Tuberculosis surveillance and health information system in Brazil , 2001-2003

RESULTADOS: Em cerca de 8% dos municípios brasileiros, mais do que 10% dos casos de tuberculose em moradores de um município foram notifi cados por serviços de saúde de outro município. Em aproximadamente 950 municípios, a freqüência de pacientes com tuberculose pulmonar que realizaram baciloscopia foi inferior a 90%. Amazonas, Pará e Amapá tinham mais municípios classifi cados nos grupos de baixa qualidade de vigilância. Pernambuco, Ceará e Bahia, no Nordeste, tinham menor nível de qualidade da vigilância. As regiões Sul e Centro-Oeste tinham maior número de municípios com melhores níveis de vigilância da tuberculose.


INTRODUCTION
The purpose of tuberculosis (TB) surveillance is to be aware of the cases of the disease in the population, so that measures to stop transmission to susceptible individuals can be taken.However, cases diagnosed and notifi ed by the health service may represent only a proportion of the TB cases. 1 Estimating the number of TB cases in a certain population has been a challenge to epidemiologists and planners on TB control.*Current indirect methods of estimation depend on the accuracy on other TB data, such as the number of people infected by Mycobacterium tuberculosis, and the number of deaths.These data depend on the good work of health services.** ,*** Studies on the prevalence of TB cases are rare, complex in methodology and, therefore, not very often performed.In this context, methodological approaches to assess the actions on TB surveillance are welcome because they can identify situations in which data on TB notifi cation are different from its actual incidence.
Even though all sick patients may be acknowledged by the health services, notifi cation of these cases by the information system may not occur or the record may be incomplete.Thus, even with the diagnoses of the TB cases, the notifi cation in the information system may be poor, preventing to know the incidence of the disease. 4Therefore, the diagnosis ability and the complete notifi cation of TB cases are necessary so that the data on epidemiological surveillance are accurate.Likewise, data on notifi cation records must be good quality.
In Brazil, assessment of the actions to control TB is encouraged because the proportion of cure for the disease is below the goal of 85%.*Good quality of information is essential to assess the program, since it enables to see the magnitude of the disease transmission in the country, and to defi ne the goal of how many cases will be treated.Poor quality of information does not enable to see whether actions have been taken or if the data have not been suitably entered onto the information system.Additionally, the quality of information may refl ect on the disease surveillance.In these situations, it is not possible to clearly defi ne if the problem of controlling TB lies on a poor surveillance, or if the problem is partial or total lack of data.
Regarding epidemiologic surveillance in itself, it is important to distinguish between two types of data.The fi rst type concerns notifi cation of the disease, that enable to assess if districts can detect and notify cases; and the second is the follow-up data of the district from notifi cation until the patient is removed from the system.The present study did not aim at assessing the problems of TB underreporting, but rather the epidemiologic surveillance of TB in Brazil.The indicators assessed were based on the quality of detection and follow-up of cases in the region, the quality of information given by the information system of the cases, and the burden of morbidity.

METHODS
The ecological approach was used to assess the characteristics and quality of TB surveillance in the Brazilian districts, especially in priority ones.Apart from double notifi cations, all information was obtained in the data bank of Sistema de Informação de Agravos de Notifi cação (Sinan -Information System for Notifi able Diseases) from 2001 to 2003, made available in March 2005.**Measures that could best qualify these cities were selected, resulting in synthetic indicators.
Spatial smoothing of the rates was used to minimize the infl uence of underreporting.This procedure was to estimate built spatial averages, using as a neighboring criteria the nearness among the districts as geographical unities. 3,4,*** Specifi c indicators for each of the three assessment dimensions of surveillance were: quality of detection and follow-up of cases (surveillance -ability to get cases and grant adherence to treatment); quality of treatment information (completeness of data of the system of information and duplicate reporting) and burden of morbidity (incidence rate).This last dimension was included in the assessment of TB surveillance because it considered that the magnitude of the disease (the intensity of its expression) may also infl uence its recognition by the health services.The quality of surveillance is believed to have different repercussion in regions with high and low number of TB notifi cations.
Initially, eight indicators were established for quality of detection and follow-up of cases.Three of them were excluded, either because they were repeated information, or because of the low distinction among cities: proportion of pulmonary cases whose sputum shows bacilli on smear in the second month; proportion of smear positive pulmonary cases that underwent HIV test, and proportion of smear positive pulmonary cases with short term treatment information directly observed, called DOTS (Directly Observed Treatment Short Course).Indicators of detection and follow-up of selected cases are listed in Table 1.As the TB surveillance systems are managed by the districts' authorities, it is hoped that, in the ideal situation, sick dwellers from one city are diagnosed and notifi ed by the health service of their own city.
To assess the burden of information, 16 indicators were studied.An initial assessment of the data indicated that seven of them did not distinguish the districts satisfactorily and were excluded: proportion of cases with not informed situation when starting treatment (if it was a new case); cases with no information regarding the clinical form; extra pulmonary TB cases with no information regarding the clinical form; regarding the performance of sputum examination, or serum examination for HIV; with no information on the date the treatment started.All districts had similar proportions of absence of information on control sputum examination in the fourth and sixth month, and the indicator of the outcome in the end was maintained, being more informative.Thus, the indicators of quality of information that remained were those of proportion of cases without information on: treatment surveillance, control sputum examination in the second, and outcome sputum examination when the treatment ended.
Based on the initial assessment, six indicators that recognized duplicate reporting were obtained from correlation of records (linkage).Three of them were excluded because they presented high collinearity with the others: real duplicate reporting proportion; proportion of duplicate reporting due to lack of information, and proportion of duplicate reporting with no explanation.Completeness indicators which discriminated well the situation of duplicate reporting are presented in Table 2. Four indicators of morbidity were structured, presented on Table 3. Incidence rate of pulmonary TB was excluded because it was considered little discriminatory information.
Assessment of indicators considers the existence of minimum conditions of quality of surveillance, represented by the cut-off points presented on Table 4.
According to this table, districts that met the criteria of at least three of the indicators of quality of detection, and at least four criteria of the indicators of quality of information, were considered as presenting "good quality of surveillance".Districts with at least two conditions of the indicators of burden of the expressed disease were considered as presenting "concerning epidemiological TB situation".Cities were classifi ed according to the quality of epidemiological surveillance of the disease, compared and grouped according to federal unity and geographical region, resulting in four groups of districts: • Group I (GI) -regular TB epidemiological situation (moderate to low incidence) and good quality of surveillance.
• Group II (GII) -concerning TB epidemiological situation (high incidence) and good quality of surveillance.
• Group III (GIII) -concerning TB epidemiological situation and poor quality of surveillance.
• Group IV (GIV) -regular TB epidemiological situation and poor quality of surveillance.Dispersion charts have been made to present the correlation between the incidence and the quality of TB surveillance.Incidence was represented by the quality of detection and follow-up of cases (surveillance -ability to capture cases and assure adherence to treatment).Quality of information was represented by completeness of data placed on the information system and duplicate reporting.

RESULTS
Regarding patients seeking health services in other districts, over 10% of the reported cases of residents from 316 districts (7.5%) were notifi ed by other city in 2001.This percentage was kept in the following years, suggesting a poor quality of TB surveillance (Figure 1).
Figure 2 shows the classifi cation of districts regarding the proportion of new pulmonary TB cases that did not undergo sputum examination.Based only on the examination of the data bank, it was impossible to distinguish when examinations were not performed, and when the record was not placed on the report form.Indirectly, this measure refl ects the ability to capture new cases, since if only a small proportion of pulmonary TB cases undergo sputum examination, the examination of respiratory symptoms may not be properly conducted.Figure 2 points up that many districts presented poor percentage of patients under- Figure 3 shows the proportion of pulmonary TB cases with positive smear and information about the end of the The assumption is that this increasing frequency of districts reveals the poor capacity of following-up patients undergoing treatment in the health unities.
A general assessment of all aspects of quality of surveillance enables a summary of TB control in Brazil, presented in Table 5.In this table the number of districts is classifi ed into the four groups, according to levels of surveillance and TB severity.The frequency of districts classifi ed in group II and IV (poorer quality of surveillance) increased from 2001 to 2003.At the same time, the number of cities with good surveillance and regular level of incidence decreased.
The trends were different among the regions: while North, Northeast, and Southeast worsened, the South improved and the Midwest kept similar conditions in the period.
The States of Amazonas, Pará, and Amapá presented more districts in the group of poor quality of surveillance (groups III and IV) than the other states in the North.Lower frequency of districts with low disease burden and good surveillance was observed in the Amazonas, with high incidence rate.The states of this region presented the worst conditions of TB control with little more than half (62% in 2001 and 53% in 2003) of the districts in group I.The state of Tocantins presented the best surveillance performance and a reduced number of cases (Table 6).
In the Northeast, the states with greater disease burden were also those which presented poorer quality of surveillance: Pernambuco, Ceará, Bahia, Maranhão, and Rio Grande do Norte.On the other hand, Sergipe presented most of the districts classifi ed as group I, with good level of surveillance and regular incidence rate (Table 7).
The States of the Southeast region presented decrease in the capacity of surveillance and probably in TB control.
In Rio de Janeiro, high incidence rates of the disease with low capacity to capture cases and poor quality of data reporting and follow-up of reported cases.The State of Espírito Santo was the one that reduced less its quality of TB surveillance in the Southeast (Table 8).
In the South a clear trend of worsening TB surveillance was not seen.The State of Rio Grande do Sul presented the less suitable conditions, and Santa Catarina presented the best situation (Table 9).
Similar to the South, the Midwest also has states with good levels of TB surveillance.However, the regional levels of the indicators were infl uenced by Mato Grosso Classifi cation (or stratifi cation) of the districts enabled to reveal those which should be more carefully assessed.The districts with low TB notifi cation and evidence of poor quality of surveillance may have high disease burden and still be "quiet", preventing being acknowledge as a priority.
Studies on the assessment of TB surveillance in Brazil that could be compared with our outcomes have not been found.However, some assumptions have been made from the assessment of the epidemiological behavior of TB.The fi rst is: if TB is determined by poor socioeconomic conditions, it is expected that the low level of surveillance would be an explanation for the absence of cases reported in some Brazilian districts.Another assumption is that, if the spatial distribution of TB is according to the distribution of its determinants in the space, the existence of places free from the disease in areas where a high rate is expected is not justifi ed.The low level of surveillance justifi es this fact.
The approach used in the present study is an attempt to know the level of TB control in the districts, but some limitations must be considered.The fi rst is the use of data available in the system of TB information, rather than a fi eld investigation that included the observation of local conditions of TB surveillance in the districts.As previously mentioned, data were not extracted from charts that could distinguish failure to perform surveillance procedures, from failure to record on report forms, and follow-up.The second limitation refers to the arbitrariness of the cut-off points used in the criteria to classify districts.The third refers to the complete underreporting, that is, absence of report, making it impossible to calculate the quality of data indicators.
In this case, the absence of problems with the quality of these data was assumed.Last, the use of spatial smoothing tried to minimize severe underreporting, enabling to recognize high and low rate situations, and different levels of quality of epidemiological TB surveillance.Although the classifi cation of the district does not change when spatially smoothened rates were used, evidences of spatial dependence of TB suggest that some level of underreporting has been corrected.Consistency of the outcomes may be confi rmed by the agreement between the levels of capture of cases and the quality of data in the studied districts, these aspects formed the assessment of the quality of surveillance.
To conclude with, the outcomes of the present study enabled to stratify the Brazilian districts according to their quality of surveillance.This classifi cation may help building control strategies in areas that have not been seen as priority, such as the improvement of surveillance actions.
of pulmonary cases that underwent sputum examination TB incidence rate (all forms), 2002 TB incidence rate forms), 2003 Proportion of pulmonary cases that underwent sputum examination Proportion of pulmonary cases that underwent sputum examination

Figure 3 .
Figure 3. Classifi cation of districts, according to information of ending of pulmonary cases.Brazil, 2001-2003 (A-C).

Table 1 .
Indicators of quality of detection and follow-up of cases.

Table 2 .
Indicators of quality of information (completeness of cases).

Table 3 .
Indicator of the burden of morbidity.
IndicatorNumerator/Denominator Incidence rate of tuberculosis in all forms Number of new cases in the year X 100.000Population of the respective year Incidence rate of smear positive pulmonary tuberculosis Number of new cases with smear positive pulmonary tuberculosis in the year X 100.000Population of the respective year Proportion of HIV positive among new cases Number of new cases with HIV positive test in the year X 100 Number of new cases in the year

Table 4 .
Cut-off points of the indicators of surveillance and magnitude of tuberculosis in Brazil.

Table 5 .
Distribution of districts according to the classifi cation of tuberculosis surveillance and burden of the disease.Brazil,  2001-2003.
GI -districts with regular epidemiological situation of tuberculosis (for example: moderate to low incidence) and good quality of surveillance; GII -districts with concerning epidemiological situation of tuberculosis (for example: high incidence) and good quality of surveillance; GIII -districts with concerning epidemiologic situation of tuberculosis (for example: high incidence) and poor quality of surveillance; GIV -districts with regular epidemiological situation of tuberculosis (for example: moderate to low incidence) and poor quality of surveillance.

Table 6 .
Frequency of districts in the North according to classifi cation of tuberculosis surveillance and burden of the disease.2001-2003.

Table 7 .
Frequency of districts in the Northeast according to classifi cation of tuberculosis surveillance and burden of the disease.2001-2003.

Table 8 .
Frequency of districts in the Southeast according to the classifi cation of tuberculosis surveillance and burden of the disease.2001-2003.

Table 9 .
Frequency of districts in the South according to the classifi cation of tuberculosis surveillance and burden of the disease.2001-2003.treatment.It is possible to observe that many districts did not record the conditions in the end of the treatment, in 9.9% of the district in 2001, and another 35% in 2003.

Table 10 .
Frequency of districts in the Midwest according to the classifi cation of tuberculosis surveillance and burden of the disease.2001-2003.doSul.In this state, lower frequencies of districts in group I, and greater frequency of districts in groups III and IV were observed.In Distrito Federal the best levels of surveillance and disease control were observed (Table10).