Development of a 12-item short version of the HIV stigma scale

Background Valid and reliable instruments for the measurement of enacted, anticipated and internalised stigma in people living with HIV are crucial for mapping trends in the prevalence of HIV-related stigma and tracking the effectiveness of stigma-reducing interventions. Although longer instruments exist, e.g., the commonly used 40-item HIV Stigma Scale by Berger et al., a shorter instrument would be preferable to facilitate the inclusion of HIV stigma in more and broader surveys. Therefore, the aim of this work was to develop a substantially shorter, but still valid, version of the HIV Stigma Scale. Methods Data from a psychometric evaluation of the Swedish 40-item HIV Stigma Scale were reanalysed to create a short version with 12 items (three from each of the four stigma subscales: personalised stigma, disclosure concerns, concerns with public attitudes and negative self-image). The short version of the HIV stigma scale was then psychometrically tested using data from a national survey investigating stigma and quality of life among people living with HIV in Sweden (n = 880, mean age 47.9 years, 26% female). Results The hypothesized factor structure of the proposed short version was replicated in exploratory factor analysis without cross loadings and confirmatory factor analysis supported construct validity with high standardised effects (>0.7) of items on the intended scales. The χ2 test was statistically significant (χ2 = 154.2, df = 48, p < 0.001), but alternate fit measures indicated acceptable fit (comparative fit index: 0.963, Tucker-Lewis index: 0.950 and root mean square error of approximation: 0.071). Corrected item-total correlation coefficients were >0.4 for all items, with a variation indicating that the broadness of the concept of stigma had been captured. All but two aspects of HIV-related stigma that the instrument is intended to cover were captured by the selected items in the short version. The aspects that did not lose any items were judged to have acceptable psychometric properties. The short version of the instrument showed higher floor and ceiling effects than the full-length scale, indicating a loss of sensitivity in the short version. Cronbach’s α for the subscales were all >0.7. Conclusions Although being less sensitive in measurement, the proposed 12-item short version of the HIV Stigma Scale has comparable psychometric properties to the full-length scale and may be used when a shorter instrument is needed.


Introduction
HIV-related stigma is prevalent in many parts of the world and affects the quality of life of people living with HIV [1][2][3][4]. HIV-related stigma is also a common barrier to HIV testing and treatment adherence [5][6][7][8][9][10]. Valid and reliable instruments for the measurement of enacted, anticipated and internalised stigma in people living with HIV are crucial for mapping trends in the prevalence of HIV-related stigma [10]. There are several instruments designed to measure HIV stigma, where Berger et al.'s [11] 40-item HIV Stigma Scale is the most commonly used and is one of only a few instruments that cover all stigma mechanisms affecting people with HIV [10]. We recently adapted and validated this scale for the Swedish context and, after removing one item, a 39-item scale showed satisfactory construct validity and reliability [12]. The over-determination of the full-length scales, with high Cronbach's α (0.883-0.958), indicated that the number of items could be reduced. Furthermore, the original 40-item scale may take up to 25 min to complete [13]. Shortened versions, which respectively cover 25 and 32 items of the HIV Stigma Scale, have been published previously [13,14]. However, to facilitate the inclusion of HIV stigma in more extensive surveys, a shorter instrument would be preferable. Although short forms exist for children and young adults [15][16][17], beyond the abovementioned examples, no other shorter versions of the HIV Stigma Scale have been published for adults living with HIV. The aim of this work was therefore to develop a substantially shorter version of the HIV Stigma Scale with psychometric properties retained from the full-length scale.

Methods
The short version of the HIV Stigma Scale was developed in two phases: 1) Our data from the validation of the 40item HIV Stigma Scale in Sweden (n = 132, 55 female, 77 male, age 23-74; mean 48.3, SD 11.0) [12] were reanalysed in order to select items for a short version. 2) To ensure construct validity and reliability of the shorter version, psychometric analysis was also performed on data from an additional sample who had responded to our proposed short version of the HIV Stigma Scale.

Phase 1. Item reduction
The HIV Stigma Scale consists of four subscales intended to measure personalised stigma, disclosure concerns, concerns with public attitudes and negative self-image [11]. Berger et al. [11] described that each of these subscales contains between two and three main aspects, as summarised  (6);518-29 [11] in Table 1. The intention when developing a short version of the HIV Stigma Scale was to select items in the instrument that showed good psychometric properties, but also to maintain as many as possible of the aspects of HIV stigma that the original instrument was intended to cover. The item reduction process is presented schematically in Fig. 1 and in more detail below.
Step 1. Removing items with underfit All items in the full-length scale were examined with item response theory methods to find items with underfit [18]. Partial credit models were calculated for each subscale separately using the package eRm [19] in R statistics [20] and item fit statistics were assessed. Items with outfit or infit mean square values exceeding 1.2 were considered to have underfit [18] and were not considered for the short version of the HIV stigma scale.
Step 2. Removing cross-loading items Remaining items were evaluated regarding their loading in our previously published exploratory factor analysis performed on data from 132 persons living with HIV in Sweden [12]. Cross loading items were not considered for the short version of the HIV stigma scale.
Step 3. Keeping as many aspects as possible A group of professionals working in academia and HIV care and with expertise in HIV and psychometrics discussed which of the remaining items best represented the different aspects of HIV stigma that Berger et al. [11] intended the instrument to cover. They agreed on three selected items from each of the four subscales to be included in the Phase 2 assessment of a tentative 12-item short version of the HIV Stigma Scale. The same response format from the original scale was used, i.e. a 4-point Likert scale, ranging from strongly disagree (1) to strongly agree (4). Responses were summed to calculate subscale scores with a possible range of 3 to 12; higher scores reflect a higher level of perceived HIV-related stigma.

Phase 2. Psychometric evaluation of the short version of the HIV stigma scale
The proposed short version of the HIV Stigma Scale was distributed as part of a longer self-administered anonymous questionnaire to a sample of individuals participating in the nationwide study 'Living with HIV in Sweden' [21]. This nationwide study investigated the quality of life of people living with HIV in Sweden and was performed December 2013 through August 2014.

Participants
The inclusion criteria were as follows: 1) >18 years of age and 2) having been diagnosed with HIV >6 months.

Construct validity
The sample of 880 completed questionnaires was randomly divided into two groups of equal size, where the first part was analysed in an exploratory factor analysis and the second part in confirmatory factor analysis, to ensure factor stability. The exploratory factor analysis was performed in SPSS 23 with alpha factoring, oblimin rotation. The confirmatory factor analysis model that represented the short version of the HIV Stigma Scale was then set up and analysed with maximum likelihood  Factor loadings <-0.32 or >0.32 in bold using the lavaan package [23] in R statistics [20]. Goodness of fit was evaluated using χ 2 testing and was expected to be non-significant if the data had a good fit to the model; root mean square error of approximation (RMESA), where a score below 0.05 indicates good fit; and comparative fit index (CFI) and Tucker-Lewis index (TLI), with a desired value of >0.90 for both indexes [24]. Corrected item-total correlation coefficients were calculated for each item, which were expected to exceed 0.4 and also have a variation in range to ensure that the broadness of the measured concept had been captured by the short version of the scale [25]. Floor and ceiling effects were calculated and compared to the Swedish 39item version of the HIV Stigma Scale (where less than 15% of participants had the lowest or highest possible score [12], which is considered acceptable [26]).

Reliability
Cronbach's α was assessed for the subscales to ensure internal consistency and was considered acceptable if >0.7 [27].
Twelve items showed underfit (Table 2) and were removed.
Step 2. Removing cross-loading items. One of the remaining items, item 25, cross-loaded in the exploratory factor analysis ( Table 2) and was removed.
Step 3. Keeping as many aspects as possible.
From the remaining items, three items were chosen from each subscale. We chose a selection of items that covered as many aspects of the concepts as possible. If more than one item covered an aspect, the item with highest loading was chosen. The selected items and the aspects they cover are shown in Table 1. For personalised stigma and disclosure concerns, one aspect was lost in each subscale due to cross-loading items or items with underfit (regrets that a person can have over disclosing one's HIV status and worry that someone else will disclose one's HIV status).

Construct validity
In the exploratory factor analysis (alpha factoring, oblimin rotation) the factor structure suggested for the short version of the HIV stigma scale was replicated without cross loadings (Table 3). Eigenvalues for the four factors  Table  4 on the item level and subscale level. Corrected itemtotal correlation coefficients exceeded 0.4 for all items and had a variation in the range 0.62-0.84 (Table 4), indicating that the broadness of the intended stigma concepts had been captured. Floor or ceiling effects exceeded 15% for Personalized stigma (28% of participants had lowest possible score), Disclosure concerns (22% of participants had highest possible score and Negative self-image (24% of participants had lowest possible score (Table 4).

Reliability
Cronbach's α for the final combination of items for the subscales were all above 0.7 and considered acceptable ( Table 4).

Discussion
This report describes the development of a 12-item short version of the HIV Stigma Scale, that mainly preserves the broad concepts and internal consistency of the original subscales [11]. Since cross-loading items were excluded from the short version and no cross-loadings appeared in the exploratory factor analysis, we believe that this short version may have better psychometric properties than the original full length HIV Stigma Scale, which has exhibited overlap of items between several subscales [14]. The subscales in the short versions are highly intercorrelated, which reflects the psychometric properties of the original full length HIV stigma scale. However, a significant χ2 test of the confirmatory factor analysis indicated a misfit between the short version of the instrument and the data. Although it is known that even minor differences can generate a statistically significant χ2 value [28], we suggest further psychometric testing of the short version of the HIV stigma scale to examine whether signs of overlap of the variance between subscales will occur. Nevertheless, the other measures of model fit used, including standardised maximum likelihood estimates above 0.7 in the expected direction together with alternate fit measures within an acceptable range, supported the construct validity of the short scale in a Swedish context; whether this holds true for different populations remains to be investigated. The sample used for testing of the proposed short version had a high proportion of men, which reflects the gender distribution of persons living with HIV in Sweden. Specific gender related properties of the instrument can therefore have been missed and should be observed in future studies.
The exploratory factor analysis that formed the basis for item selection was performed on a relatively small sample (n = 132). However, the overdetermined factors, together with a wide range of communalities, supported reliability of the resulting solution [29]. When choosing between using the full length scale or the developed short form, the higher floor and ceiling effects shown for the short version should be taken into account; as expected, the short version is less sensitive in detecting different levels of perceived stigma. Although less sensitive than the full-length scale, we consider the proposed 12-item short version of the HIV Stigma Scale to have essentially the same psychometric properties as the full-length scale and propose that it may be used when a shorter instrument is desirable. The short version could, for example, be used if there is a wish to include a brief stigma component in longer surveys investigating the life situation of people living with HIV, in clinical contexts as a brief screening measure for signs of stigma-related problems or to serve as a basis for discussions with individual clients.

Conclusions
Although being less sensitive in measurement, the proposed 12-item short version of the HIV Stigma Scale has comparable psychometric properties to the full-length scale and may be used when a shorter instrument is needed.
Abbreviations CFI: Comparative fit index; RMSEA: Root mean square error of approximation; TLI: Tucker-Lewis index