Self-efficacy scale for Brazilians with type 1 diabetes

ABSTRACT CONTEXT AND OBJECTIVE: Diabetes is a public health problem and good glycemic control is able to prevent or contain its complications. Self-efficacy is a key factor in successfully achieving behavior goals. The aim of this study was to analyze the psychometric properties of the insulin management diabetes self-efficacy scale (IMDSES) on type 1 diabetes patients from southern Brazil. DESIGN AND SETTING: Validation study in two cities in southern Brazil. METHODS: The psychometric properties of IMDSES were evaluated in a population of type 1 diabetes patients (n = 213), from September to December 2004, who were attended within the Brazilian public healthcare system. Principal component analysis was conducted to develop the subscales. Cronbach’s alpha was used as the reliability coefficient. RESULTS: The analysis of psychometric properties resulted in an IMDSES consisting of 20 items and three subscales: diet (alpha: 0.83), insulin (alpha: 0.92) and general management (alpha: 0.78) and accounted for 53% of the variance. Criteria validity was investigated through two parameters: glycohemoglobin, which showed significant association with self-efficacy on the insulin subscale (p = 0.04), and the variable "adherence", which was significantly associated with self-efficacy on two subscales (p < 0.05). CONCLUSIONS: This study shows that the IMDSES is valid and reliable, and can be used to measure results from diabetes educational programs and to measure self-efficacy relating to diabetes management, for possible interventions.


INTRODUCTION
Diabetes mellitus is a public health problem both in developed and in developing countries. There are about 5 million diabetics in Brazil, although half of them are unaware of their condition. 1 It has been estimated that, by 2025, this disease will affect 11.6 million Brazilians and that Brazil will be among the 10 countries with the highest prevalence of this disease. Diabetes is among the 10 major mortality causes in the country. 2 It is well documented that individuals with type 1 diabetes face a complex treatment regimen after their disease has been diagnosed, encompassing a series of daily decisions that demand active behavioral involvement by the patient. One study has indicated that patients consider the daily self-care regimen to be more diffi cult than the diagnosis. 3 Adherence to treatment is low according to data in the literature, 4 and inadequate control of the disease may lead to acute and chronic complications in several organs, such as the kidneys, eyes and heart, and in the circulation. 5,6 Therefore, it is important to have good glycemic control, and self-effi cacy is the key factor in successfully achieving behavioral goals.
According to Bandura, self-efficacy reflects the individual's belief in his/her ability to successfully perform specifi c activities. Belief infl uences goals, motivation and perseverance relating to the quality of analytical thought and causal attribution of success and failure. 7 The sense of self-efficacy is based on four sources: the individual's own signifi cant experiences, resulting from interpreting proposed performance; vicarious experiences that alter opinions about effi cacy through transmission of competencies and comparison with other patients' accomplishments; verbal persuasion that the individual possesses certain abilities, and similar types of social infl uence; and physiological and affective states, from which people partially judge their abilities, resistance and vulnerability to dysfunction. 8 Belief in personal ability to undertake certain behavior infl uences choices, situations that will be attempted or avoided and perseverance in performing the task. 9 The value of self-efficacy in predicting self-care behavior relating to diabetes has been verifi ed in the literature, through studies in which self-effi cacy was associated with self-reporting of adherence to treatment among adolescents [10][11][12][13][14][15] and adults. 16,17 It has been related to good glycemic control 18 and to perceived improvement of general health and social functioning. 19 Self-effi cacy has been successfully used in educational programs for diabetes patients. [20][21][22][23] The lack of instruments adapted and validated for the Brazilian population has been a barrier for conducting studies relating to this population. We found in the literature many instruments that evaluate self-effi cacy in diabetes cases: the Diabetes Empowerment Scale (DES) and the Diabetes Self-Effi cacy Scale (DSES), which evaluate self-effi cacy in type 1 and 2 diabetics; 24,25 the Self-Effi cacy in Adolescent Girls and Boys With Insulin-Dependent Diabetes Mellitus scale, which is specifi c for type 1 diabetics; 16 and the Insulin Management Diabetes Self-Effi cacy Scale (IMDSES).
The instrument chosen for adaptation was the IMDSES, which is from the United States and was developed by Hurley and Harvey to evaluate self-effi cacy among diabetics. 26

OBJECTIVE
This study was carried out in order to translate, adapt and evaluate the psychometric properties of the IMDSES, in a sample of type 1 diabetics from southern Brazil.

Sample Sample
The sample was made up of 213 individuals. The sample size was calculated using an average of 7 individuals per scale item with a margin of 10%. 27 The participants were individuals with a diagnosis of type 1 diabetes who were registered within the Brazilian national health system (Sistema Único de Saúde, SUS) in two cities in southern Brazil. The study was conducted between September and December 2004.
The patients were selected at random according to their respective registrations in public service pharmacies that provided insulin. The inclusion criteria were: age of 14 years or older (data in the literature indicates that from this age on the individual realizes his/her need for self-help 16 ); type 1 diabetes (defi ned as beginning before reaching 40 years of age and treated with insulin from the time of diagnosis); diabetes diagnosed more than six months before the study period; and self-care of diabetes, meaning that the patients were looking after themselves without the aid of third parties. Pregnant women were excluded.

Original instrument Original instrument
The IMDSES instrument evaluates selfeffi cacy relating to self-care of diabetes, among subjects who need to take insulin. It was devised in the United States and contained 28 items, with questions for which the responses were on a Likert-like scale of six points, from 1 (strongly agree) to 6 (strongly disagree). 28 Seven types of behavior were evaluated: general management, diet, exercise, care with feet, glycemia monitoring, insulin administration and detection, prevention or treatment of hypoglycemia/hyperglycemia. It was originally validated among 142 adults who were insulin-requiring diabetics. The psychometric properties of the original version were adequate. Cronbach's alpha for the whole scale was α = 0.82, and for the three domains (described below) it was: general management, α = 0.67; diet, α = 0.78; and insulin, α = 0.77. The test-retest stability was acceptable (r = 0.58, p < 0.002) and was followed up two weeks later.
An item-by-item self-reported assesment instrument called the insulin management diabetes self-care scale (IMDSCS) was developed and applied to all participants, before they responded to the IMDSES and underwent the glycohemoglobin test. 26

Procedures for adaptation Procedures for adaptation
After approval of the study by the Research Ethics Committee of Hospital Santa Casa de Misericórdia de Pelotas, in Pelotas, Rio Grande do Sul, the work of adapting the instrument started from a theoretical analysis of the items. 27 The instrument was translated from English to Portuguese by two bilingual experts, and then a back translation was made by other two translators who had not seen the original scale, in order to verify the equivalence of terms between the two versions.
For the theoretical analysis of the items, 20 doctors from Rio Grande do Sul who were specialists in diabetes evaluated the adequacy of these items and their pertinence to the respective behaviours. Any item with an 80% agreement between these judges was considered satisfactory. 26 Semantic analysis was done among 10 insulin-requiring patients, to analyze their understanding of the questions and instructions of the scale, and then a pilot study was conducted with another 10 patients. In this study, we found that there was some diffi culty in understanding some items about insulin management, but not with regard to type 1 diabetics, and therefore we decided to adapt the scale for use only among type 1 diabetics.
Some changes were made to the instrument to improve understanding. All the items were set in an affi rmative mode, whereas the original instrument had positive and negative items. We changed the Likert scale to use only four points: 1-strongly agree; 2-agree; 3-disagree; and 4-strongly disagree. We also introduced the possibility of marking a fi fth alternative, named "not applicable", which was computed as missing information. Items 6 and 10 were changed by adding examples. The fi nal instrument is presented in Annex 1.
For interpreting the scale, reverse scoring was used. This corresponded to the mean scores computed for each item of the respective subscale. As in the original scale, no cutoff point was recommended in this scale, because self-effi cacy is a dynamic concept. If we had had to recommend a cutoff point it would have been located in the upper third of the total score for the scale, because within this range there was greater correlation with high levels of self-effi cacy in the statistical analysis.
To the initial instrument was added a questionnaire on the sociodemographic characteristics of the study population, which included the following questions: name, address, telephone number, skin complexion, birth date, sex, schooling in completed years of study, social class as evaluated by the criteria of the Brazilian Association for Population Studies (Associação Brasileira de Estudos Populacionais, Abep), date of diabetes diagnosis, date of starting insulin use, types of insulin used, amounts of insulin and number of bottles used in a month, and whether the individual carried out his/her selfhelp without the aid of third parties.
After agreeing to take part in the study, the patients signed a consent form and, in the case of patients under 18 years of age, their parents gave their consent in writing. The data were collected in the laboratory recommended by the researcher or in the homes of interviewees who lived out of town or who had diffi culty in reaching the indicated location. The instrument was administered by previously trained psychology students from the Catholic University of Pelotas (Universidade Católica de Pelotas, UCPEL) in the form of a dialogue interview. Blood was collected to carry out the glycohemoglobin test.
Two months later, the retest was performed on 54 patients who were selected randomly from among the initial sample.

Data analysis Data analysis
The Statistical Package for the Social Sciences (SPSS) for Windows software, version 8.0, 29 was used for processing the analysis.
To study the construct validity of the instrument, i.e. whether the test does in fact measure what it proposes to, multivariate analysis was carried out using principal components analysis. 30,31 Univariate analysis (frequencies and means) was conducted to observe the distribution and to establish cutoff points when necessary, both for sociodemographic variables and for subscales of the instrument.
To study criterion validity, two parameters were utilized: glycohemoglobin (GHb) and "adherence". Unlike in the original IMDSES scale, the IMCSCS scale was not used because this is not considered to be valid in Brazil. GHb was measured using the high-performance liquid chromatography (HAPLY) method through ion exchange (reference: < 7%) in a single laboratory and was treated as an ordinal variable. The variable "adherence", which is behavioral, was measured by calculating the relationship between number of units used and number of insulin bottles consumed during the period. The latter was given by an external criterion (number of bottles supplied by pharmacies to diabetics during that period). Therefore, a categorical variable was established (yes or no). Analogies for using this criterion was made with studies that have correlated adherence with the accurate method of counting pills. [32][33][34][35] It was established that, for statistical analysis, "self-effi cacy" would be converted into a variable divided into terciles. For the ordinal variable GHb we used analysis of variance (ANOVA), and for the "adherence" variable, the chi-squared test (bivariate) and logistic regression (multivariate) to evaluate possible confounding factors in the results.
The reliability of the instrument was determined by internal consistency analysis, using Cronbach's alpha. To evaluate temporal stability we used test-retest and Pearson's correlation.

General characteristics
General characteristics of the sample of the sample 221 subjects had been selected to take part in the study, but eight dropped out: one because he was dependent on third-party help and seven because they refused to take part. Thus, the resultant fi nal sample was 213, of which 47.9% were men and 52.1% were women. Their mean age was 33.9 ± 14.97 years and most of them were fairskinned (90.1%). The predominant social class was C (44.1%), and the mean schooling level in completed years of study was 8.9 ± 0.27 years. The length of time since diagnosis was 12.1 years. The marital status of 49% was single. The mean GHb level was 10.28 ± 2.3%.
The retest sample comprised 54 patients. This sample presented some differences in characteristics, in relation to the test sample, with majorities of men (58.2%) and singles (61.8%). The mean age was lower: 30.9 ± 15.2 years. The predominant social classes were B and C (73%) and 45% had studied for up to eight years. The mean length of time since diagnosis was 11.4 years and the mean GHb level 9.7 ± 2.6 %. However, the differences between the test and retest samples were not signifi cant.

Psychometric characteristics Psychometric characteristics of the instrument of the instrument
In the initial analysis with 28 items, it was found that eight of them presented factorial loads of less than 0.30, which led to their removal.
In the study of construct validity with 20 items, factorability was shown by KMO coeffi cient of 0.86, Bartlett's sphericity test (p < 0.001), signifi cant inter-item correlation coeffi cients (p < 0.001) and communalities that oscillated from 0.55 to 0.86.
Five components presented eigenvalues greater than 1.0 in the initial principal components solution. The scree plot showed three distinct visual components that accounted for 53.6% of the variance (Figure 1). After considering the interpretability of the components and analyzing possible solutions, it was decided to extract three subscales (diet, insulin and general management). Their factorial loads ranged from 0.49 to 0.90, and it had been previously decided that only items with factorial loads greater than 0.30 would be considered 27 (Table 1).
In the criterion validity analysis for the GHb level , ANOVA testing on the insulin subscale was signifi cant (p = 0.040), showing a linear tendency (p = 0.01) between groups: the higher the self-effi cacy level was, the lower the GHb level was (Figure 2). Regarding "adherence", it was seen that the higher it was, the higher the levels of self-effi cacy were on the two subscales of diet and insulin. The chisquared test showed a signifi cant association on the subscales of diet (p = 0.04) and insulin (p = 0.03), but not on the general management subscale (p = 0.24). In the logistic regression analysis, after adjustment for age, sex, social class and schooling, it was seen that subjects with low self-effi cacy had approximately twice the chance of showing "non-adherence" on the subscales of diet and insulin, in comparison with the 33% who showed the best self-effi cacy on these subscales.
The reliability of the instrument was determined by internal consistency analysis using Cronbach's alpha (α), which verifi ed whether the subscales identifi ed were consistent and homogeneous. The coeffi cients were as follows: diet (α = 0.83), insulin (α = 0.92) and general management (α = 0.78).

DISCUSSION
The results from this investigation relating to psychometric properties are promising. The content validity, as verifi ed by theoretical analysis of the items and semantic analysis of the scale, was adequate. In the semantic analysis, we tried to evaluate the items with subjects from extremes of social classes, so that the language would be adequate for all levels, as recommended in the literature. 27 In the pilot study, it was decided to make changes in three items and to reduce the number of scores on the Likert scale, in order to improve understanding. Diffi culty in understanding Likert-like scales with many score options have been reported in the literature among populations of low socioeconomic level. 28 Regarding construct validity, three subscales were maintained, as in the original instrument, but the number of items was reduced to 20 and the instrument obtained more reliable measurements. The wording of item number one was modifi ed, after being identifi ed in the diet subscale. In the original instrument, it had belonged to the general management subscale (Annex 1).
Among the parameters used for analyzing the external criterion validity, GHb is an important test that refl ects the mean glycemic control over the preceding three months. Its mean in our sample was above the recommendations for the method (10.28 ± 2.33) and the mean from the original scale (10.99 ± 2.36), thus showing ina-dequate metabolic control in both populations.
If the GHb level is adequate, it may prevent or delay diabetes complications. 36,37 It was found that there was a signifi cant association between the self-effi cacy level on the insulin subscale and GHb, which was confi rmed by ANOVA. There was a linear trend in which the higher the selfeffi cacy level was, the lower the GHb level was. In the original instrument, there were signifi cant associations with GHb on all subscales, which was not observed in the present study.
Studies in the literature have correlated self-effi cacy with glycemic control using GHb 14,15 but, according to Glasgow et al. (1999), 36 it is inappropriate to use GHb as an indicator for a patient's behavior. High GHb may be attributable to behavioral problems or other factors (inadequate prescription or diseases, among others) that must be analyzed. Adherence is an important contributor towards good control, but it is not the same as control and should not be interpreted on the basis of a single laboratory test. 36 After searching for a behavioral characteristic that would reflect diabetes management, the variable "adherence" was defi ned. It was observed that there was a signifi cant association between "adherence" and self-effi cacy in the correlation analysis, on the subscales of diet (p < 0.02) and insulin (p < 0.02). This was also found in the logistic regression analysis, thus indicating that the scale has criteria validity, i.e. the subscales of diet and insulin predict "adherence" or diabetes management. Hence, in the multivariate analysis, after adjustment for age, sex, schooling and social class, and using a cutoff point of 33% for subjects with better self-effi cacy levels, there was a signifi cant association between self-effi cacy and "adherence" (adequate behavior regarding the use of insulin and diet).
In the reliability study, Cronbach's alpha for the subscales ranged from 0.92 for insulin factor to 0.78 for general management. Alpha values can range from 0 to 1, and 0.65 is the lowest acceptable value for the scale to be considered reliable. 38 But considering 0.80 as the expected index for a measurement with less error, the value for the general management subscale was acceptable. 27 The evaluation of temporal stability using test-retest did not show any signifi cant association with the insulin subscale. It is likely that the two-month period was too long and interfered with the results, considering that self-effi cacy is a dynamic concept and can change with time. In the original instrument, the second test was performed after two weeks because of this possibility.
There is evidence that behavioral patterns relating to diabetes care are relatively independent. Characterization for the specifi c behavioral area that presents impairment allows more effective interventions. 37 It is known that selfeffi cacy is specifi c for each behavioral pattern, 7 thus explaining the fi ndings from this study.
Our study presented certain limitations such as the way in which the instrument was applied: it was done by interview and not selfapplication as the original, and this may have affected the replies. 27 Considering that the original instrument was designed for insulin-requiring diabetics, whereas for our population it was decided to adapt it only for type 1 diabetics, no generalization for all diabetics is possible.

CONCLUSION
It was concluded that, despite the modifi cations made to the instrument, in relation to the original questionnaire, there was no loss in internal consistency in its adaptation. It presents adequate psychometric parameters and can be used to evaluate self-effi cacy in the management of type 1 diabetes.
It is suggested that further studies in different regions of the country should be conducted to take the cultural diversity into account, using different parameters to verify the criterion validity, as well as several samples of diabetics with investigation of temporal stability over a shorter time interval.