Context

During the past decades, the Brazilian health system has made important progress, particularly on improving maternal and child health. There is preliminary evidence suggesting that expansion of primary health care, through the Family Health Programme (PSF), may have had a positive contribution to the advances in population health (Rocha and Soares 2010). The PSF was adopted in 1994, being based on multi-professional Family Health Teams (FHT) regularly visiting households to monitor community health and perform diverse health promotion and prevention activities. The evolution of PSF shows that the programme covered approximately 62,929,042 individuals in 2,490 municipalities during 2000, growing to 116,590,182 individuals registered in the program in 5,018 municipalities during 2014, mainly centred among low income households (DATASUS 2017).

There is scarce empirical evidence related to the health impacts of PSF (Rocha and Soares 2010) and there are endogeneity issues that were not addressed in previous literature. The absence of a suitable identification strategy in the methodology adopted in such studies weakens their results in terms of unbiasedness and robustness.

The study design of our impact evaluation study is based on the use of both individual and municipal-level data to assess the impacts and the costs of the PSF. This should allow us to generate evidence on the efficiency and effectiveness of primary health care in Brazil. The adoption of quasi-experimental methods to assess the PSF helps us deliver information to support health system stakeholders and inform evidence-based decision-making in public policy regarding community health intervention. The interventions to be implemented are the presence vs. absence of the PSF within municipalities, considering also different levels of coverage as a continuous treatment.

Research Questions

The project will develop and apply methodologies to investigate two key research questions:

1) What has been the quantitative impact of PSF expansions on broad indicators of individual health status (including non-communicable diseases) and on equity in health care access and utilisation?

2) What is the cost-effectiveness ratio of the PSF and its variation according to scale?

Methods

Question 1 will be investigated through propensity score matching for continuous treatments, suited to examining the issue of multiple programme coverage levels and which has not been applied before in a similar context.

The dose-response method aims to investigate if, and to what extent, the health indicators are sensitive to the level of PSF coverage (assuming continuous treatment). To implement the method, we proposed a two-stage approach: firstly, a Generalised Propensity Score (GPS) is estimated and conditioned on a set of covariates (e.g. economic product per capita, population, degree of urbanisation, schooling, political alignment of majors and federal government, and other variables coming from municipality level); and secondly, a dose-response function whose dependent variable is the health indicator and conditioned on GPS.

We are applying these methodologies to secondary micro- and macro-datasets: the Brazilian National Health Survey (2013-2014), the National Household Survey (PNAD 1998/2003/2008), and the Ministry of Health data repository (DATASUS). These datasets will allow us to carry-out comprehensive analyses at both the individual and aggregate (municipality) levels.

The DATASUS financial data and results from question 1 will help us address question 2. We plan to conduct simple cost-effectiveness modelling of costs per health gain obtained for a given condition (e.g. monitored high blood pressure), and linking this to results on expected future benefits (e.g. mortality avoided) from the available relevant literature.

Intended Outcomes and Impact

We propose to address the PSF health impact endogeneity challenges by using treatment variables related to PSF adoption, timing of adoption and coverage level in macro-datasets at municipality level. This will allow us to assess health impacts related to chronic diseases using panel data models. We also plan to address identification of health impacts and their intermediate channels using micro-datasets at individual level, dealing with unobserved heterogeneity by adoption of dose-response.