Psychometric properties of the adapted instrument European Health Literacy Survey Questionnaire short-short form

Objective: to investigate the psychometric properties of the Brazilian Portuguese version of the health literacy questionnaire European Health Literacy Survey Questionnaire short-short form (HLS-EU-Q6) in Brazilian adults. Method: the instrument was translated and pre-tested in a sample of 50 individuals. Subsequently, it was applied to a sample of 783 adult individuals. The data went through an appropriate process of testing the properties, with the combination of techniques of Exploratory Factor Analysis, Confirmatory Factor Analysis and Item Response Theory. For the assessment of reliability, the Cronbach's alpha and McDonald's Omega indicators were used. Cross-validation with full data analysis was applied. Results: the majority of the participants was female (68.1%), with a mean age of 38.6 (sd=14.5) years old and 33.5% studied up to elementary school. The results indicated a unidimensional model with an explained variance of 71.23%, adequate factor load levels, commonality and item discrimination, as well as stability and replicability of the instrument to other populations. Conclusion: the Brazilian version of HLS-EU-Q6 indicated that the instrument is suitable for indiscriminate application in the population to which it is intended to assess health literacy levels.


Introduction
Health Literacy (HL) is a construct related to the use of multiple forms of health information in the most varied contexts (1) . Although there are several definitions, which include personal characteristics, social resources and the role of the health services in this process (1)(2) , for the World Health Organization (WHO), HL concerns the knowledge, motivations and skills of people to access, understand, judge and apply health information, in order to make decisions that help them navigate the health systems, as well as promote, prevent and care for their health (1) .
A number of studies indicate associations between individuals with low levels of HL and less participation in activities that promote health and related to disease prevention, less assertive health choices, worse self-control of chronic diseases, higher frequency of hospitalizations and cases of morbidity and mortality, with a consequent increase in costs for the health systems (1,3) . In view of this, HL is considered by the WHO as an important social determinant of health, influenced by socioeconomic and cultural characteristics and by the functioning of the health systems (1) .
Several instruments have already been developed to measure this construct in individuals and populations (4)(5) ; however, most assess only the functional characteristics of HL, that is, the personal skills to read and understand written and oral health-related information (1,6) . In order to overcome this gap, a European consortium of research institutions developed a multidimensional and integrative model of the HL and developed an instrument for its measurement consisting of 47 items, called HLS-EU-Q47 (1,7) . The HLS-EU-Q47 questionnaire assesses individual skills in understanding, evaluating and applying health-related information and was developed based on a conceptual literacy model that integrates three domains: health care (16 questions), health promotion (16 questions) and disease prevention (15 questions). Its answer options are arranged on a four-point Likert scale that ranges from 1 for very difficult to 4 for very easy (7)(8)(9) .
Although most of the research studies on HL are concentrated in the European continent, North America and Australia (1)(2) , there has been an expansion of studies in other parts of the world in the last decade, as in Brazil (15)(16) , including the creation of the Brazilian Health Literacy Network (Rede Brasileira de Letramento em Saúde, REBRALS). Bearing in mind the low level of schooling and the difficulties in understanding the professional recommendations by the Brazilian population (15)(16) , it is important that there are simple and short instruments to measure the construct of HL in this context, in order to make it applicable in the practice of the services.
In our country, the term literacy has been translated as alfabetização, literacia and letramento (17) . However (17) , although both are inseparable processes, alfabetização must be understood as the "process of acquisition and appropriation of the writing, alphabetical and orthographic system" while literacy as "the development of practical skills of reading and writing in social practices involving the written language, and of positive attitudes in relation to these practices".
Despite this fruitful research context, it is also noted that, to date, few instruments have been validated to measure HL in the Brazilian population that add broader aspects of the construct to, in addition to measuring its functional aspects, that are easy and quick to apply (18) .
Thus, the aim of the present study was to analyze the evidence of the psychometric properties of the HLS-EU-Q6 instrument, validated for Brazilian Portuguese.

Method
The research project was submitted to and approved by the Research Ethics Committee (CAAE: 58131216.5.0000.5418). Initially, Professor Kristine Sørensen, the author responsible for the instrument, was asked to authorize its translation into Brazilian Portuguese.
The instrument was translated and adapted according to the literature recommendations (19)(20) . To this end, the original version of the HLS-EU-Q47 questionnaire was translated from English into Brazilian Portuguese by two English teachers and a health researcher with knowledge of the English language.
The consensus version was then back-translated into English (back-translation) by two native Englishspeaking translators who did not participate in the first stage of the translation. A committee of experts, To calculate the sample size, a proportion of at least 15 adults was considered for each question in the questionnaire, higher than the general recommendation of 10:1 found in the literature, which allows for more accurate analyses (21) .
The HLS-EU-Q6 questionnaire is called short-short form and consists of six questions from HLS-EU-Q47 (7,9) [On a scale that goes from "very easy" to "very difficult", health promotion (7,9) .
The final individual score is a mean calculated by summing up the answers to the six questions divided by the number of items answered. The score is calculated as long as at least five of the six questions are answered differently from 1, and varies between 1 and 4, with higher values indicating better levels of HL. According to the authors of the instrument, the final score values classify individuals according to three levels of HL: inadequate (≤ 2); problematic (> 2 and ≤ 3); and sufficient (> 3) (7,(9)(10) .
For statistical analysis, data went through an extensive and robust process of testing the properties, with the combination of techniques of Exploratory Factor Analysis (EFA), Confirmatory Factor Analysis (CFA) and Item Response Theory (IRT), aiming at searching for strong evidence of validation in the construction stage and its stability for other subsamples. EFA requires the fulfillment of several stages, such as: data inspection techniques, the factor analysis method, the retention and rotation technique and the factor quality indexes (22) .

Dimensionality testing was performed by Robust
Parallel Analysis using Optimal implementation of Parallel Analysis with a minimun rank factor analysis that minimizes the common variance of the residuals (23) . The robustness of the test was determined by associating a bootstrap with a sample extrapolation to 5,000. The estimation of the polychoric matrix was performed using the Bayes Modal Estimation (24) .
Dimensionality, in the exploratory factor analysis (unrestricted model), was tested by Parallel Analysis, which has been considered one of the most effective and accurate techniques for testing the number of factors/ dimensionality (25)(26)(27) . The factors were extracted using the RULS (Robust Unweighted Least Squares) technique, which reduces the residuals of the matrices (27) .
As a complementary analysis to test the number of factors, the following techniques of unidimensionality/ multidimensionality were applied (28)  These techniques were applied to the instrument and the items. In the case of the items, they were used to guarantee and assess whether the item would adhere in a unidimensional or multidimensional manner, that is, if there was a possibility that the item would load significantly in more than one dimension. The explained variance of the instrument should be around 60% and the initial factorial loads around 0.30 (22) . In addition, mean commonality values between 0.40 and 0.60 are found (29) .
The maintenance or removal of an item from the model will depend on the magnitude of the commonality, the factor loads, the sample size and the degree with which the item can measure the factor and the absence of crossloading.
To confirm the adjustment of the factorial loads, the Normal-Ogive Graded Response Model (30) technique was used for polytomous structure, by means of the Item Response Theory. The discrimination index of the item (a) was adopted, which measures the association strength between the item and the latent variable and has a similar interpretation to the factorial loads of the exploratory factor analysis (31) to complement it. Baker's recommendation (32) was adopted that "a" <0.65 is considered to have low discrimination power; between 0.65 and 1.34, moderate discrimination, between 1.35 and 1.69, high discrimination; and above 1.70, very high discrimination. The reliability of the instrument was assessed using two indicators: Alfa (33) and Omega (34) . The adoption of two indicators sought to increase the reliability of the interpretation, as numerous reliability inconsistencies have been reported through Cronbach's alpha (35)(36) .
The replicability of the construct was assessed by the Generalized G-H Index (37) with an index greater than 0.80 (28) and, for the quality and effectiveness of the factor estimation, the Factor Determinacy Index was used, pointing for an adequate estimate values greater than 0.90, EAP marginal reliability (>0.80), sensibility ratio (SR >2) and Expected percentage of true differences (EPTD >90%). The application of multiple indicators stems from the need to certify the instrument's validity evidences by various techniques. In addition, the application and interpretation of the model's adjustment indexes (Goodness-Of-Fit -GOF), by themselves, do not guarantee that the factor analysis solution is good or useful in the practice, as it is possible to obtain satisfactory solution rates based on low quality items (38)(39) .
In order to increase the reliability and replicability of the proposed model (40) ,cross-validation was applied, as well as the Holdout technique (41) . This technique divides the bank into a training sample that can vary between 10%, 30% and 50% and another set of data, called the test bank (41) . The database was divided 50/50 with random selection of items. The Random.org website (www.random.org) and the random sequence generator technique were used to divide the groups. The banks were named as follows: complete sample (CS with 783 cases); sample 1 (S1 -training bench with 392 cases) and sample 2 (S2 -test bench with 391 cases). Another modification is that, usually, in cases of application of EFA and CFA, the tendency is to use the first training bank in EFA and the test bank in CFA (22). In this study, it was decided to apply the analysis procedures in order and MIREAL (S1 = 0.24; S2 = 0.29 and CS = 0.23).
As an extensive way of testing unidimensionality, the indices were applied to the items and the results can be seen in Table 1.   Table 2 shows the values of the factorial loads, commonality and item breakdown for the three samples.     It is essential to point out that, in the three samples tested in the CFA, both the calculation of eigenvalues by correlation and covariance indicated that the models are unidimensional.
It was verified, through the concept of correlation, that the eigenvalues were 2.70, 2.72 and 2.70, respectively for S1, S2 and CS. In addition, due to the covariance concept, the values were 1.11, 1.14 and 1.12 for the three samples. There were no other eigenvalues above 1.   was assessed by the CFA in subsamples that responded to the complete instrument, and a satisfactory factor structure was observed in the most samples from participating countries (7,9) . The analyses also bring a series of indicators that are rare to be carried out in psychometric studies, some because they are recent, and are not available in commercial software programs, and the use of more extensive data analysis techniques, which incorporate the concept of evidence of validity of the external structure by multiple indicators. There has been progress in recent years in the expansion of multiple techniques. As they point out (22) , few studies applied multiple techniques for validation analysis. Therefore, there is a substantial  (13) and in France (α=0.83) (14) , as well as a study with diabetics in Belgium (α=0.797) (12) . Furthermore, the omega values corroborated the reliability of the Brazilian version of the instrument.
Among the HL dimensions that can be assessed with the Brazilian version of HLS-EU-Q6 are the following: the evaluation and application of general health information; finding, accessing and evaluating information for disease prevention and health promotion (7,9) . The application of instruments for tracking HL skills in the reception in the health services has been recommended to qualify the data collection of users and as a means for the health professionals, including nurses, to guide their care, being considered as the "sixth vital sign" (1,45) .
The percentage of individuals with inadequate HL levels in the present study (46.3%) was higher than the mean found in the countries of the European study (9%), as well as in France (5%), Italy (8.9%) and Belgium (9.8%) (7.9,12-14). This may have occurred due to different socioeconomic characteristics of the populations analyzed, since the individuals' schooling and income levels in the aforementioned studies were much higher than those of the present study. In addition, the cognitive, cultural, organizational characteristics of the educational and health system may have contributed to these differences (7,9,13) . Although it is not the objective of this study, the identification that only 2% of the participants had sufficient levels of HL, that is, they were able to find, access, understand, evaluate and use the health information indicated for the importance of the www.eerp.usp.br/rlae

Conclusion
The Brazilian version of the HLS-EU-Q6 instrument indicated diverse evidence of adequate internal structure validity for measuring the health literacy levels of Brazilian adults. Therefore, it is a tool that can be easily used in the clinical practice, capable of quickly and objectively measuring the limitations in access, understanding and use of health information, whether for disease prevention or for health promotion.