key: cord-0750711-2wb3cmy5
authors: Jesus, Gabriela S; Pescarini, Julia M; Silva, Andrea F; Torrens, Ana; Carvalho, Wellington M; Junior, Elzo P P; Ichihara, Maria Y; Barreto, Mauricio L; Rebouças, Poliana; Macinko, James; Sanchez, Mauro; Rasella, Davide
title: The effect of primary health care on tuberculosis in a nationwide cohort of 7·3 million Brazilian people: a quasi-experimental study
date: 2022-01-24
journal: Lancet Glob Health
DOI: 10.1016/s2214-109x(21)00550-7
sha: 30da1fb965bff9ce2d6ac7c7f1b60caef3ae6b39
doc_id: 750711
cord_uid: 2wb3cmy5

BACKGROUND: Universal health coverage is one of the WHO End TB Strategy priority interventions and could be achieved—particularly in low-income and middle-income countries—through the expansion of primary health care. We evaluated the effects of one of the largest primary health-care programmes in the world, the Brazilian Family Health Strategy (FHS), on tuberculosis morbidity and mortality using a nationwide cohort of 7·3 million individuals over a 10-year study period. METHODS: We analysed individuals who entered the 100 Million Brazilians Cohort during the period Jan 1, 2004, to Dec 31, 2013, and compared residents in municipalities with no FHS coverage with residents in municipalities with full FHS coverage. We used a cohort design with multivariable Poisson regressions, adjusted for all relevant demographic and socioeconomic variables and weighted with inverse probability of treatment weighting, to estimate the effect of FHS on tuberculosis incidence, mortality, cure, and case fatality. We also performed a range of stratifications and sensitivity analyses. FINDINGS: FHS exposure was associated with lower tuberculosis incidence (rate ratio [RR] 0·78, 95% CI 0·72–0·84) and mortality (0·72, 0·55–0·94), and was positively associated with tuberculosis cure rates (1·04, 1·00–1·08). FHS was also associated with a decrease in tuberculosis case-fatality rates, although this was not statistically significant (RR 0·84, 95% CI 0·55–1·30). FHS associations were stronger among the poorest individuals for all the tuberculosis indicators. INTERPRETATION: Community-based primary health care could strongly reduce tuberculosis morbidity and mortality and decrease the unequal distribution of the tuberculosis burden in the most vulnerable populations. During the current marked rise in global poverty due to the COVID-19 pandemic, investments in primary health care could help protect against the expected increases in tuberculosis incidence worldwide and contribute to the attainment of the End TB Strategy goals. FUNDING: TB Modelling and Analysis Consortium (Bill & Melinda Gates Foundation), Wellcome Trust, and Brazilian Ministry of Health. TRANSLATION: For the Portuguese translation of the abstract see Supplementary Materials section.

The Cadastro Único (CadUnico) is a national administrative system that contains information about individuals, and their families, that has selectable characteristics to apply for any social programs in Brazil. 1 To be eligible and register with CadUnico, families must receive an income of up to half a minimum wage per capita or a total family income of up to 3 minimum wages (for example, the minimum wage ranged from R $ 380.00 in 2007 to R $ 724.00 in 2014). At the end of 2017, CadUnico consisted of approximately 114 million individual registrations, representing around 50% of the Brazilian population.

At the time of registration, registrants are assigned a unique numerical identifier and researched for socioeconomic indicators. Families must update their records every two years as long as they remain enrolled in any public program. The CadUnico cohort includes 246 socioeconomic and demographic variables with information from the family and individuals registered in this data collection and storage system. 1

The Cohort of 100 million Brazilians was created with the objective of enabling studies and continuous assessments of social determinants and the effects of social programs and policies on the different health contexts in Brazil. 2 Created by the Center for Data and Knowledge Integration for Health (CIDACS / FIOCRUZ), 3, 4 the cohort is based on information from more than 114 million individuals who were registered during the years 2000 and 2017 in the CadUnico, linked with available health datasets of morbidity and mortality.

In the 100 Million Brazilians Cohort changes of participants' residence are not included and handled in the analysis because information about the date of such changes are not available. However, considering the low rate of internal migration in the country 5 we assume that it has not biased meaningfully our analyses.

In this study, we consider the intervention as the Family Health Strategy (FHS) and the adopted outcomes are: incidence rates, cure rate, case-fatality and mortality rate. New cases of Tuberculosis (TB) are registered in the Notifiable Diseases Information System (SINAN) and the records of deaths as a cause of death TB were collected in the Mortality Information System (SIM).

SINAN includes information on all mandatory reporting diseases in the country. After TB detection, health professionals, both in the public and private sectors, are required to notify the electronic system, reporting the date of clinical detection and demographic characteristics of the patient and updating information about treatment. The SIM provides microdata of deaths by age, sex, cause and municipality of residence of the deceased. The baseline of the cohort constructed by Cidacs / Fiocruz and linked to TB information from SINAN and SIM.

All individuals diagnosed with TB before the beginning of the study period (2004) were excluded from the study cohort. While this could be considered a case of left truncation,1 the inclusion of individuals with a previous event of TB (left censoring) could have biased the analyses, because their tendency to reactivate the TB and become an new incident case would have been greater than individuals without previous TB episodes, and the probability to contract TB before the study period could have been influenced by unknown previous levels of PHC coverage. 6 

The FHS was launched in 1994 and has experienced a dramatic expansion across Brazil's municipalities. FHS is currently one of the world's largest community based primary health care programme, involving interdisciplinary health care teams that include a physician, a nurse, a nurse assistant, and four to six full-time community health agents. Family health teams are organized geographically, covering populations of up to 1000 households, and each FHS team member has defined roles and responsibilities. Moreover, national guidelines help to structure and standardize FHS responses to most health problems. Each community health agent is assigned to approximately 150 households in a geographically delineated micro-area within the catchment areausually the same micro-area where the agent lives. Agents visit each household within their micro-area at least once per month and collect individual-and household-level data. During each visit they develop health promotion activities, also helping scheduling appointments, check whether prescriptions have been filled and whether patients have been taking their medications regularly. They also ask about changes to household composition, identifying signs of violence and neglect, among other problems. They also actively look for risk factors such as smoking and symptoms of common chronic disease such as hypertension and diabetes 7 .

FHS teams are located near people's homes, facilitating the access and first-contact care. Moreover, the lists of all residents in each geographic area permit delivery of longitudinal care, and each team is responsible for everyone in its catchment area. The care provided by the FHS model is proactive, since the community health agents seek out problems before patients arrive at the health post.

In Brazil, TB diagnosis and treatment is fully offered -free of charge -by the national health system (Sistema Unico de Saúde -SUS). The TB program was decentralized to primary health care units in 2004, that include both the FHS and other centres that offer the first level of care. The SUS offers screening for latent TB infection, preventive treatment, and TB treatment and all are available free of charge. Although community wide screening for active TB can take place in certain areas from time to time, its not a national recommendation to be adopted by primary health care in routine.

The annual coverage of the FHS is built from information collected from the Ministry of Health, through the Primary Care Information System (SIAB), the most updated and consolidated data on the coverage of the FHS in all 5,570 Brazilian municipalities in the period 2000 to 2018. Here we will present the data from 2004 to 2013 due to the reduction of the study period discussed in the body of the article. The FHS exposure variable was calculated as the total number of teams deployed in the municipality that year multiplied by 3,450, which represents the number of people served by each FHS team, divided by the population of the municipality. CadUnico's registrations were linked to the coverage of the municipal FHS by the municipality code. A summary table of FHS municipal coverage and national coverage is shown on the Web Table 1 , and a graph of the annual distributions of FHS municipal coverage (box plot) is shown in Web Figure 1 . 

In order to disentangle the causal relationships and detect possible biases due to the choice of the adjusting variables in the logistic and Poisson multivariable regression models, we created a Direct Acyclic Graph (DAG) of the variables selected through the theoretical framework, the dataset availability, and the better prediction of the region of common support for the propensity score (Web Figure 2 ).

The descriptive analysis of the data by population strata is shown in Web Table 2 . The subpopulations are divided into two strata for each variable, and they are described by group exposed to the FHS (<10% and 100%). The income subpopulations were built according to the median per capita family income. Education subpopulations were stratified by individuals who are illiterate or have up to four years of study and individuals who have 5 years or more of schooling. The last two subpopulations were based on sex (male and female) and age (under 15 and 15 years old or more). Primary school or less: ≤5 years of education; junior high school: 6-9 years of education; high school: ≥10 years of education.

For each dataset of the outcome variable, we estimate the probability of the individual being exposed to the FHS, that is, residing in a municipality that has 100% coverage of primary health care using multivariable logistic regression adjusted for the relevant demographic and socioeconomic covariables (Web Table 3 ). Covariables include sex, age of entry into the cohort, education, race, number of people in the family, household supplies, family income per capita, time spent receiving the conditional cash transfer policy, Bolsa Família, average municipal TB incidence, AIDS and diabetes and year of entry into the cohort.

Web Table 3 . Prediction models using logistic regression for the individual to live in municipalities with FHS coverage (<10% and 100%) in the 100 Million Brazilian Cohort for cutouts with TB outcomes. 

Web Table 4 presents the RRs for the education subpopulations, divided in less than 4 years of education and 5 years or more, for each TB outcome. 

We performed several sensitivity analyses. First, while -in order to have a more equitable ratio between exposed and not exposed individuals -we used in the main analyses individuals in municipalities with <10% FHP coverage as not exposed, we tested if using 0% FHS coverage was affecting the results. Web Table 5 presents the results for FHS 0% and 100% coverage, showing FHS effect estimates similar in direction, magnitude and statistical significance to the one presented as main results, with the exception of cure rate, probably due to the reduced number of observations (2, 067) .

Second, to evaluate the robustness of our results, we used a different quasi-experimental approach, the Propensity Score Matching, estimating the FHS impact Average Treatment on the Treated (ATT) (Web Table 6 ). All ATT measure were comparable in direction, magnitude and statistical significance with the main results of the study, except for cure rate probably due to the same reason explained above.

Third, to verify the relevance of the IPTW for unbiased FHS estimates, we run the same multivariable Poisson regressions without IPTW and compared with the ones with IPTW (Web Table 7 ). All FHS impact estimates were comparable in direction, magnitude and statistical significance with the main results of the study, however the FHS effects were considerably stronger on TB incidence and mortality without IPTW correction.

Fourth, to test the relevance of the variable that adjust for the endemic levels of TB in each municipality, we estimated the same regressions without the TB cumulative incidence as adjusting variable (Web Table 8 ). As above, all FHS impact estimates were comparable in direction, magnitude and statistical significance with the main results of the study, however the FHS effects were considerably stronger on TB incidence and mortality without adjusting for the TB average annual incidence.

Fifth, to verify the relevance of the adjustment for all covariates specifically in the Poisson regression, we run the models as Poisson bivariate regression (without IPTW) between the TB outcome and the FHS exposure (Web Table 7 ), even if we adjusted the previous logistic regression for PS with all relevant variables: the adjustment at the Poisson level, even if in the majority of the case is not changing the direction and statistical significance of the association, is an important factor of adjustment for the effects magnitude.

Sixth, to test the relevance of including other aggregated covariates, we have included the most TB-relevant municipal level variables available in our datasets (Web Table 10 ). These include: a. municipal Gini indexrepresenting the level of social inequalities, b. municipal percentage of sanitation coveragerepresenting sanitation and other municipal infrastructures, and c. municipal density of nursesrepresenting the level of coverage of healthcare professionals most relevant for the detection and followup of TB patients. The introduction of all these aggregate-level variables in the models did not meaningfully change the FHS estimates of effectiveness. We also tested the municipal Intraclass Correlation Coefficient (ICC) for TB incidence and mortality rates (Web Table 9 ): 0.005 and 0.0004 respectively, and an ICC close to 0from a pure statistical point of view -do not justify the use of a multilevel analysis.

Seventh, we have evaluated to stratify the analysis based on changes of diagnostic/treatment. No major point-of-care diagnostic innovations have been implemented in Brazil during the study period, considering the Gene Xpert diagnostic was nationally implemented in only at the end of 2013, corresponding also to the end of our study period. However, we have stratified the analyses on TB cure rates according to the introduction of the four-drug fixed-dose combination regimen in 2009 (Web Table  13 ), showing no important differences (the loss of significance of the >=2009 is possibly due to the reduced number of TB casesn.667).

Eight, in order to be sure that such exclusion was not affecting FHP impact estimates, we recoded the missing values of each categorical variable with a specific code for missing: considering they were included in the regression models as dummy variables, and the code for missing was not used as baseline category for the dummy estimation, this approach did not change the effect of each covariate. For continuous variables (age in years and months of Bolsa Familia receiving), we imputed the missing values as the median of their distribution. We successively fitted the same regression models of the study using all 8,581,358 individuals, obtaining FHP effect estimates (together with other covariate estimates) almost identical to main results presented in the manuscript, with an FHS RR 0.79 (95%CI: 0.74-0.85) versus the original RR 0.78 (95%CI: 0.72-0.84) (Web Table 11 ). We also compared the descriptive analyses of each variable of the 8,581,358 individuals versus the 7,308,968 used in the main analyses, and we found no meaningful differences (Web Table 12 ). 

The 100 Million Brazilian Cohort has a different demographic structure in comparison with the Brazilian population: with 47% of children below 15 years of age (against a national 21%). This implies different overall indicators of TB in comparison with the Brazilian population. While for TB incidence the values of the >15years are compatible with the national TB incidence (37.5 per 100,000 person-years in 2010), TB mortality rates are lower than the national values (2.5 per 100,000 person-years in 2010) because of the underrepresentation in the cohort of the older people.

Web Table 14 : Tuberculosis outcomes according to age groups and coverage.

Web Tables 15-18 are showing all the IPTW Poisson regression models, adjusted for all demographic and socioeconomic variables, for the association between Tuberculosis Incidence, Mortality, Cure, Case-fatality rates and the Family Health Strategy (FHS) coverage, overall and according to stratum of income, sex and age (Brazil, 2004 (Brazil, -2013 

Web Figure 3. Cumulative incidence of tuberculosis by percentage of FHS coverage (<10% and 100%) in the study cohort (Brazil

Cumulative Mortality of tuberculosis by percentage of FHS coverage (<10% and 100%) in the study cohort (Brazil

Assessing the accuracy of probabilistic record linkage of social and health databases in the 100 million Brazilian cohort: IJPDS

Administrative Data Linkage in Brazil: Potentials for Health Technology Assessment

Internal migration in Brazil using circular visualization

Bias due to left truncation and left censoring in longitudinal studies of developmental and disease processes

Brazil's Family Health Strategy -Delivering Community-Based Primary Care in a Universal Health System

Web Table 17 . IPTW Poisson regression models, adjusted for all demographic and socioeconomic variables, for the association between Tuberculosis Cure Rate and the Family Health Strategy (FHS) coverage according to stratum of income, sex and age (Brazil, 2004 (Brazil, -2013