Testing and Reducing Skindex-29 Using Rasch Analysis: Skindex-17
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
ORIGINAL ARTICLE
Testing and Reducing Skindex-29 Using Rasch
Analysis: Skindex-17
Tamar E.C. Nijsten1, Francesca Sampogna2, Mary-Margaret Chren3 and Damiano D. Abeni2
The Skindex is a well-studied dermatology-specific health-related quality of life (HRQOL) instrument. The
objective of this study was to test Skindex-29 using Rasch analysis and, if necessary, to refine it so that it would
fit this item response theory based model. The Skindex-29 of 454 Italian dermatological patients was subjected
to Rasch analysis to investigate threshold order, differential item functioning (DIF), and item and overall fit to
the model. The Skindex-29 did not fit the Rasch model (Po0.001). The 5-point scoring system was re-grouped
into three categories and demonstrated logical response order for all but one item. Rasch analyses of a
combined emotion and social functioning subscale of Skindex-29 resulted in a 12-item psychosocial subscale.
Five of seven items were retained in a symptoms subscale. Both subscales fitted the model (P ¼ 0.32 and 0.13,
respectively) without significant individual item misfit or DIF (P40.05). Classical psychometric properties such
as response distribution, item–rest correlation, item complexity, and internal consistency of the two subscales of
Skindex-17 were at least adequate. The Skindex-17 is a Rasch reduced version of Skindex-29, with two
independent scores that can be used in the measurement of HRQOL in dermatological patients.
Journal of Investigative Dermatology (2006) 126, 1244–1250. doi:10.1038/sj.jid.5700212; published online 16 March 2006
INTRODUCTION Prieto et al., 2003; Nijsten et al., 2006), this approach is
The Skindex is a health-related quality of life (HRQOL) unable to solve some fundamental measurement issues such
instrument designed to measure the effects of skin disease on as response order (i.e., logical ordering of the response
patients’ lives. This dermatology-specific questionnaire has categories), additivity, which requires unidimensionality of
been extensively studied and refined in different population the measurement, and differential item functioning (DIF)
samples (Chren et al., 1996, 1997; Abeni et al., 2002; (assessing the effect of external factors, including diagnosis,
Augustin et al., 2004). Although the original Skindex on item response) (McHorney, 1997; Michell, 2003; Tesio,
consisted of 61 items (Chren et al., 1996), a refinement study 2003; Tennant et al., 2004a).
demonstrated that Skindex-29 decreased respondent burden Recently, item response theory-based models such as the
and improved its discriminative and evaluative capability Rasch model have become widely available, which should
(Chren et al., 1997). The methodology in both the original encourage psychometric research to address some of these
and the refinement study was based on classical test theory to fundamental measurement issues (Rasch, 1960; McHorney,
retain or discard items. 1997; Michell, 2003; Streiner and Norman, 2003; Tennant
For decades, classical test theory has been the most et al., 2004a). Basically, these models attempt to ensure
commonly used approach to create and reduce patient-based the transition from representational (nominal, ordinal, inter-
assessments, and includes multiple psychometric features val, and ratio scales) (Stevens, 1946) to fundamental
(Chren, 1999; Guyatt et al., 2002; Nijsten et al., 2006). measurement (i.e., weight (g), distance (m), or temperature
Although classical test theory has been proven to be a valid 1C) (Michell, 1997, 2003; Chren, 2005). This transition is
and reliable methodology and there is no consensus of how important because it allows investigators to calculate mean
to reduce existing HRQOL instruments (Coste et al., 1997; scores and change of scores without the restrictions
associated with nonparametric, representational measure-
1
Department of Dermatology, Erasmus Medical Center, Rotterdam, The
ment. Rasch analysis is one among the most commonly used
Netherlands; 2Health Services Research Unit, Dermatological Institute item response theory models to create new or test existing
IDI-IRCCS, Rome, Italy and 3Department of Dermatology, University HRQOL instruments (Tennant et al., 2004a).
of California at San Francisco, San Francisco, California, USA The overall goal of this work was to use item response
Correspondence: Dr Damiano D. Abeni, Health Services Research Unit, theory to examine the responses to Skindex-29 of a large
IDI – IRCCS, via dei Monti di Creta, 104, I-00167 Rome, Italy.
E-mail: d.abeni@idi.it
sample of patients. In this study, we used the Rasch model to
Abbreviations: DIF, differential item functioning; HRQOL, health-related
test threshold order, item fit, and DIF. In addition, we aspired
quality of life to create a reduced version of this instrument that fitted the
Received 26 September 2005; revised 12 December 2005; accepted 3 Rasch model and behaved psychometrically well among
January 2006; published online 16 March 2006 a heterogenous group of dermatological patients.
1244 Journal of Investigative Dermatology (2006), Volume 126 & 2006 The Society for Investigative DermatologyTEC Nijsten et al.
Rasch Reduced Skindex
Table 1. Demographic and disease characteristics of 1.0 0
4
the study population1
Probability
Total (n=454) Sample 1 Sample 2
No. of (n=227) No. of (n=227) No. of 0.5 2 3
patients (%) patients (%) patients (%)
Gender 1
Female 270 (59.7) 131 (57.7) 139 (61.7) 0.0
–3 –2 –1 0 1 2 3
Male 182 (40.3) 96 (42.3) 86 (38.2) Person location (logits)
Figure 1. The category probability curve for item 24 (‘‘sensitive skin’’)
Age (years) (n ¼ 454).
o45 225 (49.8) 110 (48.9) 115 (51.1)
45–65 162 (35.8) 86 (37.9) 76 (33.8)
probability curves demonstrated a disordered pattern bet-
465 65 (14.4) 31 (13.7) 34 (15.1) ween 1 and 2 for all items. In addition, 20 of 29 items
(69.0%) showed a logical transition between response
Diagnosis category 2 and 3 and 25 of 29 (86.2%) items had a good
Acne 151 (33.3) 73 (32.2) 78 (34.4) transition between category 3 and 4. However, re-grouping
the original Skindex scores into 0 (0), 1 (1 and 2), and 2 (3)
Psoriasis 76 (16.7) 40 (17.6) 36 (15.9)
and 3 (4) demonstrated reversed thresholds in 8 of the 29
Seborrheic 54 (12.0) 27 (11.9) 27 (11.9)
dermatitis
items. To avoid creating individual scoring systems for these
items or deleting them from the reduced instrument, the
Alopecia 46 (10.2) 25 (11.0) 21 (9.3)
areata
original response categories were re-grouped into three
levels, as 0 (0), 1 (1 and 2), and 2 (3 and 4), which resulted
Vitiligo 27 (6.0) 13 (5.7) 14 (6.2)
in ordered thresholds for all items, except item 30.
Nevi 100 (21.7) 49 (21.6) 51 (22.5)
Individual item fit and DIF for the psychosocial subscale.
Physician global assessment Although item 30 was the first to be discarded due to reversed
Very mild 64 (14.2) 30 (13.2) 34 (15.0) thresholds, it also misfitted the Rasch model (P ¼ 0.01)
Mild 160 (35.4) 87 (38.3) 73 (32.2) and demonstrated DIF for age (P ¼ 0.03) and diagnosis
Moderate 189 (41.8) 97 (42.7) 92 (40.5)
(P ¼ 0.0005). Each of the other eight items that were deleted
demonstrated significant item misfit in the Rasch model
Severe/very 31 (6.9) 13 (5.7) 18 (7.9)
severe (Pp0.02), except for item 12, which demonstrated significant
1
DIF for both age (P ¼ 0.008) and gender (P ¼ 0.003). Items 2
Totals may vary because of missing values.
and 19 showed significant DIF for diagnosis and item 2 for
age (Pp0.006). Items 3, 9, 13, 15, and 19 were primarily
discarded because they misfitted the Rasch model.
RESULTS Overall, the 12 remaining items of the psychosocial
Study population subscale fitted the Rasch model well (total-item w2 ¼ 39.45,
The majority of the 454 patients were young to middle-aged df ¼ 36, and P ¼ 0.32) (Table S1). After removing the ‘‘Rasch
women (Table 1). More than 60% of the participants were factor’’, a principal component analysis demonstrated that
diagnosed with an inflammatory skin disease such as acne, the first component accounted for only 16.3% of the
psoriasis, and seborrheic dermatitis, and almost half of the variance.
patients were graded as having at least moderate disease
severity. No significant differences of demographic or disease Individual item fit and DIF for the symptoms subscale.
characteristics were detected between the two random Although the overall fit for the seven symptoms-related items
samples of the study population. was acceptable (total-item w2 ¼ 40.28, df ¼ 35, and P ¼ 0.13),
item 24 was deleted due to significant DIF for gender
Rasch analysis (Po0.001) and age (P ¼ 0.02). Subsequently, item 7 misfitted
Threshold order for all items. The threshold of the original the model (residual ¼ 1.40 and P ¼ 0.01) and was discar-
5-point scoring system of Skindex-29 was significantly ded. The remaining items showed a good overall fit to the
disordered for all 29 items in sample 1. For example, the Rasch model (total-item w2 ¼ 29.31, df ¼ 25, and P ¼ 0.25)
probability that a person with little HRQOL impairment (Table S2). After extracting the Rasch factor, the second factor
scored 0 was superior to that of scoring 1, which is defined as only reflected 27.7% of the variance of this subscale.
threshold order, but those with little impairment were more
likely to score 2 than 1, which is defined as threshold Validation of Rasch analyses
disorder, for item 24 (Figure 1). The order between 2, 3, and 4 The items retained in the analyses using sample 1 (presented
appeared to be logical for this item. In fact, the category data) or 2 were comparable, except that the validation
www.jidonline.org 1245TEC Nijsten et al.
Rasch Reduced Skindex
sample did not reject item 2 in the psychosocial and 7 in the original corresponding subscales (rX0.83). The items of
symptoms subscale. The overall summary fit statistics of both Skindex-17 of both subscales represented 94% of the vari-
groups of patients were nonsignificant (P ¼ 0.32 vs 0.16 and ance of the total score of Skindex-29 (adjusted R2 ¼ 0.94).
P ¼ 0.25 vs 0.12 for the psychosocial and symptoms subscale, Both new subscales captured at least 87% of the variance
respectively), with similar Cronbach’s a’s (0.81 vs 0.83 and of their original Skindex-29 subscales (adjusted R2 scores of
0.77 vs 0.78 for the psychosocial and symptoms subscales, the psychosocial items were 0.95 and 0.87 for functioning
respectively). Also, both Rasch reduced subscales captured and emotions, respectively, and 0.87 of the reduced
more than 85% of the total variance of the original Skindex- symptoms subscale compared to its original version).
29. Adding item 2 to the retained psychosocial items of
sample 1 resulted in a poor fit to the Rasch model (P ¼ 0.002 Item behaviour of Skindex-17. Of the 12 items of the psycho-
and item 2 showed item significant misfit and DIF for age) social subscale, five items (8, 14, 17, 24, and 27) showed a
and adding item 7 to the symptoms items of sample 1 resulted suboptimal response distribution with more than 70% of the
in borderline overall fit (P ¼ 0.06), but with individual item 514 participants who scored ‘‘0’’. The ‘‘floor’’ effect of these
misfit. Therefore, items 2 and 7 were not included in the items disappeared after deleting the 100 patients with nevi
final Rasch reduced Skindex, which we named Skindex-17 (21.6% of the total study population). The correlation
(Appendix S2). coefficients between the 12 items of the psychosocial
subscale ranged from 0.31 to 0.64 and for the symptom
Classical psychometric properties of Skindex-17 subscale from 0.33 to 0.57. Each of the retained items
Validity of Skindex-17. The face validity of Skindex-17 correlated moderately (rX0.40), with at least one other item
appeared to be adequate because it distinguished symptoms in its subscale (data not shown) suggesting item redundancy.
from functioning and emotions, which are likely to show For both subscales, the Cronbach’s a’s were X0.70 for the
some extent of overlap. As hypothesized, patients with acne, items and the item–rest correlation coefficients were X0.35,
psoriasis, and seborrheic dermatitis reported significantly suggesting good internal consistency and reasonable ‘‘homo-
higher scores for both subscales of Skindex-17 than nevi geneity’’ (Table 2). None of the items of the psychosocial
patients (Po0.001) (Table S3). The new Skindex-17 subscales subscale correlated equally or stronger for the symptoms
correlated very well to excellent with Skindex-29 and its subscale and vice versa, which indicates good item discrimi-
Table 2. Classical psychometric properties of the Skindex-17 (n=454)1
Factor analysis2
Factor 1, psychosocial Factor 2, symptom subscale Item–rest
Items subscale (eigenvalue=4, 7) (eigenvalue=1, 5) Uniqueness Crohnbach’s a correlation
4 ‘‘work or hobbies’’ 0.61 0.41 0.39 0.93 0.67
5 ‘‘social life’’ 0.79 0.23 0.26 0.92 0.80
6 ‘‘depressed’’ 0.64 0.27 0.43 0.93 0.66
8 ‘‘stay at home’’ 0.74 0.19 0.36 0.93 0.72
11 ‘‘closeness with loved ones’’ 0.74 0.26 0.34 0.92 0.75
14 ‘‘do things by themselves’’ 0.65 0.25 0.42 0.92 0.67
17 ‘‘showing affection’’ 0.69 0.21 0.39 0.93 0.70
21 ‘‘embarrassed’’ 0.73 0.25 0.30 0.92 0.73
23 ‘‘frustrated’’ 0.75 0.15 0.35 0.93 0.73
25 ‘‘desire to be with people’’ 0.83 0.09 0.29 0.93 0.78
26 ‘‘humiliated’’ 0.72 0.08 0.39 0.93 0.68
29 ‘‘sex life’’ 0.57 0.12 0.60 0.93 0.56
1 ‘‘hurts’’ 0.42 0.54 0.49 0.74 0.52
10 ‘‘itches’’ 0.09 0.67 0.51 0.72 0.55
16 ‘‘bothered by water’’ 0.14 0.43 0.72 0.78 0.40
19 ‘‘irritated’’ 0.33 0.58 0.53 0.72 0.57
27 ‘‘bleeding’’ 0.20 0.77 0.35 0.70 0.35
1
Except for factor analysis, the analyses were performed for the psychosocial and symptom subscales separately.
2
Principal component analysis followed by orthogonal rotation. Factors were retained if eigenvalue41.
Bold values indicate loadings X0.40.
1246 Journal of Investigative Dermatology (2006), Volume 126TEC Nijsten et al.
Rasch Reduced Skindex
nant validity (data not shown). Factor analysis of Skindex-17 that these different reduction approaches (patient-based vs
confirms its two-dimensional structure and separates the mathematics-based) may result in similar reduced instru-
psychosocial from the symptoms subscale (Table 2). Items 1 ments. Nevertheless, it would be interesting to compare the
and 4 demonstrate moderate item complexity because their outcomes of Skindex-17 and -16 in a population to study
loadings are X0.40. these different ways of model selection in more detail.
Categorization of Skindex-17 Effect of external factors on HRQOL impairment
For both subscales we performed mixture analysis to It has been demonstrated in the development of new HRQOL
categorize the two scores of Skindex-17 (Table S4). The for psoriasis and atopic dermatitis that DIF across demo-
scores of the psychosocial subscale varied between 1 and 26, graphic variables is an important consideration (McKenna
which could consist of three distinct distributions with mean et al., 2003; Whalley et al., 2004; Nijsten et al., 2005, 2006)
scores of about 2, 7, and 15, which were subsequently and Skindex-17 is the first dermatology-specific instrument
categorized as having little (o5), moderate (5–9), and high to address the issue of response differences of patients in
(49) impact on patients’ lives. The total symptoms score functions of age, gender, diagnosis, and disease severity. Of
could be dichotomized; patients who reported 5 or more on interest, none of the items of Skindex-17 showed significant
this subscale could be categorized as experiencing a lot of DIF for education and marital status (data not shown). Of
symptoms. importance for a dermatology-specific HRQOL instrument is
that the items of Skindex-17 psychosocial subscale behave
DISCUSSION similarly among patients with different diagnoses even with
Skindex-17 vs Skindex-29 varying levels of clinical disease severity. For example, a
The Skindex-17 (Appendix S2) is derived from Skindex-29, psoriasis patient who has little HRQOL impairment responds
which is a well-designed and extensively studied HRQOL in a similar way to each of the items as a vitiligo patient with
instrument in dermatology. However, a Rasch reduced the same level of impairment. In contrast to the psychosocial
version of Skindex-29 has practical advantages such as subscale, the scores of the symptoms subscale should be
decreased response burden and theoretical advantages such compared between different skin diseases cautiously because
as fit to an item response theory model. This transition from symptoms scores depend on patients’ disease. Intuitively, it is
representational (Stevens, 1946) to fundamental measure- not surprising that items that assess skin symptoms behave
ment theory has multiple implications, such as that it ensures differently among different skin conditions. For example, DIF
unidimensionality, additivity, response order, DIF, and speci- analyses demonstrated that individuals with seborrheic
fic objectivity (Michell, 1997, 2003), and is considered to be dermatitis were significantly more frequently bothered by
the new standard in psychometric methodology (McHorney, water but significantly less by bleeding compared to patients
1997; Michell, 2003). Although the retained items with with psoriasis or acne, despite the same level of HRQOL
response categories in Skindex-17 are identical to the original impairment (Po0.001). Ideally, the symptoms subscale
items, the scores should be collapsed into a 3-point scoring should only be used to compare the impact of symptoms
system and the three subscales with an overall score have within and not between skin diseases. Although to a lesser
been replaced by two subscales (psychosocial and symptoms) extent than for diagnosis, the scores of the symptoms subscale
with separate summing scores. In HRQOL assessments, it is should be compared with caution across gender because
not uncommon to report multiple summing scores without a women are more likely to give higher scores, but not signi-
total overall score (Hemingway et al., 1997; Houssien et al., ficantly, than men for most symptoms items (Table S2).
1997) or to collapse scoring categories (Piccinelli et al., 1993; Measurement of (very) limited HRQOL impairment in
Nijsten et al., 2006). Collapsing or reducing response ‘‘benign’’ skin conditions such as nevi and (senile) warts
categories does not necessarily decrease the precision of an remains challenging, because generic and dermatology-
HRQOL instrument (Piccinelli et al., 1993; De Jong et al., specific instruments may not be sensitive enough. Despite
1997; ). The fact that the subscales of Skindex-17 reflected the floor effect of 9 of 12 psychosocial items and three of five
more than 85% of the total variance of its counterparts of symptoms items (data not shown) among nevi patients and
Skindex-29 and that the correlations between the (sub)scales that none of the respondents with nevi reported ‘‘often’’ or
was excellent suggest that deleting 12 items from Skindex-29 ‘‘all the time’’ for any of the 17 items (except for item 21,
has resulted in a minimal loss of information. Moreover, this ‘‘embarrassed’’), Skindex-17 was able to select those who
reduction process did not substantially alter the excellent indicated at least moderate psychosocial impairment and
psychometric properties of the original Skindex-29 (Chren those with more than a few symptoms (12 and 5% of the nevi
et al., 1997). patients, respectively). Re-doing the Rasch analysis in sample
Comparing Skindex-16, which focuses on degree of bother 1 without the 49 nevi patients resulted in exactly the same
rather than frequency and is shortened based on patients’ item retention with adequate fit to the model (P ¼ 0.25,
perspectives (Chren et al., 2001), with Skindex-17, which is Cronbach’s a ¼ 0.93). In the future, it would be interesting to
the result of the statistical model, is of interest. Of the 12 study the effect of cultural background on the HRQOL tools
items that remained relatively the same in Skindex-16 and that are used in international collaborations (Bullinger
-29, nine items (75%) were both in Skindex-17 and -16. et al., 1993; Tennant et al., 2004b). In addition to static
Overall, a simple and preliminary comparison suggests assessments, change in HRQOL impairment may also depend
www.jidonline.org 1247TEC Nijsten et al.
Rasch Reduced Skindex
on external factors such as age, education, and marital status groups: (1) acne; (2) psoriasis; (3) seborrheic dermatitis; (4) alopecia
(Hemingway et al., 1997; Unaeze et al., in press). Therefore, areata; (5) vitiligo; and (6) nevi. These diagnostic groups were
the effect of patients’ characteristics should be carefully selected because they represent relatively homogenous groups
evaluated before comparing HRQOL scores of different of diagnoses, may be associated with comparable HRQOL impair-
patient populations, as is done in clinical trials or surveys, ment, and consisted of at least 25 patients. As there were relatively
especially if the used tool has not been tested for DIF more patients with nevi, we randomly selected 100 of the 258
previously. individuals whose major concern was nevi. Subsequently, the 454
subjects were divided into two groups. Both selection procedures
Sensitivity analyses used the random number generator method. The sampling of
To test whether combining the functioning and emotion two smaller subsets was done because the w2 statistics used in the
subscales did affect our findings, we re-did the Rasch Rasch model are very sensitive to the sample size (if n is large, even
analyses accepting the three dimensions of Skindex-29 and slight deviations from unidimensionality may be statistically
analyzing each separately and combined. Neither approach significant) (Tesio, 2003) and it enabled us to compare the findings
improved the reduction process of this instrument (data not in both samples.
shown). For example, the emotion subscale did not fit the Age was categorized into three levels (o45; 45–64; and X65
overall Rasch model (Po0.0001), and even after shortening it years). The 5-point physician global assessment re-grouped into four
to three items based on item fit and DIF its overall fit to the levels ((1) very mild; (2) mild; (3) moderate; and (4) severe and very
Rasch model remained very poor (P ¼ 0.004). Alternatively, severe), because the very severe score was reported in only 1% of
Rasch analyses of the three subscales combined resulted in the patients. Ethical approval was granted by the institutional review
significant DIF for diagnosis (Pp0.006) for six out of seven board and the study was carried out in compliance with the
symptoms-related items. These a posteriori Rasch analyses Declaration of Helsinki Principles.
demonstrate how the selection of the included items affects
the result because item fit and DIF are determined by the total Statistics
amount of underlying construct (i.e., HRQOL impairment Rasch analysis. Statistical analyses have been described in
due to skin conditions), which entirely depends on the items detail previously (Nijsten et al., 2005, 2006). The same procedure
selected. This underlying methodology also plays a role in the of Rasch analysis was followed for samples 1 and 2. To ensure
order of deleting items that demonstrate significant item misfit a homogeneous and ordered scoring system, the threshold order
and/or DIF because it may change the outcome substantially. was examined for each of the 29 items. Owing to practical reasons,
the scoring categories were not altered for each item separately,
CONCLUSION but according to the most optimal ordering for the vast majority of
The Skindex-17 is a Rasch-reduced version of Skindex-29 that the items.
overcomes some of the fundamental issues in measurement. Based on a cluster analysis that included Skindex-29 (Sampogna
This reduced questionnaire behaved psychometrically very et al., 2004) that showed a high correlation between functional and
well in this study population and may be a valuable tool in emotional subscales, we combined these two subscales (psychoso-
the measurement of patient-based outcomes in dermatology, cial subscale), and analyzed them separately from the symptoms
but its psychometric properties and clinical value should first subscale (21 and 7 items, respectively). As the data of the two
be tested in other populations. subscales of sample 1 and 2 did not fit the rating scale model
(Po0.001), we used the unrestricted Rasch model for all analyses. In
MATERIALS AND METHODS the Rasch model, individual item misfit was present if an item
Participants and settings showed a significant w2 statistic or a large positive (42.5) or negative
The design of the study included the characteristics of its participants (o2.5) standardized residual. To ensure that responses to
who were all aged 18 or more and attending the outpatient clinics of individual items were unaffected by external factors (age, gender,
dermatology and dermatological surgery of a large Italian dermato- diagnosis, and disease severity) to the measurement tool, DIF
logical hospital on predetermined days. More than 80% of the analyses were performed (Angoff, 1993). For the symptoms subscale,
patients are referred to this institution by their general practitioner. DIF for diagnosis was not assessed because preliminary analyses
After signing an informed consent, the Italian version of Skindex-29, showed that all symptom-related items had significant DIF for
which was created according to the guidelines for the cross-cultural diagnosis (Po0.0001), except items 1 and 26 (P ¼ 0.38 and 0.05,
adaptation of the HRQOL measures (Guillemin et al., 1993) and respectively). Overall fit of the data to the model was assessed using
validated (Abeni et al., 2001, 2002), was administered (Appendix item–trait interaction.
S1). After the visit, the physician registered the diagnosis and rated To confirm the unidimensionality of the retained items that fitted
the severity of the skin condition by answering the question ‘‘In your the Rasch model, a principal component analysis was performed. If
experience, among all patients you have seen with this condition, the first factor after discounting the ‘‘Rasch factor’’ accounted for
how severe is this patient’s condition?’’, which was defined as the o40% of the variance, the scale was assumed to be unidimensional.
physician global assessment. The possible responses were: very
mild, mild, moderate, severe, and very severe. The 2,242 Classical psychometric features of Skindex-17. To test the
dermatological patients presented various dermatological diagnoses. validity of Skindex-17, we examined the relation between the
However, in the present study, we restricted the analyses to 454 original and the short version of the Skindex using Pearson’s
patients whose diagnoses were part of the following diagnostic correlation coefficients (r) and the adjusted R2 in a linear regression
1248 Journal of Investigative Dermatology (2006), Volume 126TEC Nijsten et al.
Rasch Reduced Skindex
model. Construct validity was assessed by testing the assumption Table S4. Categorization of the psychosocial and symptoms subscale of
Skindex-17 based on mixture analysis (n ¼ 454)*.
that patients with inflammatory dermatoses such as psoriasis and
Appendix S1. Skindex-29 items.
acne would have higher scores than patients with nevi. In addition,
Appendix S2. Skindex-17 items.
multiple classical test theory -based psychometric features of
Skindex-17 and its items were studied, including response distribu-
REFERENCES
tion, item–rest correlation, item discriminant validity, item complex-
Abeni D, Picardi A, Pasquini P, Melchi CF, Chren MM (2002) Further
ity, and Cronbach’s a’s (Nijsten et al., 2005, 2006). A principal
evidence of the validity and reliability of the Skindex-29: an Italian study
component analysis followed by oblique rotation was performed to on 2,242 dermatological outpatients. Dermatology 204:43–9
test the proposed subscaling of Skindex-17. Factors with eigenvalues Abeni D, Picardi A, Puddu P, Pasquini P, Chren MM (2001) Construction and
of 1 or more were retained. validation of the Italian version of Skindex-29, a new instrument to
measure quality of life in dermatology (in Italian, abstract in English).
Categorization of scores. The scores of the new Rasch reduced G Ital Dermatol Venereol 136:73–6
psychosocial and symptoms subscale of the Skindex were entered Angoff WH (1993) Perspectives on differential item functioning methodology.
In: Differential item functioning (Holland PW, Wainer H, eds), Hillsdale,
into a mixture analysis using computer-assisted mixture analysis
NJ: Lawrence Erlbaum
(Haughton, 1997). This statistical method detects whether multiple,
Augustin M, Wenninger K, Amon U, Schroth MJ, Kuster W, Chren M et al.
distinct subdistributions are present in the distribution of the scores (2004) German adaptation of the Skindex-29 questionnaire on quality of
of all the individuals and would, therefore, facilitate the interpreta- life in dermatology: validation and clinical results. Dermatology 209:
tion of the new scores. We used the same procedure as described 14–20
previously for the categorization of another HRQOL instrument Boehning DB, Schlattmann P (2005) Computer assisted analysis of mix-
tures (C.A.MAN). Available at: ftp.ukbf.fu-berlin.de/sozmed/caman.html
(Nijsten et al., 2006). In brief, we estimated the weights for the
(accessed 20 August)
maximum number of grid points suggesting Poisson distributions,
Bullinger M, Anderson R, Cella D, Aaronson N (1993) Developing and
which were subsequently used as starting values for the expecta- evaluating cross-cultural instruments from minimum requirements to
tion–maximization algorithm. Then, a much smaller number of optimal models. Qual Life Res 12:451–9
weights and support points were computed with nonparametric Chren MM (1999) Understanding research about quality of life and other
maximum likelihood estimations of 1 and by combining support health outcomes. J Cutan Med Surg 3:312–6
points that are very close to each other or have very low weights. The Chren MM, Lasek RJ, Sahay AP, Sands LP (2001) Measurement properties of
support points and weights that remained were then entered in a Skindex-16: a brief quality-of-life measure for patients with skin diseases.
J Cutan Med Surg 5:105–10
fixed mixture to classify each observation in one of the different
Chren MM (2005) Measurement of vital signs of skin diseases. J Invest
mixture components using posterior probabilities. Dermatol 125:vii–ix
All statistical tests were two sided. Significance was assessed at Chren MM, Lasek RJ, Flocke SA, Zyzanski SJ (1997) Improved discriminative
an a level o0.05, with the exceptions outlined above. Stata version and evaluative capability of a refined version of Skindex, a quality-of-life
7.0 (Stata Corp., College Station, TX) was used to estimate instrument for patients with skin diseases. Arch Dermatol 133:
Cronbach’s a, the correlation coefficients, and the principal axis 1433–40
analyses. SPSS version 10.0 for Windows (SPSS Inc., Chicago, IL) Chren MM, Lasek RJ, Quinn LM, Mostow EN, Zyzanski SJ (1996) Skindex,
a quality-of-life measure for patients with skin disease: reliability,
was used to randomize the sample in two groups. Rasch analyses validity, and responsiveness. J Invest Dermatol 107:707–13
were performed using RUMM2020 (RUMM Laboratory Pty Ltd,
Coste J, Guillemin F, Pouchot J, Fermanian J (1997) Methodological approa-
Perth, Australia). To categorize the Rasch version of the Skindex, ches to shortening composite measurement scales. J Clin Epidemiol
computer-assisted mixture analysis 2.0 was used (Boehning and 50:247–52
Schlattmann, 2005). de Jong Z, van der Heijde D, McKenna SP, Whalley D (1997) The reliability
and construct validity of the RAQoL: a rheumatoid arthritis-specific
quality of life instrument. Br J Rheumatol 36:878–83
CONFLICT OF INTEREST
The authors state no conflict of interest. Guillemin F, Bombardier C, Beaton D (1993) Cross-cultural adaptation of
health-related quality of life measures: literature review and proposed
guidelines. J Clin Epidemiol 46:1417–32
ACKNOWLEDGMENTS
This study was financially supported, in part, by the Italian Ministry of Health, Guyatt GH, Osoba D, Wu AW, Wyrwich KW, Norman GR, Clinical
Rome, Italy. Dr Chren was supported by an Independent Scientist Award Significance Consensus Meeting Group (2002) Methods to explain the
(#K02 AR 02203-01) from the National Institute of Arthritis and Musculoske- clinical significance of health status measures. Mayo Clin Proc 77:
letal and Skin Diseases, National Institutes of Health. Tamar Nijsten was 371–83
supported by a grant from the Fund for Scientific Research-Flanders, Belgium Haughton D (1997) Packages for estimating finite mixtures: a review. Am Stat
(FWO-Vlaanderen). 51:194–204
Hemingway H, Stafford M, Stansfeld S, Shipley M, Marmot M (1997) Is the
SUPPLEMENTARY MATERIAL SF-36 a valid measure of change in population health? Results from the
Table S1. Individual item fit to the Rasch model and DIF for the items of the Whitehall II Study. BMJ 315:1273–9
psychosocial subscale of Skindex-17 across demographic and disease Houssien DA, McKenna SP, Scott DL (1997) The Nottingham Health Profile
characteristics (sample 1, n ¼ 227 patients). as a measure of disease activity and outcome in rheumatoid arthritis.
Table S2. Individual item fit to the Rasch model and DIF for the items of the Br J Rheumatol 36:69–73
symptoms subscale of Skindex-17 across demographic and disease character- McHorney CA (1997) Generic health measurement: past accomplishments
istics (sample 1, n ¼ 227 patients). and a measurement paradigm for the 21st century. Ann Intern Med
Table S3. For the total and the 2 random samples, the mean and median (25th 127:743–50
and 75th percentiles) of Skindex-17 scores among patients with different skin McKenna SP, Cook SA, Whalley D, Doward LC, Richards HL, Griffiths CE
conditions. et al. (2003) Development of the PSORIQoL, a psoriasis-specific measure
www.jidonline.org 1249TEC Nijsten et al.
Rasch Reduced Skindex
of quality of life designed for use in clinical practice and trials. Br J severity, quality of life, and psychological distress in patients with
Dermatol 149:323–31 psoriasis: a cluster analysis. J Invest Dermatol 122:602–7
Michell J (1997) Quantitative science and the definition of measurement in Stevens SS (1946) On the theory of scales. Science 103:677–88
psychology. Br J Psychol 88:355–83
Streiner DL, Norman GF (2003) Health measurement scales: a practical
Michell J (2003) Measurement: a beginner’s guide. J Appl Meas 4:298–308 guide to their development and use. 3rd edn. Oxford: Oxford
Nijsten T, Unaeze J, Stern RS (2006) Refinement and reduction of the University Press
Impact of Psoriasis Questionnaire: classical test theory vs Rasch analysis. Tennant A, McKenna SP, Hagell P (2004a) Application of Rasch analysis in
Br J Dermatol. 16 January [E-pub ahead of print, doi:10.1111/j.1365- the development and application of quality of life instruments. Value
2133.2005.07066.X]
Health 7:S22–6
Nijsten T, Whalley D, Gelfand J, Margolis DJ, McKenna SP, Stern RS (2005)
Tennant A, Penta M, Tesio L et al. (2004b) Assessing and adjusting for cross-
The psychometric properties of the Psoriasis Disability Index in United
cultural validity of impairment and activity limitation scales through
States patients. J Invest Dermatol 125:665–72
differential item functioning within the framework of the Rasch model:
Piccinelli M, Bisoffi G, Bon MG, Cunico L, Tansella M (1993) Validity and
the PRO-ESOR project. Med Care 42:I37–48
test–retest reliability of the Italian version of the 12-item General Health
Questionnaire in general practice: a comparison between three scoring Tesio L (2003) Measuring behaviours and perceptions: Rasch analysis as a tool
methods. Compr Psychiatry 34:198–205 for rehabilitation research. J Rehabil Med 35:105–15
Prieto L, Alonso J, Lamarca R (2003) Classical test theory versus Rasch Unaeze J, Nijsten T, Murphy A, Ravichandran C, Stern RS (in press) Impact of
analysis for quality of life questionnaire reduction. Health Qual Life psoriasis on health related quality of life decrease over time: an 11-year
Outcomes 1:27 prospective study. J Invest Dermatol
Rasch G (1960) Probabilistic models for some intelligence and attainment Whalley D, McKenna SP, Dewar AL, Erdman RA, Kohlmann T, Niero M et al.
tests. Chicago: University of Chicago Press (reprinted, 1980) (2004) A new instrument for assessing quality of life in atopic dermatitis:
Sampogna F, Sera F, Abeni D, IDI Multipurpose Psoriasis Research on international development of the Quality of Life Index for Atopic
Vital Experiences (IMPROVE) Investigators (2004) Measures of clinical Dermatitis (QoLIAD). Br J Dermatol 150:274–83
1250 Journal of Investigative Dermatology (2006), Volume 126You can also read
