This cross-sectional study was carried out in MRJ, correlating TB cure with family health coverage and socioeconomic, demographic, and epidemiological variables. The municipality is located in the southeast of the country, with exclusively urban population estimated at 6,476,631 inhabitants in 201514.
The study population is comprised of new cases of tuberculosis residing in the MRJ notified in the period 2012–2014. All cases were eligible to be part of the study to minimize selection bias.
The geo-referencing technique was used from the residence address to identify family’s health coverage, assigning a geographical position for each record (latitude and longitude). The PHC coverage maps were developed by the Municipal Health Secretariat of Rio de Janeiro and correspond to the existing coverage in 2012, 2013 and 2014, with reference to the month of December. Georeferencing of addresses was done using the “Geocode” tool made available by Google Maps through a free Application Programming Interface (API).
This process utilized the Google’s streets and location base. The accuracy of georeferencing can be evaluated from a score ranging from 0 to 10, (0-not found, 1-country level, 2-state, 3-subregion, 4-city, 5-Zip Code, 6-streets, 7-intersection between streets, 8-address, 9-name of the building or trade, 10-maximum precision). Addresses with a score lower than “5” were considered as losses. Records with scores between “8” and “10” were considered with acceptable accuracy. The remaining records were manually reviewed. Of the 14,384 georeferenced records, the precise geographic coordinates of 3,484 records (24.22%) could not be determined, characterizing losses of the georeferenced sample. The total number of georeferenced records was 10,900 new closed cases.
The outcome variable considered in this study was the “TB cure outcome” (yes/no) obtained from the Notifiable Diseases Information System for Tuberculosis (SINAN-TB) and the exposure variable was “PHC coverage”, expressed by the “time (in months) between the implantation of the heath teams and TB diagnosis”.
The selection of variables was based on the theoretical TB cure model based on the dimensions “environment”, “individual factors”, “access to healthcare service” and “social status”. The model includes demographic, social, epidemiological, access and use of health services variables.
The variables selected from the SINAN-TB were: age; gender; race/color (white/non-white); schooling; “HIV coinfection” (yes/no); “alcohol abuse history” (yes/no); “Contact search” (yes/no); “serology for HIV” (positive, negative, not performed); “supervised treatment” (yes/no).
Considering that there is a large number of socioeconomic variables from the 2010 demographic census (IBGE, 2011)15, a multivariate analysis was performed using the main components analysis technique. Thus, the TB cure-related realms (“agglomeration”, “household conditions”, “demographic”, “schooling”, “income”) were established from the theoretical model related to the “social condition” and for each realm, the variables that, to a lesser extent, represent the other variables of that realm were selected.
The socioeconomic and demographic variables resulting from this analysis were: “average monthly income of the person in charge (R$)”; “average number of residents per household”; “population density in the census sector”; “Density of dwellers/rooms”; “proportion of permanent private households with bathrooms for the exclusive use of residents or water closet and sanitary sewage via general sewer or rainwater network”; “Proportion of permanent households with electricity”; “average number of toilets per permanent private residence”; “aging rate”.
The variables obtained from the Demographic Census represent the averages and proportions of each census tract. These values were repeated for each individual resident in the same census tract since these variables in this level of aggregation show high homogeneity. The other variables were analyzed from the individual level.
R software (R Development Core Team, 2016)16 was used for data descriptive analysis through bivariate logistic regression (gross analysis), exploratory analysis, multivariate analysis, kernel estimation and spatial analysis was performed using the R software.
During TB spatial exploratory analysis, we observed the distribution of points and identification of possible clusters through the point density estimation technique, defined as Kernel density estimation, which consists of generating a point density surface within a region of influence, weighted by the distance of each from the location of interest, for the visual identification of “hot areas” on the map. Bandwidths from 500 m to 3,000 m were tested, with 250 m increments. A matrix of 500 × 500 points and a radius of 2,500 m was used because it was considered the most appropriate for highlighting strategic areas. We generated maps with estimates of TB incidence rate through the kernel ratio between reported cases of TB and the kernel of the population.
Spatial analysis was based on the generalized additive model (GAM), which can be considered an extension of generalized linear models, with the inclusion of a non-parametric element by smoothing functions. The model has the great advantage of being more flexible and relatively simple to interpret17. The Thin Plate Regression Splines smoothing technique was used.
The construction of the generalized additive models was performed through manual selection, based on the essential factors of the theoretical framework, not only considering the statistically significant variables in the bivariate analysis, but also those of epidemiological importance.
We did not only consider the p-value of each association, but the importance previously described of each variable and the impact on the explanatory power of the model. Only those variables with a clear negative impact under the explanatory power of the model, observed through deviance, were removed.
The software for constructing family health coverage maps and spatial data queries was ArcGIS, version 10.2.218 in Latlong/WGS84 projection, available in the shapefile extension.
During the analyses, the following additional packages of R were used: car19, mgcv20, descr21, sp22, sdep23, maptools24, splancs25, fields26, RColorBrewer27, ggplot228.