AGREEMENT OF NHP AND SF-36 SOCIAL DOMAINS: AN EXPLORATORY STUDY

| Introduction: Health Related Quality of Life (HRQoL) is an important outcome measure to investigate and monitor patients with chronic diseases. In order to achieve such goal, it is essential to choose an appropriate tool to evaluate the peculiarities of each population. Instruments to investigate HRQoL, NHP and SF-36 are alike in several aspects and are considered by the scientific community as interchangeable amongst each other. However, there are doubts regarding the agreement of their social domains. Objective: The present study aim to assess whether the social domains of the NHP and SF-36 correspond when applied in a population with chronic disease. Methods: The present research was made by applying the two instruments to a population with chronic disease. The social domains agreement was evaluated by Pearson correlation, ROC curve, AUC, Youdex Index, and Bland-Altman plot. Results: It was achieved a weak correlation between the two social domains (r = 0.3), confirmed by ROC curve with small AUC (0.416) and Youdex Index around 0.0. The Bland-Altman plot reaffirmed there is no agreement between the two subscale by achieving mean difference equal to 29.37 (± 38.51), which made the concordance Interval vary from 106.39 to -47.65. Conclusion: it was concluded that social domains of NHP and SF36 do not agree to each other. However, reasons for that are not explained and need to be investigated in futures studies. The current research was approved by the Research Ethics Committee of the BAHIANA School of Medicine and Public Health under the protocol CAAE 516.42315.5.0000.5544. Key-words: Data Accuracy; Quality of Life; Surveys and

Quality of life (QoL) is defined by WHO as "the individual's perception of their position in life in the context of culture and value systems in which they live, and in relation to their goals, expectations, standards and concerns" 1 . From this definition it is possible to notice that QoL concept is closely related to the context of culture and value system in which the individual lives. This understanding comes from the context in which the concept of QoL was first formulated: sociology and anthropology 2 . Originally, QoL was not studied by health sciences, but when such appropriation happened some elements were added to the inicial concept: functional losses, changes in relationships with society as a result of morbidity situation or physical limitation, the perception of the relationship with the health care system in its economic and political organization, as well as the value attributed by individuals to life expectancy when it is modified by the perception of physical or psychological constraints 3 .
Thus, to investigate Health Related Quality of Life (HRQoL), it is important to ascertain whether and to what extent morbidity is interfering with the individual's perception of their position in life, society and culture in which he/she lives, as well as his/hers objectives, expectations and concerns 1 . It is understandable therefore the great importance that the assessment of the social dimension have when investigating HRQoL.
Among the instruments to assess general population HRQoL, the Nottingham Health Profile (NHP)4,5 and the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) 6,7 are well accepted by the scientific community and patients. Such instruments are alike in several aspects 8,9 and are considered by the scientific community as replaceable each other. However, there were doubts regarding the agreement of their social domains 6,8,10,11 .
In order to clarify this question, there was made a systematic review that showed clear results concerning the correlation coefficients between social domains 12 . However, there are severe criticisms to this statistical technique when applied to compare two instruments. It is claimed that the correlation test is appropriate to assess the association between two variables.

METHODS
It was conducted a study to test the agreement of SF-36 and NHP social domains. The population was of individuals over 18 years with confirmed diagnostic of rheumatoid arthritis (RA), from a reference outpatient clinic for treatment of collagenosis, in the city of Salvador, Bahia, Brazil. Participants were interviewed between October 2011 and July 2012. The exclusion criteria covered limited understanding of the research instruments, or chronic degenerative comorbidities with potential to be confounders such as neurological, orthopedic, cardiac or pulmonary disorders.
Potentially includible individuals were invited to participate in the survey through telephone contact, when it was booked the first personal contact. This was followed by the presentation of research objectives with subsequent signing of the Informed Consent by those who agreed to participate. Data collection was made by primary source.
Participants visited three stations to respond the questionnaires through face to face interview in a private room. At each station there was an However, its application to evaluate agreement among tools could hide biases and disagreements between them 13 . Accordingly, it should be used more appropriate statistical techniques to assess the agreement between assessment tools. However, it has not been explored by the authors reviewed.
Taking into consideration that the instruments in question (NHP and SF-36) have implications for HRQoL evaluation, it is clear that health professionals' need to have full understanding of them as evaluation tools, which makes it imperative to clarify the presented question. Aiming to fill this scientific gap, it was developed the current research: an exploratory study to assess whether the social domains of NHP and SF-36 correspond when applied in a population with chronic disease.
interviewer responsible for applying a questionnaire: the socio-demographic questionnaire, NHP and SF-36. Each interviewer was responsible for applying the same questionnaire during the whole period of data collection in order to avoid collection bias. The interview was conducted by a previously trained staff to read the questionnaires how printed, without changing the words and without further explanation. For each questionnaire application a time of approximately 10 minutes was necessary, in accordance with the literature 5,7 .
The socio-demographic questionnaire consisted of a categorical variable about sex (male or female); age in years; color of skin categorized in white, red, black or brown; socio-economic class categorized as A1, A2, B1, B2, C1, C2 and D according to ABEP (Associação Brasileira de Instituto de Pesquisa de Mercado); schooling was classified as below and above eight years of study.
Using data from SF-36 and NHP, it was possible to get the domain scores of these scales, including social isolation (NHP) and social functioning (SF-36). The NHP is a generic tool to assess HRQoL, originally developed for patients with some chronic involvement. In order to address the various aspects of HRQoL, the instrument investigates six different areas: Energy (EN/NHP), Pain (P/NHP), Emotional Reactions (ER/NHP), Sleep (SL/NHP), Social Isolation (SI/NHP) and Physical Mobility (PM/NHP). NHP is composed of 38 dichotomous questions, to which a point is scored for each affirmative answer. Thus 38 points can be scored, which indicates a worse HRQoL condition compared to the individual who scores zero. The same way of calculating can be applied to the domains, which helps identify areas in which the individual is less or more committed.5 For comparison with SF-36, scores can be transformed into a score ranging from 0 to 100, which is often adopted. Emotional Limitation (EL/SF-36) and Mental Health (MH/SF-36). A mathematical formula proposed by SF-36 creators allows computing participants' scores. The total score relates to the individual's HRQoL. However, the calculation can be performed individually per domain. Score ranges from 0 (worst score) to 100 (best score) 7 . Both questionnaires should preferably be selfadministered 5,7 . The application by interview is susceptible to collecting bias, especially when the questions are too long or subjective. However faceto-face interview is acceptable in case of illiterate or functionally illiterate individuals, provided that all care is taken for the interviewer does not suggest the respondent answer. In this research, it was decided to apply the questionnaires to all patients through interviews to better standardization.
For the sample calculation 0.70 correlation between social subscales was used. This value is the minimum considered as a good correlation in the studies reviewed 8,[14][15][16][17][18][19] . Thereby using G * Power 3.1.9.2 software, alpha of 5%, power of 80% and 0.70 correlation, it was calculated a sample of 84 subjects. From the service records, with 456 registered patients, 97 participants were randomly selected using a table of random numbers. If the individual contacted was not found or did not want to participate, the next table number was included in the list of participants until the sample reached the estimated size.
Data were analyzed using SPSS (17.0). The sociodemographic variables were treated in absolute numbers and percentages or average and standard deviation, since it has little transgressed the normal distribution of frequency. Adopting alpha 5% and 80% power for all analyzes, the Pearson correlation coefficient of SF-36 and NHP domains was calculated in order to investigate whether the social areas of NHP and SF-36 correlated. The Cronbach's alpha of the two scales social domains was also calculated to assess the internal consistency. It was made the ROC curve of Social Isolation (NHP) taking as reference line Social Functioning (SF-36); and it was calculated the Area Under the Curve (AUC) in order to verify the sensitivity and specificity of a domain relative to each other. Subsequently, the Youden index was calculated to estimate the sensitivity and specificity of Social Functioning (SF-36) domain cutoffs.
The Youden index indicates the lower total proportion of possible misclassification, it is the cut-off point with the lowest number of incorrect diagnoses (false positives added to false negatives). The index ranges from -1 to + 1. The calculation can be performed manually:

RESULTS
The sample consisted of all patients registered in a Reference Clinic for Treatment of Collagen, in the city of Salvador, Bahia, Brazil, who met the inclusion criteria. There were no refusals and only three patients were not found from the chart data. As a result, data were collected from 97 patients with confirmed diagnosis of RA. The sample was characterized by women (92.8%), mean age of 52.5 years old (SD±11), brow skin (47.4%), C2 socio-economic class (39.2%), having studied for eight years or more (64.9%) ( Table 1). There were no missing data.
Afterwards, it was built Bland-Altman plot 13 to assess agreement between the two instruments. For this evaluation, the scores of SI/NHP were reversed, so that the social domains of the SF-36 and NHP scored zero to the participant with the worst possible performance in sub-scale and 100 to the individual with the best possible performance. The study followed the definitions of Resolution number 466/12 of the Brazilian National Council of Health for Research in Humans, guaranteeing anonymity, non-maleficence and beneficence to participants. The biggest benefit generated by the project is the best knowledge of this population profile for future interventions for their wellbeing. In addition, the current research allows better knowledge of the available tools to evaluate the social aspect of HRQoL. Furthemore, it was offered as a directly and immediately benefit to the volunteers, a lecture given about their disease, RA, as well as the daily care needed to maintain a good quality of life. Regarding the psychometric evaluation of tested tools, the assessment of internal consistency by Cronbach's alpha, SF/SF-36 domain showed α=0.50, while SI/ NHP hit α=0.73 (Table 2). When calculating the Pearson correlation to verify the convergent validity between social isolation dimension of NHP and social functioning of SF-36, it revealed a significant positive correlation (r=0.305; p=0.006), opposite of expected. Furthermore It was found a higher correlation between SI/NHP and MH/SF-36 (r=-0.433, p<0.001), this time a negative value, as expected. The results of correlations between all areas of NHP and SF-36, especially the theoretically expected correlations, are presented in Table 3.
The ROC curve of Social Isolation (NHP) compared to Social Functioning domain (SF-36) showed AUC of 0.416 (SE=0.8) (p-value=0.28) (Figure 1), indicating low sensitivity and specificity of a domain to each other. When checking the Youden index, It was noted that most of the scores showed low sensitivity and low specificity rating close to zero or negative, indicating that the tolls do not discriminate well each other (table 4).    Bland-Altman plot, comparing NHP and SF-36 social domains, showed mean difference equal to 29.37 (±38.51), and LOA from -47.65 to 106.39. Three outliers were found (Figure 2). The NHP and SF-36 instruments are designed to evaluate the HRQoL and have been treated in the scientific literature as comparable, replaceable one another. However, the current study demonstrated that NHP and SF-36 social domains do not match when applied to a population with RA accompanied in a reference unit for the treatment of collagen in the city of Salvador, Bahia, between the months October 2011 to July 2012.

DISCUSSION
The first psychometric property tested to answer the research question, was internal consistency. It was necessary to know whether the instruments assessed what each of them intended to, before checking if they evaluated the same thing. In this exploratory study SI/NHP had good internal consistency, which corroborates previously published studies 8,10,14,16 .
However, SF/SF-36 showed moderate internal consistency, which was also observed by some authors who investigated hemodialysis patients8 or ischemia of the lower limbs 16

CONCLUSION
The investigation of the assumed comparability of NHP and SF-36 social domains, by a quantitative method of analysis of questionnaires applied to a population with RA from a reference unit for the treatment of collagen in the city of Salvador, Bahia, from October 2011 to July 2012 and subsequent use of the Pearson correlation coefficient tests, Youden index, ROC curve, AUC and Bland-Altman plot, indicated no agreement of the two investigated domains.