Applicability of the Spoken Knowledge in Low Literacy Patients with Diabetes in Brazilian elderly

ABSTRACT Objective To translate, adapt and evaluate the properties of a Brazilian Portuguese version of the Spoken Knowledge in Low Literacy Patients with Diabetes, which is a questionnaire that evaluate diabetes knowledge. Methods A cross-sectional study with type 2 diabetes patients aged ≥60 years, seen at a public healthcare organization in the city of Sao Paulo (SP). After the development of the Portuguese version, we evaluated the psychometrics properties and the association with sociodemographic and clinical variables. The regression models were adjusted for sociodemographic data, functional health literacy, duration of disease, use of insulin, and glycemic control. Results We evaluated 129 type 2 diabetic patients, with mean age of 75.9 (±6.2) years, mean scholling of 5.2 (±4.4) years, mean glycosylated hemoglobin of 7.2% (±1.4), and mean score on Spoken Knowledge in Low Literacy Patients with Diabetes of 42.1% (±25.8). In the regression model, the variables independently associated to Spoken Knowledge in Low Literacy Patients with Diabetes were schooling (B=0.193; p=0.003), use of insulin (B=1.326; p=0.004), duration of diabetes (B=0.053; p=0.022) and health literacy (B=0.108; p=0.021). The determination coefficient was 0.273. The Cronbach a was 0.75, demonstrating appropriate internal consistency. Conclusion This translated version of the Spoken Knowledge in Low Literacy Patients with Diabetes showed to be adequate to evaluate diabetes knowledge in elderly patients with low schooling levels. It presented normal distribution, adequate internal consistency, with no ceiling or floor effect. The tool is easy to be used, can be quickly applied and does not depend on reading skills.


INTRODUCTION
Diabetes is a chronic degenerative disease commonly observed in elderly, and its prevalence has increased fastly in recent decades. (1) Strong evidence demonstrated the need of appropriate glycemic control to prevent microand macrovascular complications. (2,3) Inadequate glycemic control in adults is due to cognitive impairment, sensory loss, polypharmacy, depression and poor compliance. (4)(5)(6) Management of type 2 diabetes also involves learning about the disease and adopting self-care. (7) In this context, the individuals with limited educational background and insufficient health literacy tend to present more difficulties. (8,9) Several tools designed to evaluate diabetes knowledge have been developed in the last decades. Unfortunately most of these involve reading ability and are based on complex scales, with limited clinical applicability, especially for individuals with low schooling level. (10,11) Rothman et al. developed the Spoken Knowledge in Low Literacy Patients with Diabetes (SKILLD), (12) a tool originally designed in English to assess knowledge on this disease. Since SKILLD is applied verbally, it tests parameters that are independent of reading capacity. The questions are written in simple language and the level of difficulty has been adapted for individuals with low schooling level. Furthermore, the open questions allow the individuals to explain their answers in their own words.
Since its publication, SKILLD had neither been adapted to Portuguese language nor used as a questionnaire for elderly people in Brazil, a country where the population aged over 60 years presents low schooling levels. (13) OBJECTIVE To translate, adapt and assess the psychometric properties of the Spoken Knowledge in Low Literacy Patients with Diabetes.

METHODS Participants
We recruited individuals treated at the reference outpatient clinic for elderly at the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (FMUSP), between June 2011 and July 2012. We enrolled the individuals who had enough time available to answer the questions. A total of 225 type 2 diabetics aged ≥60 years were assessed. All individuals received free medications from the pharmacy of the organization. None of the participants was participating in diabetes education programs, and all had access to the same healthcare services.
One researcher talked to participants in the waiting room, where they waited for routine medical appointment with geriatricians or residents under training at the outpatient clinic. After accepting the oral invitation to participate in the study, the participants were taken to another room, away from external factors that could impact on their attention and concentration, to answer the questions. All information obtained in the study was classified. The same researcher interviewed all participants.
The inclusion criteria were age ≥60 years, diagnosis of type 2 diabetes under treatment with oral medication or insulin, (7) glycosylated hemoglobin (HbA1c) measured during the last six months, and oral fluency in Portuguese. We excluded individuals with diagnosis of dementia, based on the medical chart report, and those with visual, auditory, motor or language impairment that hindered interaction with the examiner. The situations that could affect accuracy of HbA1C measurement were additional exclusion criteria: hemoglobin <11mg/dL; thyroid stimulating hormone (TSH) <0.1 or >10mU/L. and glomerular filtration rate estimated by Cockcroft-Gault formula <30mL/min/1.73m 2 . (14)(15)(16) Furthermore, we also excluded participants who presented frailty syndrome, because less aggressive glycemic control targets have been proposed for this group. (7,17) For this purpose, we used the criteria described in the Study of Osteoporotic Fracture (SOF), (18) which include more than 5% weight loss in the previous year; inability to sit on a chair and stand up five times in a row; reported feeling of lack of energy assessed by question: "Do you feel full of energy?".

Spoken Knowledge in Low Literacy Patients with Diabetes
This tool consists of ten questions associated to diabetes knowledge. Only complete answers are accepted and the score varies between zero and 100%. Higher scores are associated with better understanding of the disease. At first, the participant is given 10 to 15 seconds before answering each question. Should the participant fail to answer at this first attempt, the question is reformulated for easier understanding, asked again and further 15 seconds are given for answer. The total time to answer the questionnaire varies from 5 to 10 minutes. (12) The Brazilian version of the SKILLD was independently translated and adapted by two native Brazilian physicians who are fluent in Portuguese and English. The physicians were aware of the aims of the study and had good knowledge of the target population this tool would be applied to. A third physician, who is an experienced member of the group for discrepancy resolution, revised both translations and produced a final version. A pilot study was conducted to test the final version in ten diabetic subjects, who were chosen by convenience to identify possible language and cultural problems. This pilot study yielded satisfactory results revealing that the tool required no changes.

Sociodemographic and clinical data
The following sociodemographic data were obtained: age, years of schooling, race, marital status (married versus non-married), previous occupation (blue collar worker or not). The socioeconomic classes were assessed according to the Criterion for Economic Classification Brazil (CCEB -Critério de Classificação Econômica Brasil), (19) which provides a continuous scale that is calculated based on the scores associated with the amount of household items and the schooling level of the person with the highest income in the house. Based on the score, we obtained assessment intervals, which are classified into five subgroups: A (35 to 46), B (23 to 34), C (14 to 22), D (8 to 13) and E (zero to 7).
Participants were also assessed on the duration of diabetes, type of treatment (oral medication or insulin therapy) and assistance with medication (help to obtain, organize, remember when to take medication, and those who were totally dependent).
The most recent HbA1C measurement taken within the last six months was used to assess control of diabetes. Therefore, we used the high performance liquid chromatography (HPLC), and the results were extracted from the patient's electronic file available at the service. For the purpose of this study, glycemic control was considered inadequate when HbA1c ≥7%.
Previous studies demonstrated that depressive symptoms may influence glycemic control. (5) To assess depressive symptoms we used the Geriatric Depression Scale with 15 items (GDS-15). The GDS-15 is a questionnaire with dichotomous answers (yes or no) and provides a continuous measure of severity. We used the version that had been validated for Brazil, comprising 15 items. A score ≥5 indicates depression. (20) For the assessment of functional health literacy we used the Short Assessment of Health Literacy for Portuguese-speaking Adults (SAHLPA-18), (21) which is a tool validated for Portuguese that measures understanding and pronunciation of relatively common medical words.
During the literacy assessment cards are used. On the top of each card, a medical term is written in bold letters. At the bottom of the card, there are two additional words, but only one is associated with the word on the top of the card. At first, the interviewee is asked to read out loud the word on the top of the card. Then the interviewer reads out aloud the words on the bottom of the card, and asks which is associated with the medical term. Each individual is given a set of 18 cards, and a score is attributed to each item considering correct association and pronunciation. The final score varies from zero to 18. Individuals who score ≤14 are classified as people with insufficient health literacy.

Statistical analysis
The description of interval variables included mean and standard deviation. After normality assessment test, based on histogram graphs, we carried out a bivariate analysis with parametric statistics.
In order to evaluate the association between the level of knowledge about diabetes with sociodemographic and clinical characteristics, the participants were divided into two groups: adequate knowledge (SKILLD >50%) and inadequate (SKILLD ≤50%). (20) To compare category variables between these two groups we used the χ 2 test. For comparison of interval variables, the Student´s t test for independent samples was employed.
The SKILLD score was also evaluated according to intervals. Univariate and multivariate regression models were created to investigate the associated factors regardless of the level of knowledge on diabetes. According to this model, the SKILLD score was defined as a dependent variable. The explanatory variables considered for insertion in this model were age, sex, schooling, insufficient functional literacy, marital status, economic level, duration of disease, use of insulin, and glycemic control. We used the backward strategy to insert variables that had a significance level <0.1 in the bivariate analysis and those with great clinical relevance in relation to SKILLD.
In order to evaluate the internal consistency of the tool we used Cronbach's a. To determine the properties of each item we calculated adjusted item-total correlation, changes in Cronbach after removal of each item, point biserial correlation coefficient associated with schooling and SAHLPA-18. To investigate the factorial structure of the instrument, we carried out an analysis of the main components in a tetrachoric correlation matrix. The sedimentation graph is derived from the self-scores attributed to visual inspection, and Horn's parallel analysis was conducted to compare the dimension of self-scores obtained by the analysis of main components with those obtained by 100 random samples.
Two-tailed test were used to assess data. A p value of 0.05 was considered statistically significant. The analysis were carried out using Statistical Package Social Sciences (SPSS) version 20.0 and Stata version 13.0 This protocol was approved by the Research Ethics Committee of the organization and is registered under 0534/11. All participants signed an Informed Consent Form. In the case of illiterate participants, the form was read out and explained aloud to them, in the presence of an impartial witness. The form was signed by the participant, or by the legal representative.

RESULTS
We invited 225 type 2 diabetic individuals to participate in the study. A total of 90 were excluded for presenting one of the following characteristics: dementia (51), frailty syndrome (13), severe visual impairment (10), severe hearing impairment (1), renal failure (9), lack of fluency in Portuguese (3), anemia (2), aphasia (1). Of the remaining individuals, six refused to participate in the study. Thus, a total of 129 participants were included in the analysis.

Variables associated with Spoken Knowledge in Low Literacy Patients with Diabetes
In assessing the differences between both groups classified according to SKILLD regarding the level of knowledge about diabetes (sufficient versus insufficient), we observed a statistically significant difference in schooling level (6.6 (±5.  Cronbach's a coefficient was 0.75, revealing adequate internal consistency. As shown on table 3, nine out of ten questions presented question-total correlation ≥0.4, which indicates good discrimination power. The question associated with correct frequency of physical exercise was the only one that presented a lower question-total correlation (0. 20), and the only one that resulted in an increase in Cronbach's coefficient when removed (0.76).  In the multivariate linear regression model corrected for blue-collar work, the following variables were independently associated with SKILLD: schooling (B=0.193; p=0.003), use of insulin (B=1.326; p=0.004), duration of diabetes (B=0.053; p=0.022), and SAHLPA-18 score (B=0.108; p=0.021). The determination coefficient was 0.273, and p<0.001 (Table 2).

Psychometric properties of Spoken Knowledge in Low Literacy Patients with Diabetes
Spoken Knowledge in Low Literacy Patients with Diabetes presented normal distribution and mean value of 42.1% (±25.8%). The final translated model including ten questions and answers is described in chart 1. The questions with the highest percentage The main component analysis revealed two factors with self-value above 1, the first of 4.79, explaining 47.9% of variance, and the second of 1.29, explaining 13% of additional variance. Visual inspection suggested a one-dimensional structure. Horn's parallel analysis revealed only one factor with a self-value greater than the mean corresponding to random samples (4.79 versus 1.46), confirming the one-dimensional nature of the tool (Figure 1).

DISCUSSION
The SKILLD score proved to be appropriate for this sample of elderly people. It presented normal distribution, adequate internal consistency and with no ceiling and floor effect. The SKILLD has some advantages when compared to other instruments used to evaluate diabetes knowledge. Since it is a simple oral test, it is easy to use in individuals with low schooling level. Moreover, other tools include questions associated with the pathophysiology of the disease, which do not seem to be essential for adequate self-care. (10,11,22) The frequency of errors observed in the SKILLD reflects the low schooling of the sample population. Other authors who evaluated individuals with low schooling observed a similar proportion of right and wrong answers related to the topics of the questionnaire. (12,23) In our study, 82.9% of participants had incomplete high school. This percentage is even higher than that observed in the SKILLD validation study, where 40% of participants had incomplete high school. (12) Only 10.9% of individuals answered correctly the question about the ideal value of glycosylated hemoglobin. The correct answer to this question was glycosylated hemoglobin should be below 7.0%. Nevertheless, it is important to point out that current guidelines suggest that this cutoff point is not appropriate for all elderly individuals, because there are risks associated with intense glycemic control, especially in more fragile individuals, who present loss of functional capacity. (7,17,24) We also observed that some participants had trouble answering the question about exercise, because according to them exercise should be more often and more intense than indicated. The analysis of the psychometric properties of the score revealed that this question presents low item-total correlation and that withdrawing it would increase internal consistency, suggesting that this question is of limited use and could possibly be excluded.
Regression model analysis showed an association between SKILLD and schooling, use of insulin, duration of disease and functional health literacy. The determination coefficient of the final model was 0.273, which indicates that the variation in SKILLD is for the most part explained by factors that were not represented in this model. Despite the fact that SKILLD is an oral test, schooling and literacy influenced the results, because these individuals probably have less access to information. (25) Rothman et al., reported similar results, and observed a correlation between the SKILLD and the variables schooling level, health literacy, duration of disease, and use of insulin. Contrary to our results, these authors also observed a negative correlation between HbA1c levels and SKILLD score. It is important to point out that the validation study did not test the association of each variable and the results in an independent manner. (12) We must also mention some limitations of our study. It was not possible to establish causality in the relations between SKILLD and other variables, for being a cross sectional study. We underline the absence of criterion validation, since the translated SKILDD version has not been compared with another questionnaire about diabetes knowledge. This study did not allow evaluation of this criterion, because most tools that measure diabetes knowledge were originally described in English, and the instruments validated in Brazil have not been adjusted for low-schooling populations, for involving reading ability and complex questions on pathophysiology of disease. (25) Neither did we evaluate reliability parameters, such as inter-examiner agreement and test-retest stability. Another limitation of the study was the subjective influence of the professional when scoring the answer of participants. In order to avoid measurement biases, the same researcher carried out all SKILLD evaluations.
The participants of our study were recruited by convenience, in a single tertiary care hospital, which limits generalization of results. Some factors that might be associated with knowledge on diabetes were not included in this study, and must be taken into account in future studies. Among these factors, we include cognitive performance, complexity of therapeutic regimen, diabetes education, motivation and attitude regarding treatment. (4,7,17) Further studies must be carried out to compare SKILLD with other questionnaires on diabetes knowledge, and to assess the inter-examiner and test-retest reliability.

CONCLUSION
The Portuguese version of Spoken Knowledge in Low Literacy Patients with Diabetes is adequate to evaluate diabetes knowledge in elderly patients with low schooling. This tool is easy to use, can be quickly applied and does not depend on reading ability of patients. In our sample, Spoken Knowledge in Low Literacy Patients with Diabetes was independently associated with the following variables: schooling, use of insulin, duration of disease and functional health literacy.

CONTRIBUTION OF AUTHORS
Jonas Gordilho Souza participated in the development of the study, data collection, statistical analysis and drafting of the manuscript. Daniel Apolinario took part in the development of the study, statistical analysis, and manuscript revision. José Marcelo Farfel, Omar Jaluul, Regina Miksian Magaldi, Alexandre Leopold Busse, Flávia Campora and Wilson Jacob-Filho, participated in the development of the study, and manuscript revision.