Development and validation of the self‐stigma scale in people with diabetes

Abstract Aim To develop and initially validate the Diabetes Self‐Stigma Scale for assessing self‐stigma in people with diabetes. Design Scale development and evaluation. Methods Participants were 399 patients with diabetes. In phase 1, initial items were generated based on the concept analysis of diabetes self‐stigma. Moreover, content validity was established by diabetes experts. Phase 2 evaluated structural validity through item analysis, exploratory factor analysis and confirmatory factor analysis. Reliability was evaluated by examining stability and internal consistency. Results The findings revealed that the self‐stigma scale for patients with diabetes is a valid and reliable instrument. The Diabetes Self‐Stigma Scale was confirmed with 16 items. It consists of four domains: comparative inability, social withdrawal, self‐devaluation and apprehensive feeling. The scale developed in this study can measure self‐stigma in diabetes patients and can be used as an intervention to reduce self‐stigma.

In Japan, the validity and reliability of related tools to measure selfstigma for diabetes have been established (Kato et al., 2014). However, since the Japanese version of the Self-Stigma Scale was initially developed for minorities in Hong Kong (Mak & Cheung, 2010), it does not fully reflect the characteristics of people with diabetes who experience self-stigma. In addition, self-stigma is created by the internalization of social stigma and social stigma is influenced by the culture to which they belong (Rüsch et al., 2005;Seo & Song, 2019). Therefore, there is a limit to its usefulness in measuring self-stigma in Korean people with diabetes. Despite the need to measure and manage self-stigma of diabetes patients, there are no tools to measure self-stigma for diabetes patient in Korea. Therefore, it is necessary to develop and validate the Diabetes Self-Stigma Scale (DSSS) for people with diabetes in Korea that accurately reflects the properties of self-stigma.

| BACKG ROU N D
It has been found that stigma reduces self-esteem (Park et al., 2019;Rüsch et al., 2005). However, not all social stereotypes or prejudices reduce an individual's self-esteem. To do this, the stereotypes or prejudices of society must be internalized. Such internalization is called self-stigma. Self-stigma means that individuals accept socially shared stereotypes and prejudices and devalue themselves, believing that they are merely socially tolerated (Vogel et al., 2006).
Self-stigma occurs in three stages: awareness, agreement and application (Corrigan, 2004;Corrigan & Watson, 2002). This can be explained by an individual identifying one's own traits with those of the stigmatized group and applying stereotypes and prejudices related to them (Eisenberg et al., 2009). Individuals who experience social stigma may be aware of stereotypes that others apply to them, but they do not internalize all of them. In other words, self-stigma is not simply the awareness of prejudice or stereotype but the process of internalization.
A previous study (Link et al., 1989) emphasized the importance of self-stigma by suggesting a "modified labelling theory." This means that people with self-stigma experience negative consequences, such as damage to their self-esteem, that make them vulnerable. Therefore, individuals with self-stigma have a diminished self-esteem or self-worth (Vogel et al., 2006). They perceive themselves as inferior, inadequate or weak (Nadler & Jeffrey, 1986).
In this context, people with diabetes often develop self-stigma through the treatment process rather than because of the symptoms of the disease. Self-stigma is more important than social stigma for diabetes because the general public does not think that having diabetes is stigmatized and people with diabetes do not stigmatize other diabetes patients. In contrast, diabetes patients develop negative attitudes towards themselves because they feel judged and monitored by others (Schabert et al., 2013). In addition, patients with diabetes can hide their disease from others because their disease is not visible (Song & Ah, 2016).
Most studies on stigma of diabetes patients have been conducted to define the concept simply as stigma or social stigma without considering its level (Abdoli et al., 2018;Browne et al., 2016;Schabert et al., 2013). Nishio and Chujo (2017) conducted a study on self-stigma in diabetes patients that was limited to patients with type 1 diabetes and may not be generalizable to patients with type 2 diabetes. Additionally, the tool that was designed to measure diabetes self-stigma is the only DSSS used in Japan (Kato et al., 2014). This tool was designed for minorities with mental illness and homosexuals in Hong Kong at the time of the tool's development. It is therefore limited in that it does not sufficiently reflect the characteristics of diabetes self-stigma. In addition, since self-stigma affects public perception about the disease (Kato et al., 2014), it is necessary to have a self-stigma tool that applies to Korean culture.

| Design
The design of this study involved a development phase and an evaluation phase of the scale. The participants were patients with diabetes recruited from the outpatient endocrine departments at two university hospitals in D city and from two health centres located in cities C and S. A sample of 399 people was included in the study. The study was explained to them; they agreed to participate and provided written consent. The sample was randomly divided into set A for exploratory factor analysis (EFA) and set B for confirmatory factor analysis (CFA).
The inclusion criteria for participants were as follows: (a) received a diabetes diagnosis from a physician, (b) participated in functional self-care and possible social activities, (c) were not diagnosed with psychiatric illnesses and were able to communicate, (d) had the cognitive ability to express feelings and (e) were voluntarily participating in the study.
This study was conducted in two phases ( Figure 1) (Netemeyer et al., 2003). In the development phase, preliminary questions were developed based on the definitions and attributes of the concepts presented through concept analysis (Seo & Song, 2019). The preliminary tool underwent content validation with the expert group and diabetes patients group. In the evaluation phase, structural validity was evaluated through EFA and CFA. In addition, the final instrument of diabetes self-stigma was confirmed by evaluating reliability using internal consistency.

| Development phase
The initial items of the DSSS were developed based on the results of concept analysis of diabetes self-stigma to improve the availability of the contents. The concept analysis for self-stigma was conducted through three phases (theoretical phase, fieldwork phase and final analysis phase) of a hybrid method for Korean individuals with diabetes, reflecting the self-stigma of diabetes based on Korean culture. The self-stigma concept included three categories with nine attributes: affective (negative feelings and feeling sorry for others who have concerns about me), cognitive (low selfesteem and self-efficacy, perceived weakness, low expectations for the future, worry for children and disease burden) and behavioural (social withdrawal and avoiding disease disclosure) factors.
Detailed methods and results can be found in our main published paper (Seo & Song, 2019). Based on this, 40 questions were derived using existing self-stigma tools and in-depth interview data of patients with diabetes.

| Evaluation phase
In this phase, the content validity index (CVI) was calculated. The content was validated with experts and people with diabetes. Ten experts participated including endocrinologists, nurses specializing in diabetes and nursing professors. Ten diabetes patients also participated in the content verification. They were asked to respond on a 4-point Likert scale as to whether the question was appropriate for measuring diabetes self-stigma. In addition, they were required to pick out words that were used inappropriately or sentences that were difficult to understand. CVI measures the number of experts who gave a 3 or 4 rating for each item after measuring "1 = not relevant," "2 = somewhat relevant," "3 = relevant" and "4 = very relevant." That is, the proportion in agreement about the relevance was measured and an item with a score of 0.8 or higher was adopted for the scale. Items with scores below 0.5 were considered less valid and were excluded from the list by obtaining a non-consensus ratio of relevance and items with values of 0.5-0.8 were revised based on the opinions of experts (Polit et al., 2007).
Structural validity was confirmed through item analysis, EFA and CFA. The EFA determines the number of dimensions by grouping related questions. CFA is a method of verifying support by modelling theoretically proven dimensions of self-stigma for diabetes patients and items tied to EFA (Kang, 2006). In addition, the final questions were confirmed after evaluating their reliability using internal consistency.

| Participants
This study was conducted with adults over 20 years of age with type 2 diabetes. Convenience sampling was conducted for those who understood the purpose of the study and voluntarily agreed to participate in the study among patients with diabetes who visited the University Hospital Endocrinology Department in D City and the public health centre in C Province. Data were collected from a total of 430 subjects; this number was reached based on the evidence that 5-10 times (or more) the number of items was appropriate for a stable test of the reliability and validity of the measurement tool including factor analysis (Tabachnick et al., 2001). Of these, data were analysed using the survey results of 399 people, excluding the data from 31 people with insufficient responses. Of the 399 responses, data from 183 people (Dataset A) were used for exploratory factor analysis and the data from 216 people (Dataset B) were used for confirmatory factor analysis.
The administered questionnaire included sociodemographic information (i.e. gender, age, educational level, marital status, having religion, having a job, cohabiting status, economic satisfaction, social activities, perceived health status) and characteristics of diabetes (i.e. duration of illness, type of medical care, how diabetes is managed, number and type of complications, receiving diabetes education).

| Statistical analysis
The developed tool was analysed using IBM SPSS (version 22.0) and AMOS (version 22.0). Item-total correlation coefficients were calculated to determine whether each item reflects the concept of self-stigma measurement. Items with low contributions in the domain with the item-total score correlation coefficient of less than 0.20 were deleted (Field, 2005).
The validity of this tool was analysed using EFA and CFA. EFA examined the suitability of the factor analysis using the Kaiser-Meyer-Olkin (KMO) index and Bartlett's spherical test. EFA was then conducted to determine the number of dimensions by grouping related items. To conduct EFA, it was necessary to determine the factor extraction and rotation methods. Factor extraction includes principal component analysis, principal axis factor method and maximum likelihood method. Among them, the principal factor method assumes a factor analysis model and extracts factors so that the covariance of the measured variables can be explained as much as possible. Rotation was then performed to simplify the structure of the factors. Both orthogonal and oblique rotation methods were used. Oblique rotation is more common in the field of social science because it places a greater significance on the interpretation of the factor structure in terms of the disconnection of factors (Woo, 2012). In this study, factors were extracted using the principal factor method. Thereafter, the oblimin method, which is an orthogonal pre-transaction method, was used. Standards for loading values vary between 0.3, 0.4 and 0.5, but in this study, items with communality and loading value size of 0.4 or less were removed (Song, 2011).
In CFA, indices that explain the adequacy of the model include chi-squared to degree of freedom ratio (CMIN/df), root mean square residual (RMR), root mean square error of approximation (RMSEA), goodness-of-fit index (GFI), adjusted goodness-of-fit index (AGFI) and comparative fitness index (CFI). When CMIN/df < 2.0, RMR < 0.08, RMSEA < 0.05, GFI > 0.9, AGFI > 0.9, or CFI > 0.9, the test model is judged to be suitable. The significance of the factor load of each item was verified to confirm the concentration validity.
The reliability of this tool was analysed using Cronbach's α, which was also used to identify the internal consistency. If the Cronbach's α value is 0.7-0.8, internal consistency is good; if it is 0.8-0.9, it can be regarded as very high. However, if it is 0.9 or higher, the number of questions should be reduced (DeVellis, 2016).

| Ethics
The study was approved by the Institutional Review Board of the nursing college of [blinded for peer review] (No. IRB 2-1046881-A-N-01-201705-HR-012). When recruiting participants, the purpose of the study was explained to them and data were collected only from those who had understood and agreed in writing. Participants could withdraw from the study at any time, if desired, with no consequence. Participants were informed that all data collected during the study would be used for research purposes and discarded two years after the completed study. They were also informed that the results could be shared if requested by the government or a review agency. All participants were provided with incentives such as an eco-bag and toothbrush. Table 1 shows the general and disease-related characteristics of 399 participants. There were 42.9% and 57.1% male and female participants, respectively. Their average age was 65.88 years. The mean duration of diabetes was 11.52 years, and the most common cases were managed by university hospitals (45.9%). Of the participants, 74.4% managed their blood sugar with oral medication and only 6.8% used insulin therapy. The data were divided into two sets for EFA and CFA analysis through randomization.

| Development phase
The initial question of the DSSS was developed based on the definition, dimensions and attributes obtained during concept analysis of self-stigma for diabetes (Seo & Song, 2019). Results of concept analysis revealed that the scale was composed of nine attributes in three dimensions: affective, cognitive and behavioural factors. The nine attributes were divided into 23 indicators; using these indicators and in-depth interview data from patients with diabetes, the first 40 questions were produced. The response format involved a five-point Likert scale to continuously measure the opinions, attitudes and characteristics of the respondents.

| Content validity
After content validity, the 8th and 12th items, which had CVI lower than 0.5, were eliminated. Moreover, the 14 th , 22 nd , 26 th , 29 th , 31 st and 35 th items were incorporated into similar items. In addition, the 17th item, which the patients considered irritating, was deleted.
Thus, altogether, 14 items were eliminated, and one item was added according to expert opinion; the original 40 items were reduced to 27 items.   Table 2.

| Confirmatory factorial analysis
In the second sample (N = 216), five items were deleted and the model's goodness-of-fit test was performed according to the criteria that the factor loading should not be less than 0.40 or more than 0.95 (Song, 2011). When analysing 16 items of the four factors that were divided into comparative inability, social withdrawal, self-devaluation and apprehensive feeling, the chi-squared degree of freedom (χ 2 /df) was 2.742. The result of the model's  goodness of fit is shown in Table 3. Models were found to be suitable for CFI and RMR, but GFI, AGFI and RMSEA were found to be inadequate. Standardized factor loadings ranged from 0.54-0.95 (p < .001) (Figure 2).

| Reliability
Cronbach's coefficient of the developed tool was 0.89. Cronbach's values for each factor are as follows: comparative inability 0.88, social withdrawal 0.84, self-devaluation 0.80 and apprehensive feeling 0.74. The tool showed good internal consistency, indicating that it was reliable.

| D ISCUSS I ON
The purpose of this study was to develop and test a measure of selfstigma, which negatively affects the psychological well-being of pa- It is a comparative inability factor, suggesting that a role cannot be performed due to a lack of health. Agreement correlates with the factor of self-devaluation, as it indicates that a person identifies with social stigma. Application refers to the behavioural aspect of selfstigma and is consistent with the social withdrawal factor, which includes difficulty in or discomfort pertaining to attending meetings due to diabetes. Harm to self-esteem can be explained by the emotional aspects that result from such cognition and behaviour, consistent with factors of the apprehensive feeling that include embarrassment, resentment and fear caused by diabetes (Corrigan et al., 2012). Therefore, the DSSS covers both the formation and the consequent aspects of self-stigma.
For refining the items, the content validity was examined with diabetes specialists and patients with diabetes, the target group of the tool. In this process, some participants wept in response to questions. The self-stigma scores of the participants who wept were high. The tool was confirmed with the final 16 items. This is similar to the 12-item Weight Self-Stigma Scale (WSSS) (Lillis et al., 2010) and the 16-item The Self-

| Limitations
CFI was supported by CFA to confirm the theoretical structure of self-stigma, but there is a limitation in terms of the low value of CFI and AGFI. It can be inferred that the reason for this was the TA B L E 2 Item loadings from rotated component matrix

| CON CLUS ION
Based on Corrigan's self-stigma perspective, this study developed a 16-item DSSS that can measure the multidimensional concept of self-stigma for diabetes patients. The DSSS has secured practicality by refining items, including by establishing validity and reliability. This can contribute to the development of a nursing theory of self-stigma for diabetes. It is expected that the self-stigma measurement of patients with diabetes through the DSSS will help nurses comprehensively understand their patients and provide a basis for differentiated diabetes management strategies according to the degree of self-stigma. In addition, as a strategy to alleviate diabetes self-stigma, the constitutive factors derived through this study can be used as indices constituting the sub-domains in providing nursing interventions for self-stigma.

CO N FLI C T O F I NTE R E S T
No conflict of interest has been declared by the authors.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data generated during this study are included in this published article.

F I G U R E 2
The self-stigma model for patients with diabetes mellitus