Psychometric evaluation of the WHOQOL-BREF
in community-dwelling older people in Taiwan
using Rasch analysis
Wen-Miin LiangÆ Chih-Hung Chang Æ Yi-Chun Yeh Æ Haw-Yaw ShyÆ Hung-Wei Chen Æ Mau-Roung Lin
Accepted: 10 March 2009 / Published online: 2 April 2009
Ó Springer Science+Business Media B.V. 2009
Abstract
Objective To examine the psychometric characteristics of the brief version of the World Health Organization Quality of Life (WHOQOL-BREF) questionnaire in rural-com-munity-dwelling older people in Taiwan using Rasch analysis.
Methods This is a cross-sectional study. A total of 1200 subjects aged C65 years were recruited to complete the 26-item WHOQOL-BREF. Scale dimensionality, item
difficulty, scale reliability and separation, item targeting, item-person map, and differential item functioning (DIF) were examined.
Results The four WHOQOL-BREF scales (physical capacity, psychological well-being, social relationships, and environment) were found to be unidimensional and reliable. The item–person map for each domain indicated that the spread of the item thresholds sufficiently covered the latent trait continuum being measured. However, gaps in content coverage were identified in the social domain. Analyses of the DIF revealed that one psychological item (body image) exhibited DIF across the two age groups (old–old vs. young–old) and that two social items (sexual activity and friends’ support) displayed DIF across genders and the two age groups.
Conclusions Rasch analysis is a comprehensive method of psychometric evaluation of the WHOQOL-BREF and identifies areas for improvements. Three items displaying age-related DIF (body image, sexual activity, and friends’ support) may potentially cause biased health-related QOL assessments, and their impacts on scores should be care-fully examined.
Keywords Health-related quality of life
Item response theory Older people Rasch model WHOQOL-BREF
Abbreviations
CTT Classical test theory
DIF Differential item functioning
HRQL Health-related quality of life
IRT Item response theory
WHOQOL-BREF World Health Organization Quality of Life-BREF
W.-M. Liang
Graduate Institute of Biostatistics, Institute of Environmental Health, Biostatistics Center, China Medical University, Taichung, Taiwan, ROC
C.-H. Chang
Buehler Center on Aging, Health & Society, Division of General Internal Medicine, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA
C.-H. Chang
Graduate Institute of Biostatistics, China Medical University, Taichung, Taiwan, ROC
Y.-C. Yeh
Biostatistics Center, China Medical University, Taichung, Taiwan, ROC
H.-Y. Shy
Department of Mathematics, National Changhua University of Education, Changhua, Taiwan, ROC
H.-W. Chen
Institute of Environmental Health, China Medical University, Taichung, Taiwan, ROC
M.-R. Lin (&)
Institute of Injury Prevention and Control, Taipei Medical University, 250 Wu-Hsing Street, Taipei, Taiwan, ROC e-mail: mrlin@tmu.edu.tw
Introduction
With the increasing acceptability of the concept of health-related quality of life (HRQL) as a measure of healthcare outcomes over the past two decades, many HRQL instru-ments have been developed and refined [1,2]. In 1990s, the World Health Organization (WHO) initiated a worldwide 15-country project to develop a brief version of the WHO’s quality of life instruments (WHOQOL-BREF) [3, 4]. The WHOQOL-BREF, a generic HRQL instrument designed for healthy people and patients, has been rapidly introduced and utilized as an outcome measure in many countries and cultures [5–10]. It has shown excellent psychometric and clinimetric properties, such as test–retest reliability, inter-nal consistency, discriminant validity, construct validity, and responsiveness to community-dwelling older people, primarily based on the principles of classical test theory (CTT) [11].
Item response theory (IRT), originally developed in the fields of education and psychology, has proven to be a powerful tool for health outcome assessment [12,13]. This tool comprises a set of statistical models suitable for ana-lyzing a scale or survey instrument with multiple items that measure the same construct (e.g., physical functioning). An IRT model specifies how both person–trait level and item characteristics are related to a person’s item responses. This is different from the CTT approach in which items and the person latent trait being measured are considered separately and, therefore, cannot be meaningfully and systematically compared [14, 15]. Unlike the manner in which the trait level is estimated in the context of an IRT model, in the CTT method, true score estimates are typi-cally obtained by summing responses across items. In CTT, the two assumptions made to support the summed scores (each item within the same construct is valued equally ,and the score difference between the two adjacent response scales is uniform) do not hold in most situations [16]. These limitations can be solved rationally using IRT modeling even though IRT also makes assumptions, such as unidimensionality and local independence. Many useful statistics, such as the reliability and separation index, can be calculated directly from the IRT model, and differential item functioning (DIF) or item bias can be examined for measurement invariance [14,15].
The WHOQOL-BREF is a generic measure, and this enables it to be used in different cultures and disease groups. It can be an effective tool for assessing HRQL in the context of a cost-effectiveness evaluation and health intervention in a large-scale study, such as insurance pol-icies for different diseases and welfare polpol-icies for different age groups among older people [17, 18]. However, two aspects of the measurement need to be considered when applying a generic HRQL measure to older people. First,
the items and domains in a HRQL measure should take into account the aspects of life (access to health services and environmental safety) identified as important by older people [19]. Although the HRQL aspects considered by older people be relevant and important to health may be similar to those of young people, the relative importance and definition of each domain or facet (e.g., role func-tioning may be defined in terms of roles other than work) may differ substantially [20,21]. Second, extreme low or high scores on HRQL measures may appear to be common among older people if these measures are directly used without proper adaptation. Floor effects can potentially reduce the ability to detect distinguishing features, such as those between ill and very ill people, for example, and may also reduce the ability to detect changes in HRQL scores over time or after a health intervention [22]. In addition, it is important for physicians and healthcare decision-makers to be aware of potential biases in the instruments being used. For example, some response instability may have to do with the aging process when the study group comprises older subjects [23]. An accurate and precise scoring of the WHOQOL-BREF can lead to a more efficient allocation of healthcare resources or to more efficient screening of patients for specialized geriatric care. Since findings of previous WHOQOL-BREF studies were subject to the limitations of CTT methods, we have applied the Rasch model to examine psychometric properties of the WHO-QOL-BREF among community-dwelling older people in rural Taiwan. This is a population subgroup that has to date not received as much attention as urban-dwelling elders with regard to HRQL.
Methods
Study subjects and procedures
This was a cross-sectional study. Elderly subjects aged 65 and older were recruited from six of 15 villages of the Shin-Sher Township of Taichung County in west–central Taiwan. These villages were selected according to the order of the magnitude of populations of the Shin-Sher Township and were found to be representative of rural community-dwelling older people in Taiwan [11]. A postcard describing the aim of the study and the schedule of interviews was mailed to all eligible residents in the six villages. During a 2-week interview period, 1200 of the 2072 eligible subjects signed informed consent forms to participate and completed the WHOQOL-BREF. Of the 872 eligible elderly who did not participate, 24 had died, 59 were hospitalized or bed-ridden, 252 had moved out of the area, 323 were not at home during the assessment period, and 214 declined to be interviewed. The respondents had
similar distributions in gender and educational level, but tended to be younger (P = 0.073) when compared with non-respondents [11].
Each interviewer completed a 4-h training session to ensure that they followed standardized procedures in con-ducting the face-to-face interviews. Each enrolled participant was interviewed by a trained interviewer using the structured questions from the WHOQOL-BREF, which included questions related to demographic characteristics, life habits, and medical history.
The WHOQOL-BREF
The standard WHOQOL-BREF contains 26 items: two items from the overall QOL and general health facet and one item from each of the remaining 24 health-related facets [24]. The Taiwanese version of the WHOQOL-BREF (seeAppendix) was developed in compliance with the WHO guidelines on translation procedures as well as the design and selection of appropriate items [25]. The 28 items of the Taiwanese WHOQOL-BREF include two general items (i.e., G1: overall QOL; G4: general health), 24 items universally adopted for WHOQOL-BREF to cover four domains (namely, physical, psychological, social, and environment), plus two national items that were more specific for the culture of people of Taiwan (i.e., being respected/accepted among people, and eating what one loves to eat). This translated version showed good reliabilities (including internal consistencies of 0.70–0.77 and test–retest reliabilities of 0.76–0.80) and validities (including content, criterion, discriminant, predictive, and construct validities) [26–28]. Each domain score was cal-culated by multiplying the mean of all facet scores in the same domain by a factor of four, with a higher scoring indicating a better QOL (range 4–20). We used abbrevia-tions, with a single letter representing each of the four domains to denote items for ease of reference in this paper: P for physical, Y for psychological, S for social, and E for environmental. The number attached to the letter indicated the item number in the questionnaire. For example, S21 (satisfaction with sexual life) indicates item 21, which belongs to the social domain.
Rasch analysis
Partial credit model
We used the partial credit model, an extension of the dichotomous Rasch model, as it is suitable for an ordered polytomous response scale used in the WHOQOL-BREF [29]. As modeled, an item with five response categories would have four threshold parameters. At each threshold, a person has a 50/50 chance of choosing one category over
another [16]. For example, for a 5-point Likert scale (1: strongly dissatisfied, 2: dissatisfied, 3: moderately satisfied, 4: satisfied, 5: strongly satisfied), the first threshold is modeled as the value at which the probability of choosing a response of 2 (dissatisfied) over a response of 1 (strongly dissatisfied) is equal to 50%. The item difficulty estimates within each domain were standardized with a mean of 0 and a standard deviation of 1 (in logit units) [30].
Unidimensionality
In the Rasch model, item difficulty and a person’s latent trait are modeled to share the same scale with a single unit of logit, i.e., the unidimensionality of the scale [31, 32]. Confirmatory factor analysis was used to assess the prop-erty of unidimensionality using LISREL 8.72 (Scientific Software International, Lincolnwood, IL) [33]. Unidimen-sionality, the measurement of one underlying construct, was determined by the magnitude of factor loadings, with a value [0.3 indicative of importance, and three model fit indices—goodness-of-fit index (GFI), comparative fit index (CFI), Bentler–Bonett normed fit index (NFI)—with an index [0.9 indicative of good fit [34,35]. We also applied a two-index presentation strategy suggested by Hu and Bentler [36], using the indices standardized root mean square residual (SRMR) B0.08 and CFI [0.95 to confirm the model fit. After unidimensionality was established by confirmatory factor analysis (CFA), the Rasch model analysis was employed and the item infit statistic was used to further evaluate item–level model fit. An item with an infit statistic [1.4 or \0.6 was used to indicate the lack of fit in unidimensionality [37]. Specifically, an infit value [1.4 indicates that the item does not contribute to the same underlying construct or differs from other items in the same scale in its ability to discriminate among persons. An infit value \0.6 indicates that the item is muted or often has interdependence with other items in the same scale and may occur when several items are similar or highly cor-related or when one item is dependent on another [38].
Reliability and separation
Person reliability, another fit statistic of the Rasch model, was used to test the internal consistency among items at the domain level. Similar to Cronbach’s a, a value close to 1 indicates high internal consistency and a value \0.7 indi-cates model misfitting [16,39]. The person separation is a measure of the ability of items to discriminate subjects and was used to assess the ability of items to spread the elderly subjects along the HRQL continuum being measured in this study. Person separation statistics ranging between 1.5 and 2.0 were considered to be acceptable, 2.0–3.0 to be good, and [3.0 to be excellent [39].
Other aspects regarding item difficulty and personal ability
Targeting: Perfect targeting is defined as the equivalence of the average personal trait and the mean item difficulty. A targeting index, the average of personal ability, was used to examine whether the level of difficulty of each WHOQOL-BREF domain was appropriate for our sample. A targeting index [0 indicates that the subjects tend to give ‘positive’ responses (e.g., ‘satisfied’), and a value \0 indicates that the subjects tend to give ‘negative’ responses (e.g., ‘dis-satisfied’). Values of between 0.5 and 1.0 or between -1.0 and -0.5 are considered to be slight mis-targeting, and those [1 and \-1 are considered to be substantially mis-targeting [39,40].
Range and gap: The range between the highest and lowest threshold values of all items within a domain is considered to be good when it covers at least 95% of the levels of personal ability [41]. A gap is defined as the difference in two adjacent item difficulties which are C1 logit [42], and it implies that the item calibrations for a particular domain of the WHOQOL-BREF are not evenly spread or the number of items within the domain is not sufficient.
Ceiling and floor effects: The percentages of ceiling (highest) or floor (lowest) values among all subjects for each WHOQOL-BREF domain were used to assess the extent to which the latent HRQL trait of a subject is not reliably discriminated at both extremes. The ceiling effect for a particular WHOQOL-BREF domain is defined as the number of people with a level of personal latent trait greater than the highest (i.e., the fourth) threshold, and the floor effect as the number of people with a level of a personal latent trait less than the lowest (i.e., the first) threshold [41].
Item-person map: The relationship between item diffi-culties and personal latent traits for each domain was also examined by plotting the item difficulties and person measures together along the same line, referred to as an item-person map in the Rasch model analysis. The distri-bution of item difficulties allowed us to identify regions along the latent continuum that may be lacking items for reliable assessment.
Differential item functioning (DIF)
Differential item functioning refers to an item lacking measurement equivalence in different groups or settings [16]. In this study, sets of item difficulties were compared between genders (males vs. females) and between two age groups (young–old: 65–74 years vs. old–old: C75 years) to detect DIF. Due to our large sample size and the different facets in each domain of the WHOQOL-BREF, the results may be more sensitive to statistical testing to show a
significant difference in score comparisons. We therefore employed the same methodological approach as those used in other similar studies [7]. A criterion of 0.5 logits between item difficulties in different groups was applied to determine whether an item exhibited DIF [7,43]. All Rasch analyses were performed using Winsteps software ver. 3.47 [30].
Results
Table1shows the demographic characteristics of the study sample. This study cohort (n = 1200) consisted of elderly subjects aged between 65 and 103 years (mean 73.4 years); 59% (n = 709) were male, 86% (n = 1032) had an ele-mentary school of education or less, 66% (n = 792) lived with their spouse, 68% (n = 821) had been diagnosed with Table 1 Demographic characteristics of the study sample (n = 1200 community-dwelling older people)
Variable n Percentage Age 73.4 (6.04)a 65–103b Gender Male 709 59.08 Female 491 40.92 Education
Elementary school or below 1032 86.00 Higher than elementary school 168 14.00 Living with spouse
Yes 792 66.00
No 408 34.00
Number of physician-diagnosed chronic conditions
1 353 43 2 225 27 [3 243 30 Depression No 1135 94.74 Yes (GDS score [ 10) 63 5.26 Cognitive impairment No 980 81.67
Yes (MMSE score [ 18) 220 18.33 WHOQOL-BREF domain score
Physical 13.31 (2.25)a 0.77c Psychological 12.76 (2.14)a 0.79c Social 13.16 (1.92)a 0.73c Environmental 13.08 (1.99)a 0.79c GDS Geriatric depression scale, MMSE mini-mental state examina-tion, WHOQOL-BREF World Health Organization Quality of Life-BREF
a Mean, with the standard deviation given in parenthesis b Range
at least one chronic disease (353 with one; 225 with two; 243 with three or more), and 5% (n = 63) had suffered from depression. All of the four domains of the WHOQOL-BREF showed good reliability, with all Cronbach’s a val-ues[0.7. The mean QOL scores were 13.3 for the physical, 12.8 for the psychological, 13.2 for the social, and 13.1 for the environmental domain.
Unidimensionality
Domain-specific CFA results showed that the three indices for the two domains, psychological (GFI 0.98, CFI 0.97, NFI 0.96) and social (GFI 1.00, CFI 1.00, NFI 1.00), were [0.95. The SRMR were 0.03 and 0.01 for each of these two domains, respectively. The three indices for the other two domains, physical (GFI 0.98, CFI 0.97, NFI 0.97) and environmental (GFI 0.97, CFI 0.95, NFI 0.93), increased to C0.95 when one or two pairs of error covariance were
added separately for each of the two domains; the SRMR were 0.04 and 0.04, respectively. The standardized factor loading for each item in its respective domain, namely physical (0.28–0.83), psychological (0.44–0.76), social (0.47–0.80), and environmental (0.45–0.65), was [0.4 with one exception. Item ‘P4 dependence on medication or treatment’ had the lowest factor loading of 0.28. When the item P4 was excluded in the analysis, only negligible changes in GFI, CFI, and NFI statistics were found, indi-cating that the assumption of unidimensionality of each WHOQOL-BREF domain was confirmed.
Infit statistics and item difficulties
Table2 shows the model fit index, item difficulty esti-mates, and their standard errors as well as the person reliability and separation indices for each domain. All of the infit statistics fell in the 0.7–1.4 ranges, thereby
Table 2 Results of the Rasch analysis of the four WHOQOL-BREF domains P physica, Y psychologica, S social, E environmental, SE standard error Domain/Item Infit index Item difficulty (SE) Reliability (separation) Physical domain 0.76 (1.79)
P3 Pain and discomfort 1.05 -1.1 (0.04) P4 Dependence on medication or treatment 1.35 -0.6 (0.04)
P16 Sleep and rest 1.20 0.16 (0.05)
P17 Activities of daily living 0.72 0.21 (0.05)
P15 Mobility 0.90 0.42 (0.04)
P18 Working capacity 0.83 0.42 (0.05)
P10 Energy and fatigue 0.93 0.48 (0.05)
Psychological domain 0.77 (1.83)
Y26 Negative feeling 1.38 -1.19 (0.05) Y19 Self-satisfaction 0.97 -0.61 (0.06) Y11 Body image and appearance 0.95 -0.52 (0.06) Y6 Spirituality, religion and personal beliefs 0.82 0.37 (0.06) Y7 Thinking, learning, memory, and concentration 0.95 0.67 (0.05)
Y5 Enjoyment of life 0.87 1.29 (0.05)
Social domain 0.68 (1.45)
S22 Friends’ support 0.81 -0.77 (0.07) S20 Personal relationship 0.84 -0.71 (0.07) S27 Esteem and respect 1.24 0.31 (0.07)
S21 Sexual activity 1.04 1.17 (0.08)
Environmental domain 0.78 (1.86)
E28 Eating food 1.17 -1.08 (0.05)
E9 Physical environment 1.14 -0.63 (0.05) E23 Home environment 0.80 -0.56 (0.05)
E25 Transportation 0.96 -0.03 (0.05)
E24 Health and social care: availability and quality 0.96 0.01 (0.05) E8 Physical safety and security 0.99 0.1 (0.04)
E13 Opportunities 0.86 0.29 (0.05)
E12 Financial environment 1.00 0.94 (0.04) E14 Participation and support of leisure activities 1.15 0.96 (0.04)
satisfying the unidimensionality assumption of the IRT model. The items in each domain are listed in Table2 in ascending order of their difficulty estimates. For example, the item ‘P10 energy and fatigue’ was the most difficult item in the physical domain. The ranges of the average difficulty estimates of the physical, psychological, social, and environmental categories were -1.10–0.48, -1.19– 1.29, -0.77–0.17, and -1.08–0.96, respectively. Four items in the social domain had relatively large standard errors, with item ‘S21 satisfaction with sex life’ having the largest standard error of 0.08.
Reliability and separation
As shown in Table2, the person reliability for each of the WHOQOL-BREF domain was acceptable, with reliability coefficients ranging from 0.68 (social domain) to 0.78 (environmental domain). Separation indices for the physi-cal, psychologiphysi-cal, social, and environmental domains were 1.79, 1.83, 1.45, and 1.86, respectively, suggesting that all but the social domains had acceptable separation properties.
Item-person map
Figures1, 2, 3, and 4 show the four item–person maps depicting the person latent trait estimates (upper panel) and sets of threshold parameter estimates (represented by the numbers 1–4 for each item in the lower panel) jointly located along the same ‘logit‘ scale. For example, the fourth threshold of the item ‘P3 pain and discomfort’ shown in Fig.1 had a lower logit value (1.23) than the fourth threshold of the item ‘P10 energy and fatigue’ with a
higher logit value (5.30). This indicates that a subject with a latent trait level of between 1.23 and 5.3 logits is more likely to choose the fifth response category for item ‘P3 pain and discomfort’ and the fourth response category for item ‘P10 energy and fatigue’. Furthermore, the sequences of the four threshold parameters for each item were all in order from less (category 1) to more (category 4). When the data do not fit the model, the thresholds may be out of order (e.g., 1, 2, 4, and 3). It should be noted that the mean of the threshold parameters for each item is the item dif-ficulty estimate listed in Table2.
The average values of person traits (i.e., the targeting indices) were 0.57, 0.44, 1.02, and 0.36 for the physical, psychological, social, and environmental domains, respec-tively, indicating that the study population had a higher tendency to select more ‘positive’ response categories in each domain. However, very few subjects had a HRQL score greater than the highest threshold (0.5, 0.33, 0.17, and 0.5% for the physical, psychological, social, and environmental domains, respectively) or less than the lowest threshold (0.33, 0.00, 0.08, and 0.25 for the physical, psychological, social, and environmental domains, respectively), suggest-ing that there was no significant ceilsuggest-ing or floor effect for any of the four domains. In addition, the ranges of the thresholds in each domain (-3.9–5.3 for physical, -6.7–6.8 for psy-chological, -8.4–9.7 for social, and -3.7–4.8 for environmental) covered at least 95% of subjects’ traits (which ranged from -1.5 to 3.3, from -2.4 to 3.3, from -3.3 to 5.2, and from -1.6 to 2.4 for the physical, psychological, social, and environmental domains, respectively), indicating that the WHOQOL-BREF provided a satisfactory estimate for most subjects in this study.
Fig. 1 Item–person map on the logit scale for the physical domain of the World Health Organization Quality of Life-BREF (WHOQOL-Life-BREF). Logit values for the four thresholds (1, 2, 3, 4,
respectively) and the mean item difficulty (m) are given
Examination of DIF based on gender and age
Figure5shows that no DIF items were found across gen-ders, with the exception of two items in the social domain. Item S21 (Satisfaction with sex life) located at the bottom right of the DIF plot indicated that it was the most difficult item for males but the least difficult one for females. Item S22 (Satisfaction with friends’ support) was the least dif-ficult item for males but the most difdif-ficult one for females. The DIF plot, shown in Fig.6, indicated that three items (Y11, S21, and S22) displayed DIF across the two age
groups. Item ‘Y11 Bodily image and appearance’ was more difficult for the young–old group but less difficult for the old–old group. We further examined items ‘S21 Satisfac-tion with sex life’ and ‘S22 SatisfacSatisfac-tion with friends’ support’ as they both exhibited DIF across genders and across age groups. In Fig.7, there were three DIF items (i.e., ‘S21 Satisfaction with sex life’, ‘S22 Satisfaction with friends’ support’, and ‘S20 Personal relationships’) in the young–old group but no significant DIF item in the old–old group, despite two borderline DIF items of ‘S22 Satisfac-tion with friends’ support’ and ‘S27 Esteem and respect’. Fig. 2 Item–person map on the
logit scale for the psychological domain of the WHOQOL-BREF. Logit values for the four thresholds (1, 2, 3, 4,
respectively) and the mean item difficulty (m) are given
Fig. 3 Item–person map on the logit scale for the social domain of the WHOQOL-BREF. Logit values for the four thresholds (1, 2, 3, 4, respectively) and the mean item difficulty (m) are given
Discussion
The aim of the first part of our analysis was to confirm the model fit. Each item of the WHOQOL-BREF in its respective domain had properly ordered threshold param-eters and also showed a good fit to the unidimensionality specification. The person reliability indices were moder-ately high, with the exception of the social domain, which is similar to those found in Hwang et al. [11]. As IRT focuses more on item properties, such as difficulty, order-ing, and number of items in each domain, this type of analysis provides results for a more comprehensive item evaluation.
Unlike the traditional CTT analysis, which often impo-ses continuous level assumptions on ordered response, the model in IRT analysis treats the response categories as ordinal levels of data [44]. The person reliability index is the proportion of observed variance that is explained by the Rasch model, while Cronbach’s coefficient a is based on the covariance structure of the items within the scale. Both statistics are conceptually similar and are the percentage of observed response variance that is reproducible. Cron-bach’s coefficient a is constant for all scale scores, meaning that it assumes that measurement error is dis-tributed normally and equally for all score levels; in contrast, the measure of precision for the person reliability index is estimated separately for each score level or response pattern, thereby controlling for the characteristics (e.g., difficulty) of the items in the scale. Therefore, one of the main differences between Cronbach’s coefficient a and the person reliability index occurs when there are extreme scores. In such a case, Cronbach’s coefficient a increases,
but the person reliability decreases—and the latter is with a higher precision of measurement [16,45,46].
In addition, person reliability is computationally related to the person separation statistic, which is used to indicate the number of statistically different performance strata that can be identified in the sample (i.e., person separation of two indicates two strata can be identified) [16, 39]. The overall item performance also allowed us to evaluate the properties of each item in greater detail by examining their thresholds and corresponding person latent traits, which is not possible with the traditional CTT-based analysis [43,
44,47]. For example, a person with a latent trait of three logits in the social domain is more likely to choose response category 4 (satisfied) for item ‘S22 Personal relationships’, 3 (moderately satisfied) for item ‘S21 Sex-ual activity’, 4 (satisfied) for item ‘S22 friends’ Support’, and 4 (satisfied) for item ‘S27 Esteemed and respected’ (see Fig.3).
The item–person map explicitly reveals the relationship between person latent trait estimates and item difficulty parameters. The positive mean values (targeting indices) of person traits for each of the four domains indicate that the average item difficulty was relatively low for the elderly, with the social domain in particular showing substantial mis-targeting, with a targeting index of 1.02. Those items tended to be scored as ‘satisfied’, perhaps because very elderly people may respond that they are satisfied, even though they are unsure how to answer questions about sexual life and friendship. The addition of items to the social domain that are more appropriate for the elderly would be desirable to enhance the content relevance and scale performance. It may also be worthwhile to include Fig. 4 Item–person map on the
logit scale for the environmental domain of the WHOQOL-BREF. Logit values for the four thresholds (1, 2, 3, 4,
respectively) and the mean item difficulty (m) are given
items in order to obtain an even distribution of thresholds on the scale of the social domain, with the aim of increasing the precision of estimates in the full range, as there were no threshold values in the intervals of -2–1 and 4–6 for the social domain [43,47].
In the second part of our analysis, we applied DIF analysis to examine the invariance of item parameters across different subgroups and to explore influential issues within each specific construct domain. Item ‘S21 Satis-faction with sexual activity’ performed quite differently between males and females in young–old people, but there were only negligible differences in old–old people. This
phenomenon may be attributed to the physiological dif-ferences of the aging process between males and females [48–51]. For example, females tend to have a distinct cli-macteric in the aging process, with apparent physiological and psychic changes as well as an end of the reproductive capacity and onset of menopause [52], while the climac-teric in males is less clear, with a less-pronounced loss of libido and retained fertility [53]. However, sexual desire in both genders diminishes considerably as they age, as evi-denced by the result that item ‘S21 Satisfaction with sexual activity’ was the least difficult item for both males and females in the old–old group.
Another interesting item exhibiting DIF was item ‘S22 Satisfaction with friends’ support’, which was a relatively easy item for males but the most difficult one for females, as shown in Fig.5. Further DIF analysis across genders, stratified by different age groups as shown in Fig.7, sup-ported the concept that ‘S22 Satisfaction with friends’ support’ was still a relatively difficult one for female. However, for males, ‘S22 Satisfaction with friends’ sup-port’ was a relatively easy item for young–old males but was the most difficult one for old–old males. The finding that females had a relatively low satisfaction with friends’ support may have been due to the fact that our study sub-jects lived in a rural area and were more socially isolated and socially conservative. Another possible explanation is
that the social circles of married women become smaller as they age and are often confined primarily to their husband and his relatives. Moreover, women often have a longer life expectancy than men and become widowed (e.g., in the study, the spouses of 16.8% of males and 39.5% of females had died). Previous studies have also found that elderly females residing in rural areas had a poor quality of social life, likely due to low levels of social contact support, high levels of isolation, being widowed, or living alone [54].
Another finding is that the item-difficulty locations of two items ‘S21 Satisfaction with sex life’ and ‘S22 Satis-faction with friends’ support’ for males were calibrated in the two opposite extremes separately in the young–old and old–old groups. A noticeable movement of these two items Fig. 6 Differential item functioning plot by age group and the WHOQOL-BREF domain
in their relative locations was observed between the young–old and the old–old groups, as shown in Fig.7. The violation of item parameter invariance for items S21 and S22 is potentially a consequence of getting older. Older people need to adjust to the aging process, such as the natural decrease in sexual activity that occurs with increased age and the greater social isolation as friends move away, become sick, or die. The relative importance of these two items in our study in the older male subjects changed, whereas the order of ‘difficulty’ for some other items was quite stable. For example, in both age groups, self-care tasks were easier to perform than household activities, and walking a short distance was easier to per-form than stair climbing; however, items S21 and S22 did not maintain the expected order among latent traits with increasing age. The impact of DIF on older male subjects can be further investigated by examining its influence on domain scores of the WHOQOL-BREF, especially by comparing the magnitudes of scale-level differences in the social domain between the young–old and old–old groups. There are some limitations in the study worth noting. First, the response rate of each of the 26 WHOQOL-BREF items was higher than 97.5%, with the exception of item ‘21 Sexual life’ (83.5%). However, this limitation can be adequately mitigated by the Winstep program for the Rasch model as it can handle missing item response data. Second, the study cohort may not be representative of the general elderly population in Taiwan as compared with the general population in Taiwan, the study subjects had a lower education level and there was a higher percentage of males. The study subjects were also from a conservative rural
area, and certain factors, such as the role of women in the family and job status before retirement, would be affected differently than were the subjects from urban areas. The reason why we chose this specific population group is because HRQL in rural-dwelling elderly people is less well understood and this group is in greater need of health promotion, including HRQL. Lastly, very sick older people were not available for this study; therefore, our findings may only reflect the QOL among older people who are relatively healthy.
Conclusions
Health-related quality of life has become an increasingly important measure in the clinical evaluation of the treat-ment and cost-effectiveness of health policies [12,55]. As the WHOQOL-BREF expands its use as a generic HRQL instrument, it is of great importance to evaluate its psy-chometric properties and suitability in specific populations. Using Rasch analysis, we conducted a systematic evalua-tion of the WHOQOL-BREF in rural community-dwelling older people in Taiwan. While the results confirmed the suitability of the WHOQOL-BREF for this elderly popu-lation, they also identified areas for improvements, especially for items in the social domain. Differential item functioning analysis detected certain items, such as body image, sexual activity, and friends’ support, as performing differently with increasing age. Researchers and clinicians need to be aware that the results for certain items, such as friends’ support and sexual life, may depend on aging and Fig. 7 Differential item functioning plot by age-stratified gender group in the social domain
gender. If one ignores the age effect in the evaluation of a health promotion program, for example, the benefit of the program among the older elderly may be under-estimated due to their lower satisfaction with certain age- and gender-sensitive items, such as friends’ support among older male individuals. Therefore, it is essential to conduct the anal-ysis with and without these age- and gender-sensitive items to confirm consistency or with a stratified method by age and gender to avoid problems caused by DIF. We also suggest that both domain and item scores be used if there are DIF differences in order to obtain a more comprehen-sive understanding of HRQL in elderly populations. A particular scoring strategy is also suggested if we want to
compare the HRQL by using this measure among old–old and old–young elderly Taiwanese in rural areas.
Acknowledgments This study was supported by the National Health Research Institutes (NHRI-EX98-9805PI and NHRI-EX95-9204PP), China Medical University (CMU96-225), and the National Science Council (NSC93-2320-B-039-013 and NSC95-2314-B-038-008) of Taiwan, Republic of China. We are most grateful to all of the subjects who participated in this study.
Appendix
See Table3.
Table 3 The WHOQOL-BREF
Code Item statement Response choices
– How would you rate your quality of life? 1. Very poor 2. Poor 3. Neither poor nor good 4. Good 5. Very good – How satisfied are you with your health? 1. Very dissatisfied 2. Dissatisfied 3. Neither satisfied nor dissatisfied
4. Satisfied 5. Very satisfied P3 To what extent do you feel that physical pain prevents you from
doing what you need to do?
1. Not at all 2. A little 3. A moderate amount 4. Very much 5. An extreme amount
P4 How much do you need any medical treatment to function in your daily life?
As above Y5 How much do you enjoy life? As above Y6 To what extent do you feel your life to be meaningful? As above
Y7 How well are you able to concentrate? 1. Not at all 2. A little 3. A moderate amount 4. Very much 5. Extremely
E8 How safe do you feel in your daily life? As above E9 How healthy is your physical environment? As above
P10 Do you have enough energy for everyday life? 1. Not at all 2. A little 3. Moderately 4. Mostly 5. Completely Y11 Are you able to accept your bodily appearance? As above
E12 Have you enough money to meet your needs? As above E14 To what extent do you have the opportunity for leisure activities? As above
P15 How well are you able to get around? 1. Very poor 2. Poor 3. Neither poor nor good 4. Good 5. Very good P16 How satisfied are you with your sleep? 1. Very dissatisfied 2. Dissatisfied 3. Neither satisfied nor dissatisfied
4. Satisfied 5. Very satisfied P17 How satisfied are you with your ability to perform your daily
living activities?
As above P18 How satisfied are you with your capacity for work? As above Y19 How satisfied are you with yourself? As above S20 How satisfied are you with your personal relationships? As above S21 How satisfied are you with your sex life? As above S22 How satisfied are you with the support you get from your
friends?
As above E23 How satisfied are you with the conditions of your living place? As above E24 How satisfied are you with your access to health services? As above E25 How satisfied are you with your transport? As above Y26 How often do you have negative feelings such as blue mood,
despair, anxiety, depression?
1. Never 2. Seldem 3. Quite often 4. Very often 5. Always S27 Do you feel respected by others? 1. Not at all 2. A little 3. A moderate amount 4. Very much 5. An
extreme amount
References
1. Cella, D., & Chang, C. H. (2000). A discussion of item response theory and its applications in health status assessment. Medical Care, 38[Suppl 9], II66–72.
2. Hays, R. D., Morales, L. S., & Reise, S. P. (2000). Item response theory and health outcomes measurement in the 21st century. Medical Care, 38[Suppl 9], II28–42.
3. Field trial WHOQOL-100. (1995). The 100 questions with response scales. Geneva: World Health Organization.
4. Field trial WHOQOL-100. (1995). Scoring the WHOQOL. Geneva: World Health Organization.
5. Ozakbas, S., Akdede, B. B., Kosehasanogullari, G., Aksan, O., & Idiman, E. (2007). Difference between generic and multiple sclerosis-specific quality of life instruments regarding the assessment of treatment efficacy. Journal of the Neurological Sciences, 256(1–2), 30–34.
6. Fang, C. T., Hsiung, P. C., Yu, C. F., Chen, M. Y., & Wang, J. D. (2002). Validation of the World Health Organization quality of life instrument in patients with HIV infection. Quality of Life Research, 11(8), 753–762.
7. Wang, W. C., Yao, G., Tsai, Y. J., Wang, J. D., & Hsieh, C. L. (2006). Validating, improving reliability, and estimating corre-lation of the four subscales in the WHOQOL-BREF using multidimensional Rasch analysis. Quality of Life Research, 15(4), 607–620.
8. Naumann, V. J., & Byrne, G. J. A. (2004). WHOQOL-BREF as a measure of quality of life in older patients with depression. International Psychogeriatrics, 16(2), 159–173.
9. Hirayama, M. S., Gobbi, S., Gobbi, L. T., & Stella, F. (2008). Quality of life (QoL) in relation to disease severity in Brazilian Parkinson’s patients as measured using the WHOQOL-BREF. Archives of Gerontology and Geriatrics, 46(2), 147–160. 10. Jamison, R. N., Fanciullo, G. J., Mchugo, G. J., & Baird, J. C.
(2007). Validation of the short-form interactive computerized quality of life scale (ICQOL-SF). Pain Medicine, 8(3), 243–250. 11. Hwang, H. F., Liang, W. M., Chiu, Y. N., & Lin, M. R. (2003). Suitability of the WHOQOL-BREF for community-dwelling older people in Taiwan. Age and Ageing, 32(6), 593–600. 12. Pickard, A. S., Dalal, M. R., & Bushnell, D. M. (2006). A
comparison of depressive symptoms in stroke and primary care: Applying Rasch models to evaluate the center for epidemiologic studies-depression scale. Value in Health, 9(1), 59–64.
13. Conrad, K. J., & Smith, E. V., Jr. (2004). International conference on objective measurement: Applications of Rasch analysis in health care. Medical Care, 42[Suppl 1], I1–6.
14. Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Hillsdale, NJ: Lawrence Erlbaum Associates. 15. Hambelton, R. K., Swaminathan, H., & Rogers, H. J. (1991).
Fundamentals of item response theory. London: Sage.
16. Bond, T. G., & Fox, C. M. (2001). Applying the Rasch model: Fundamental measurement in the human sciences. Hillsdale, NJ: Lawrence Erlbaum Associates.
17. Noyes, K., Dick, A. W., & Holloway, R. G. (2006). Pramipexole versus levodopa in patients with early Parkinson’s disease: Effect on generic and disease-specific quality of life. Value in Health, 9(1), 28–38.
18. Gunther, O. H., Roick, C., Angermeyer, M. C., & Konig, H. H. (2008). The responsiveness of EQ-5D utility scores in patients with depression: A comparison with instruments measuring quality of life, psychopathology and social functioning. Journal of Affective Disorders, 105(1–3), 81–91.
19. Lawton, M. P., Windley, P. G., & Byerts, T. O. (1982). Aging and the environment: Theoretical approaches. New York: Springer.
20. Pearlman, R. A., & Uhlmann, R. F. (1988). Quality of life in chronic diseases: Perceptions of elderly patients. Journal of Gerontology, 43(2), M25–30.
21. Stewart, A. L., Sherbourne, C. D., & Brod, M. (1996). Measuring health-related quality of older and demented populations. Phil-adelphia: Lippincott-Raven Publishers.
22. Bindman, A. B., Keane, D., & Lurie, N. (1990). Measuring health changes among severely ill patients: The floor phenomenon. Medical Care, 28(12), 1142–1151.
23. Watsford, M. L., Murphy, A. J., & Pine, M. J. (2007). The effects of ageing on respiratory muscle function and performance in older adults. Journal of Science and Medicine in Sport, 10(1), 36– 44.
24. WHOQOL-BREF. (1996). Introduction, adminstration, scoring and generic version of the assessment–field trial version. Geneva: World Health Organization.
25. Yao, K. P. (2002). Development and applications of the WHO-QOL-Taiwan version. Formosan Journal of Medicine, 6(2), 193– 200.
26. Chan, S. W., Chiu, H. F., Chien, W. T., Thompson, D. R., & Lam, L. (2006). Quality of life in Chinese elderly people with depression. International Journal of Geriatric Psychiatry, 21(4), 312–318. 27. Yang, S. C., Kuo, P. W., Wang, J. D., Lin, M. I., & Su, S. (2006).
Development and psychometric properties of the dialysis module of the WHOQOL-BREF Taiwan version. Journal of the Formo-san Medical Association, 105(4), 299–309.
28. Yao, G., Chung, C. W., Yu, C. F., & Wang, J. D. (2002). Development and verification of validity and reliability of the WHOQOL-BREF Taiwan version. Journal of the Formosan Medical Association, 101(5), 342–351.
29. Van Der Linden, W. J., & Hambleton, R. K. (1997). Handbook of modern item response theory. New York: Springer.
30. Linacre, J. M. (2006). A user’s guide to WINSTEPS: Rasch-Model Computer Programs. Chicago:www.winsteps.com. 31. Wright, B. D., & Maters, O. N. (1982). Rating scale analysis:
Rasch measurement. Chicago: MESA Press.
32. Avery, L. M., Russell, D. J., Raina, P. S., Walter, S. D., & Ro-senbaum, P. L. (2003). Rasch analysis of the gross motor function measure: Validating the assumptions of the Rasch model to create an interval-level measure. Archives of Physical Medicine and Rehabilitation, 84(5), 697–705.
33. Jo¨reskog, K., & So¨rbom, D. (2005). LISREL 8.72. Chicago: Scientific Software.
34. Metz, S. M., Wyrwich, K. W., Babu, A. N., Kroenke, K., Tierney, W. M., & Wolinsky, F. D. (2006). A comparison of traditional and Rasch cut points for assessing clinically important change in health-related quality of life among patients with asthma. Quality of Life Research, 15(10), 1639–1649.
35. Hart, D. L., Mioduski, J. E., & Stratford, P. W. (2005). Simulated computerized adaptive tests for measuring functional status were efficient with good discriminant validity in patients with hip, knee, or foot/ankle impairments. Journal of Clinical Epidemiol-ogy, 58(6), 629–638.
36. Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis_conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1–55. 37. Fitzpatrick, R., Norquist, J. M., Dawson, J., & Jenkinson, C.
(2003). Rasch scoring of outcomes of total hip replacement. Journal of Clinical Epidemiology, 56(1), 68–74.
38. Wright, B. D., & Linacre, J. M. (1996). Reasonable mean-square fit values. In J. M. Linacre (Ed.), Rasch measurement transac-tions (p. 370). Chicago: MESA.
39. Duncan, P. W., Bode, R. K., Min Lai, S., & Perera, S. (2003). Rasch analysis of a new stroke-specific outcome scale: The stroke
impact scale. Archives of Physical Medicine and Rehabilitation, 84(7), 950–963.
40. Gallagher, P., Horgan, O., Franchignoni, F., Giordano, A., & Maclachlan, M. (2007). Body image in people with lower-limb amputation: A Rasch analysis of the amputee body image scale. American Journal of Physical Medicine and Rehabilitation, 86(3), 205–215.
41. Urbach, D. R., Tomlinson, G. A., Harnish, J. L., Martino, R., & Diamant, N. E. (2005). A measure of disease-specific health-related quality of life for achalasia. The American Journal of Gastroenterology, 100(8), 1668–1676.
42. Wolfe, F., Michaud, K., & Pincus, T. (2004). Development and validation of the health assessment questionnaire II: A revised version of the health assessment questionnaire. Arthritis and Rheumatism, 50(10), 3296–3305.
43. Lai, J. S., Cella, D., Chang, C. H., Bode, R. K., & Heinemann, A. W. (2003). Item banking to improve, shorten and computerize self-reported fatigue: An illustration of steps to create a core item bank from the FACIT-Fatigue Scale. Quality of Life Research, 12(5), 485–501.
44. Uttaro, T., & Lehman, A. (1999). Graded response modeling of the quality of life interview. Evaluation and Program Planning, 22(1), 41–52.
45. Vidotto, G., Bertolotti, G., Carone, M., Arpinelli, F., Bellia, V., Jones, P. W., et al. (2006). A new questionnaire specifically designed for patients affected by chronic obstructive pulmonary disease; the Italian health status questionnaire. Respiratory Medicine, 100(5), 862–870.
46. Vidotto, G., Carone, M., Jones, P. W., Salini, S., & Bertolotti, G. (2007). Maugeri respiratory failure questionnaire reduced form:
A method for improving the questionnaire using the Rasch model. Disability and Rehabilitation, 29(13), 991–998. 47. Garratt, A. M. (2003). Rasch analysis of the Roland disability
questionnaire. Spine, 28(1), 79–84.
48. Onder, G., Penninx, B. W., Guralnik, J. M., Jones, H., Fried, L. P., Pahor, M., et al. (2003). Sexual satisfaction and risk of dis-ability in older women. The Journal of Clinical Psychiatry, 64(10), 1177–1182.
49. Wespes, E., & Schulman, C. C. (2002). Male andropause: Myth, reality, and treatment. International Journal of Impotence Research, 14[Suppl 1], S93–98.
50. Kaiser, F. E. (1992). Sexual function and the older cancer patient. Oncology (Williston Park), 6(2 Suppl), 112–118.
51. Yang, H., Toy, E. C., & Baker, B. (2000). Sexual dysfunction in the elderly patient. Primary Care Update for Ob/Gyns, 7(6), 269– 274.
52. Kingsberg, S. A. (1998). Postmenopausal sexual functioning: A case study. International Journal of Fertility and Women’s Medicine, 43(2), 122–128.
53. Peate, I. (2003). The male menopause: Possible causes, symp-toms and treatment. The British Journal of Nursing, 12(2), 80–84. 54. Savikko, N., Routasalo, P., Tilvis, R. S., Strandberg, T. E., & Pitkala, K. H. (2005). Predictors and subjective causes of lone-liness in an aged population. Archives of Gerontology and Geriatrics, 41(3), 223–233.
55. Tengs, T. O. (2004). Cost-effectiveness versus cost-utility anal-ysis of interventions for cancer: Does adjusting for health-related quality of life really matter? Value in Health, 7(1), 70–78.
