shortened instrument with an independent
sample [20].
Another issue justifying caution when interpreting misfit
of items is that fit results depend on the power of the
tests. Power increases with the sample size and variance
of the subjects (i.e. the PSI). With our large sample and
high PSI, even small differences between observed and
expected values may result in statistically misfit data [16].
This situation can explain why some chi-square statistics
remained significant, while the comparison of person-
estimates showed that the bias in the full-item domains
were negligible.
For every item demonstrating DIF, a clinical interpret-
ation can be found. For example, Items 4 and 5 show that
going down or climbing stairs is, as expected, more
difficult for people with knee than for those with hip OA.
In contrast, dressing, cutting toenails and getting in and
out of a car are more difficult with hip OA. Some other
explanations are obvious (feel older than my years and
DIF for age), some are more subtle (been afraid of being
dependent on others and DIF for sex), but all make sense.
In the same way, all observed issues of local dependence
appear between items that are conceptually close (e.g.
need to spare oneself and take longer doing things).
These are violations of the Rasch model assumptions,
but we showed that in the present case they do not
create a significant bias for the domain scores.
Conclusion
The five domains of the OAKHQOL questionnaire show
good measurement properties by Rasch analysis and pro-
vide valid scales to measure the specific QoL for patients
with hip or knee OA that is under medical management or
those waiting for surgery.
Rheumatology key messages
. The OAKHQOL questionnaire is valid for use in its
current five-scale format.
. The OAKHQOL shows good measurement proper-
ties through Rasch analysis.
Acknowledgements
Funding: This work was supported by the Clinical
Epidemiology Center, INSERM CIE6 (Institut National de
la Sante et de la Recherche Medicale), Health Ministry
504
PHRC (Programme Hospitalier de Recherche Clinique),
Nancy University Hospital, France.
Disclosure statement: The authors have declared no
conflicts of interest.
Supplementary data
Supplementary data are available at Rheumatology
Online.
References
1 Leidy NK, Revicki DA, Geneste B. Recommendations for
evaluating the validity of quality of life claims for labeling
and promotion. Value Health 1999;2:11327.
2 Bombardier C, Melfi CA, Paul J et al. Comparison of a
generic and a disease-specific measure of pain and
physical function after knee replacement surgery. Med
Care 1995;33:13144.
3 Hawker G, Melfi C, Paul J, Green R, Bombardier C.
Comparison of a generic (SF-36) and a disease specific
(WOMAC) (Western Ontario and McMaster Universities
Osteoarthritis Index) instrument in the measurement of
outcomes after knee replacement surgery. J Rheumatol
1995;22:11936.
4 Rat AC, Coste J, Pouchot J et al. OAKHQOL: a new
instrument to measure quality of life in knee and hip
osteoarthritis. J Clin Epidemiol 2005;58:4755.
5 Rat AC, Pouchot J, Coste J et al. Development and testing
of a specific quality-of-life questionnaire for knee and hip
osteoarthritis: OAKHQOL (OsteoArthritis of Knee Hip
Quality Of Life). Joint Bone Spine 2006;73:697704.
6 Edelen MO, Reeve BB. Applying item response theory
(IRT) modeling to questionnaire development, evaluation,
and refinement. Qual Life Res 2007;16:518.
7 Tennant A, Conaghan PG. The Rasch measurement model
in rheumatology: what is it and why use it? When should it
be applied, and what should one look for in a Rasch
paper? Arthritis Rheum 2007;57:135862.
8 Rasch G. Probabilistic models for some intelligence and
attainment tests. Copenhagen: Danmarks Paedagogiske
Institut, 1960.
9 Tennant A, McKenna SP, Hagell P. Application of Rasch
analysis in the development and application of quality of
life instruments. Value Health 2004;7:S226.
10 Baumann C, Rat AC, Osnowycz G et al. Do clinical
presentation and pre-operative quality of life predict
satisfaction with care after total hip or knee replacement?
J Bone Joint Surg Br 2006;88:36673.
11 Altman R, Alarcon G, Appelrouth D et al. The American
College of Rheumatology criteria for the classification and
reporting of osteoarthritis of the hip. Arthritis Rheum 1991;
34:50514.
12 Masters GN. A Rasch model for partial credit scoring.
Psychometrika 1982;47:14974.
13 Coste J, Walter E, Venot A. A new approach to selection
and weighting of items in evaluative composite measure-
ment scales. Stat Med 1995;14:256580.
14 Andrich D, Douglas GA. Reliability: distinctions between
item consistency and subject separation with the simple
logistic model. Paper presented at the Annual Meeting of
www.rheumatology.oxfordjournals.org

the American Educational Research Association.
New York, 1977.
15 Teresi JA, Fleishman JA. Differential item functioning
and health assessment. Qual Life Res 2007;16:
3342.
16 Smith AB, Rush R, Fallowfield LJ et al. Rasch fit
statistics and sample size considerations for polytomous
data. BMC Med Res Methodol 2008;8:33.
17 Linacre JM. Optimizing rating scale category effective-
ness. J Appl Meas 2002;3:85106.
www.rheumatology.oxfordjournals.org
OA of knee and hip quality of life
18 Rat AC, Guillemin F, Pouchot J. Mapping the osteoarth-
ritis knee and hip quality of life (OAKHQOL) instrument to
the international classification of functioning, disability and
health and comparison to five health status instruments
used in osteoarthritis. Rheumatology 2008;47:171925.
19 Gustafsson J. Testing and obtaining fit of data to the
Rasch model. Br J Math Stat Psychol 1980;32:20533.
20 Coste J, Guillemin F, Pouchot J, Fermanian J.
Methodological approaches to shortening composite
measurement scales. J Clin Epidemiol 1997;50:24752.
505