A multilevel model for the study of breastfeeding determinants in Brazil

The benefits of breastfeeding for the children`s health have been highlighted in many studies. The innovative aspect of the present study lies in its use of a multilevel model, a technique that has rarely been applied to studies on breastfeeding. The data reported were collected from a larger study, the Family Budget Survey-Pesquisa de Orcamentos Familiares, carried out between 2002 and 2003 in Brazil that involved a sample of 48 470 households. A representative national sample of 1477 infants aged 0-6 months was used. The statistical analysis was performed using a multilevel model, with two levels grouped by region. In Brazil, breastfeeding prevalence was 58%. The factors that bore a negative influence on breastfeeding were over four residents living in the same household [odds ratio (OR) = 0.68, 90% confidence interval (CI) = 0.51-0.89] and mothers aged 30 years or more (OR = 0.68, 90% CI = 0.53-0.89). The factors that positively influenced breastfeeding were the following: higher socio-economic levels (OR = 1.37, 90% CI = 1.01-1.88), families with over two infants under 5 years (OR = 1.25, 90% CI = 1.00-1.58) and being a resident in rural areas (OR = 1.25, 90% CI = 1.00-1.58). Although majority of the mothers was aware of the value of maternal milk and breastfed their babies, the prevalence of breastfeeding remains lower than the rate advised by the World Health Organization, and the number of residents living in the same household along with mothers aged 30 years or older were both factors associated with early cessation of infant breastfeeding before 6 months.

Multilevel analysis of hepatitis A infection in children and adolescents: a household survey in the Northeast and Central-west regions of Brazil

Background The objectives were to estimate the prevalence of hepatitis A among children and adolescents from the Northeast and Midwest regions and the Federal District of Brazil and to identify individual-, household- and area-levels factors associated with hepatitis A infection. Methods This population-based survey was conducted in 20042005 and covered individuals aged between 5 and 19 years. A stratified multistage cluster sampling technique with probability proportional to size was used to select 1937 individuals aged between 5 and 19 years living in the Federal capital and in the State capitals of 12 states in the study regions. The sample was stratified according to age (59 and 10- to 19-years-old) and capital within each region. Individual- and household-level data were collected by interview at the home of the individual. Variables related to the area were retrieved from census tract data. The outcome was total antibodies to hepatitis A virus detected using commercial EIA. The age distribution of the susceptible population was estimated using a simple catalytic model. The associations between HAV infection and independent variables were assessed using the odds ratio and corrected for the random design effect and sampling weight. Multilevel analysis was performed by GLLAMM using Stata 9.2. Results The prevalence of hepatitis A infection in the 59 and 1019 age-group was 41.5 and 57.4, respectively for the Northeast, 32.3 and 56.0, respectively for the Midwest and 33.8 and 65.1 for the Federal District. A trend for the prevalence of HAV infection to increase according to age was detected in all sites. By the age of 5, 31.5 of the children had already been infected with HAV in the Northeast region compared with 20.0 in the other sites. By the age of 19 years, seropositivity was 70 in all areas. The curves of susceptible populations differed from one area to another. Multilevel modeling showed that variables relating to different levels of education were associated with HAV infection in all sites. Conclusion The study sites were classified as areas with intermediate endemicity area for hepatitis A infection. Differences in age trends of infection were detected among settings. This multilevel model allowed for quantification of contextual predictors of hepatitis A infection in urban areas.

Prediction in Multilevel Logistic Regression

The purpose of this article is to present a new method to predict the response variable of an observation in a new cluster for a multilevel logistic regression. The central idea is based on the empirical best estimator for the random effect. Two estimation methods for multilevel model are compared: penalized quasi-likelihood and Gauss-Hermite quadrature. The performance measures for the prediction of the probability for a new cluster observation of the multilevel logistic model in comparison with the usual logistic model are examined through simulations and an application.