Multilevel analysis of self-perception in oral health and associated factors in Southern Brazilian adults : a cross-sectional study

The aim of this study was to evaluate the association between individual and contextual variables related to self-perception in oral health among residents in the municipality of São Leopoldo, Rio Grande do Sul State, Brazil. The cross-sectional design involved 1,100 adults in 38 census tracts. The self-perception was evaluated using the Oral Health Impact Profile (OHIP-14) tool. A logistic multilevel analysis was performed. The multivariate analysis revealed that those who are of the female gender, older, with lower scores of quality of life and less social support, with poor healthy eating habits, smokers and those living in low-income census tracts presented higher odds of reporting worse oral health self-perception (OHIP-1). We concluded that individual and contextual variables are associated with oral health self-perception. This is essential information for planning health services wishing to meet the health needs of the population. Multilevel Analysis; Oral Health; Quality of Life ARTIGO ARTICLE


Introduction
Self-perception of oral health conditions related to people's quality of life presumably incorporates distinct degrees of subjectivity on the part of the subjects involved.In order to understand the distinct implications relating to subjectivity, such as the perception of social inequities and their effects on one's health or illness, there is a need to go beyond clinical indicators of morbidities and to include multidimensional aspects of health 1,2,3,4 .With regard to general health in the Brazilian population, research has been conducted in the quest for more solid explanations about one of the aspects of this issue, i.e. the possible effects of individual and contextual variables on diverse health outcomes 5,6,7,8 .Recent contributions provide evidence of the impact of social, economic, psychosocial and other factors on health inequities 9,10,11,12 .
Analyses have taken into account macro and micro considerations, and these relations and implications have also been investigated in the area of oral health 13,14,15 .The association between oral health problems and socioeconomic and geographical inequities suggests that negative determinants, acting at the individual level and at the level of people's lives, are associated with a less favourable oral health profile, although the relative weights of these determinant variables and the more precise nature of these associations may not have been fully elucidated 16,17 .
It is a challenge for scientific knowledge to determine the impact that oral health problems have on people's everyday lives, including consequences on the quality of their lives 2 .This, therefore, also has implications for the evaluation of oral health needs and the organization of dentistry policies, programmes and services in general 18 .
Furthermore, there is evidence that higher rates of oral diseases and complaints are related to worse social inequity indicators, thus increasing the impact of the chronic negative factor on the social determinants of health 13,19 .The findings of these studies generate new hypotheses that seek to establish how the relationship between subjective aspects and quality of life occurs, in association with the general and oral health condition of individuals and populations, in view of the lack of consensus on this issue 20,21 .
The scientific debate on the concept of quality of life, as well as how it is put into practice in studies, frequently refers to its multidimensional complexity, so that it almost becomes an abstraction.The events relating to the term are predominantly subjective and, moreover, the way in which individuals perceive their quality of life can vary according to their social, cultural and political context 17,22 .The same can be observed in individuals' perception of their general state of health, since individual characteristics and the characteristics of the context in which they live appear to bear influence 23 .
The psychometric instrument referred to as the Oral Health Impact Profile (OHIP) was developed in order to evaluate the impact of the oral health condition of individuals on their quality of life.Originally with 49 questions, the profile was reduced to a 14-question version called OHIP-14 24 , which has proved to be equally robust in methodological terms 25 .
Studies of greater complexity in Brazil, using multilevel analysis, have proposed the elucidation of hypotheses of association between contextual and individual factors in subjects' perception of their general 23 and oral health 26 .Their authors have compared the different geographical structures of data (administrative regions, neighbourhoods or census tracts) with the purpose of verifying the effects of individual and contextual variables on the odds of individuals having a negative or positive perception.Both suggest that there are variations in the perception of general or oral health which are dependent on aspects relating to individuals and their contexts.
Aware that individual and contextual factors may be associated with oral health outcomes with consequent implications for quality of life, this study aimed at testing in a multilevel analysis the existence of such association in individuals living in São Leopoldo, Rio Grande do Sul State, Brazil.

Methodology
This multilevel cross-sectional study was conducted in 2006-2007  The sample for this cross-sectional study was obtained based both on a pilot study and using customary statistical parameters.To compensate losses and to control for confounders, an increase of 20% in the sample size for the number of households was done.As a result, 1,512 households from 40 census tracts were randomly selected, in a strategy of probability proportional to size, thus we obtained a self weighted sample.A total of 38 census tracts were effectively reached, and the final sample was made up of 1,100 subjects.Only adults over the age of 18 were included.During the visits, if the respondent was absent, the interviewers returned three more times in order to attempt to collect the data 23 .
The respondents answered a structured, standardized and pretested questionnaire that included all the individual variables used.
Poor self-reported oral health was assessed using the OHIP-14.The OHIP provides a comprehensive measure of self-reported dysfunction, discomfort and disability arising from oral conditions.It is an adaptation of the World Health Organisation's classification of impairments, disabilities and handicaps 27 .A shortened version (OHIP-14) was later developed based on 14 items 24 .Based on a guiding question, it asks how frequently respondents experience problems related to taste, speaking, chewing, eating, relaxing, or socializing due to their teeth, mouth or dentures.Possible replies were: never, rarely, sometimes, frequently and always.The outcome OHIP-1 represented those who replied "frequently" or "always" at least once to any of the items , whilst OHIP-0 was attributed to the remainder, that is, none of items were perceived in the worst response categories 28,29 .
The individual explanatory variables were divided into categories: gender and ethnicity (skin color), dichotomized into male/female and white/non-white, respectively.Regarding ethnicity, in the questionnaire the possible selfreported skin colors were yellow, white, black and light-skinned black, as per categories used by the Brazilian Institute of Geography and Statistics (IBGE).After the initial explanation of this variable, considering the rate of each category, the dichotomization was adopted as described earlier.
Age was stratified into 10-year intervals.Family income (expressed in the currency Brazilian Reals), was categorized into: high (the highest 25% of scores), medium (the 50% of intermediary scores) and low (the lowest 25% of scores).The psychosocial variables of quality of life and social support also followed the same categorization.Data on quality of life was obtained by questions to obtain information about personal satisfaction 30 .The answers varied from "very satisfied" to "very unsatisfied", ranging from 0 to 4. The minimum and maximum scores obtained by adding the points for each answer are 0 and 32, respectively.Higher scores indicate higher quality of life.The social support followed the lines proposed by the Medical Outcomes Study (MOS) 31 and was collected by means of 19 questions comprising material, affective, emotional, positive social interaction and information factors.The answers range from 0 to 4 and the total possible score can range between 0 and 76.Higher scores mean a higher level of reported social support.
The contextual explanatory variable, i.e. those relating to the census tract in which the sample subject lived, were grouped into sociodemographic and infrastructure variables, these being collected based on the IBGE demographic census available at the time the study was prepared (2000) (IBGE.Cidades@.http://www.ibge.gov.br/cidadesat/topwindow.htm?, accessed on 17/Aug/2008).The demographic variables at this level were: residents (or population resident in the census tract), average monthly income of the heads of the permanent private households and the average number of years of schooling of the heads of the permanent private households.Infrastructure was evaluated based on the following information: permanent private households with garbage collection, bathroom or toilet and sewage collection via general sewer or rainwater networks.These variables were then categorized into: high (the highest 25% of scores within the census tracts), medium (the 50% of intermediary scores) and low (the lowest 25% of scores).
For the purposes of the statistical analysis, the variables of the original database obtained through the application of the questionnaire in the cross-sectional study were included in the IBGE secondary database mentioned above.
The chi-square test (p < 0.05) was used in the bivariate analysis in order to identify significant associations between the explanatory variables and their ability to predict the outcome of interest, i.e.OHIP.Analyses at this stage were performed using SPSS version 15.0 (SPSS Inc., Chicago, USA).
Multilevel logistic regression with a second order predictive quasi-likelihood (PQL) approach was used, with the aim of estimating the odds ratios and their 95% confidence intervals (95%CI).The adoption of this model was primarily based on the response variable used in this manuscript that is dichotomous.
Two levels were incorporated into the model.The first level included the context variables (variables attributed to the census tract) whilst the second level included the individual variables.Initially the null model was adjusted.Fol-lowing this the model in which the census tract variables were included was adjusted.Census tract variables having significance in Model 1 were included in Model 2, as well as the individual variables.Finally, Model 3 was comprised of the variables found to be significant in Model 2. This analysis was performed using MLwiN 2.24 (Centre for Multilevel Modelling, Bristol, UK).A 5% significance level was used to detect associations between outcome and exposure.

Results
71.8% of the participating individuals were women and 83.8% were white.Average age was 44.2 years (SD = 15.8), the maximum age being 90 and the minimum 18.Average time of schooling was 8.2 years (SD = 4.1).The prevalence of the worst classification (OHIP-1) was approximately 16%.
With regard to the contextual characteristics, the average census tract population was 746 inhabitants (SD = 358), whilst average income was R$ 955.00 (SD = 586) and average length of schooling was 8 years (SD = 3).The average frequency of households with garbage collection and which were also connected to the sewerage network was 41%.
Under the bivariate analysis, the odds of individuals of the female gender having the worst oral health condition classification was 2.5 times higher (OR = 2.53; 95%CI: 1.63-3.93)than for males.Similarly, older individuals showed the same behaviour, with greater impacts in those aged 40 and over (p < 0.001).Low levels of schooling also showed themselves to be a variable favouring the perception of bad oral health conditions (OR = 2.73; 95%CI: 1.48-5.03).The psychosocial, quality of life and social support variables also had significant association with the outcome studied, whereby individuals who reported lower scores for these conditions were more subject to a worse oral health classification.Similarly, those who smoked also provided worse reports (OR = 2.22; 95%CI: 1.53-3.21)(Table 1).
At this stage of the analysis, no significant differences were found in relation to the individual variables of skin color, healthy eating habits and alcohol consumption, nor as to the census tract variables of schooling and garbage collection/ connection to the sewerage network (Tables 1 and 2).
Under the multivariate analysis, the null model indicated that the census tracts differ in relation to the probability of the outcome being classified as OHIP-1 (worst condition).In Model 1, in which the variables were adjusted by each other, low income in the tracts was associated with the outcome (OR = 0.26; 95%CI: 0.07-0.96).When this variable was adjusted for the individual variables (Model 2) statistical significance was maintained (OR = 0.23; 95%CI: 0.09-0.60).In the final model (Model 3) the variable maintained this behaviour, whereby the odds of the OHIP-1 condition was 0.24 higher among those living in low-income census tracts than among those living in moderate to high-income census tracts.The gender, age, quality of life, social support and smoking habit variables also presented statistical significance in this model.The individual healthy eating habit variable was significant in Models 2 and 3, having odds of 0.51 (95%CI: 0.31-0.84) in the latter model (Table 3).

Discussion
This study investigated the association between individual and contextual variables and oral health self-perception in adults.The instrument chosen for the study, OHIP, is referred to as a relevant oral health measure since it considers the social consequences of oral health problems that are important to individuals 2 .The literature also reveals consistency as to people's self-reported oral health alterations 33 .
The results show that the worst reports of oral health conditions were made by individuals of the female gender, older people, those with less schooling, lower quality of life and social support scores, as well as smokers.
With regard to gender, the findings are consistent with the literature showing higher poor oral health perception in women compared to men 17,34 .
With regard to increased age, this variable was inversely associated with oral health perception in all stages of analysis, i.e. increased age is associated with a worse perception of oral health whereby oral disorders are more prevalent 34,35 .Despite only having been significantly associated with the outcome in the bivariate analysis, family income may favour worse perceptions 36,37 .The loss of this variable's explanatory capacity may have occurred as a result of other variables being more determinant; e.g.gender, which may serve as a proxy for material conditions, i.e. gender in some way reflecting socioeconomic inequities, which the income variable on its own is not able to capture.In relation to the psychosocial variables, quality of life considered to be less favourable leads to a worse perception of oral health 38,39 .
Consistent conclusions are not found in the literature on this issue as it is a factor characterized by multidimensional complexity, as well as being Note: bold values are statistically significant (p < 0.05).
Table 3 Multilevel analysis for poor self oral health perception Oral Health Impact Profile (OHIP-1) according to individual and contextual variables (N = 1,100).of a subjective nature and liable to the action of the social, cultural and political context in which individuals find themselves 17,22 .Regarding social support, there is evidence of the effects generated by the links between people providing a favourable psychosocial environment, culminating in positive or negative behaviours or choices, influencing the absence or presence of diseases in adult life 8,9,40 .Deleterious habits, such as smoking, consuming alcohol or using other drugs, were associated with worse oral health conditions relating to quality of life 41 .In this study impacts were most reported by smokers.On the other hand, in the multilevel analysis healthy eating habits had significant association with the outcome and this may have resulted in the majority of the individuals being concentrated in the moderate consumption category.

Variables
The adjustments made during the multilevel analysis for the individual variables revealed that only the census tract income variable was associated with higher odds of reporting the worse oral health condition (OHIP-1).Other reports, aimed at verifying the effects of contextual variables on the odds of the individuals' negative perception of oral health, with control of individual dimension factors, have concluded that individual socio-economic attributes have greater explanatory power as to the variation in perceived oral health than the socioeconomic contexts of the areas of residence, when comparing the different geographic structures of the data, such as administrative regions, neighbourhoods and census tracts 26,39 .
Another line of study provides evidence of the existence of an association between the environment in which individuals live and their exposure to greater risks 10 .A study with the same database used in the present study, but with the outcome of interest geared to the perception of general health conditions, revealed a high prevalence of worse classification in places with larger populations and lower levels of income and schooling, confirming the influence of the living environment on people's health 23 .
Brazilian studies also indicate the influence of the context in which people live on their health conditions.Places with less favourable conditions appear to influence the way in which individuals perceive their health, in keeping with the findings of this study.Census tracts with smaller populations and lower income show a greater perception of bad oral health conditions 5,6,7 .
The bivariate analysis in this study found that a larger population was associated with a worse perception of oral health.Studies indicate that this variable may bear influence since a large number of individuals, concentrated in the same census tract, suffer from similar problems and this produces effects on their perception of their health 42,43 .
The option for the method of using the census tracts as units of analysis is a useful resource for elucidating the influence of the structural and surrounding environmental characteristics on the outcome to be studied.This strategy, referred to in other studies as "neighbourhood", has been used in other research related to health conditions 39,42,43,44,45 .
The limitations of this study may be related to the design adopted and the non-inclusion of clinical variables.The former is a limiting factor in that cross-sectional designs do not express with a high degree of certainty whether or not the relationship found is a causal one.The latter is due to the fact of the possible association between clinical conditions and the variation in the perception of impact, including on quality of life, and the possibility of the sociodemographic and socio-economic factors occurring in an independent manner 46 .The small number of individuals included in the study might be another explanation for the lack of significant association between contextual variables and OHIP.
It is appropriate to clarify that in models using a hybrid structure, including data aggregated in ecological designs, there is an obvious requirement for the maximum reduction of errors to compensate for design weaknesses, as well as the choice of an adequate model that enables sufficient variable capture in the ecological exposure of interest 47 .The key concern of these methods has to be the minimization of aggregation biases capable of producing so-called "ecological fallacies".
There are recommendations in the literature for care to be taken in inferring equally to all individuals' specific conditions found in particular sectors studied 48 .In this study this does not appear to be a problem, since the census tract data are very close to the overall data for the municipality of São Leopoldo, thus permitting the results to be inferred for the local population 23 .
The multilevel analysis technique was adopted since, currently, in studies involving public health it has shown itself to be a powerful resource for helping to discern the influence of social determinants and individual variables on the health-illness process 49,50 .As such it is possible to find explanations for the impacts generated not only by individual characteristics but also by the conditions of the context in which individuals are located.This technique systematically reveals how the different levels are associated with the outcome of interest 49 .

Conclusion
The results indicate that individual and contextual variables are associated with self-perceptions of oral health.This may be important to help the development and planning of oral health care services aimed at addressing population demands and to reduce oral health inequalities.
Análisis Multinivel; Salud Bucal; Calidad de Vida Contributors M. C. L. Gabardo contributed towards the conception of the study design, interpretation of the results and writing.S. J. Moysés participated in the conception of the study design, interpretation of the results, writing and final approval of the article.S. T. Moysés contributed towards the critical review of the article.M. Olandoski and M. T. A. Olinto collaborated on the data analysis and interpretation of the results.M. P. Pattussi contributed to the critical review and final approval of the article.

Table 1
Oral Health Impact Profile (OHIP-1) distribution, prevalence and unadjusted and adjusted odds ratio (OR) according to individual variables (N = 1,100).

Table 2
Oral Health Impact Profile (OHIP-1) distribution, prevalence and unadjusted and adjusted odds ratio (OR) according to contextual variables (N = 1,100).