The Contraction That Wasn't: Brazil's Primary Care Shock and the Missing Deaths

A number that cannot be compared to itself

Brazil's coverage indicator is not one series. It is at least three, and they do not measure the same thing.

Until 2020 the Ministry of Health published coverage as the number of family health teams multiplied by a fixed parameter of 3,450 people per team, divided by municipal population. This is a capacity measure: it says how many people the installed teams were designed to serve. It moves only when teams are added or removed, and it was published monthly from 2007.

From 2021, under the financing model introduced by Portaria 2.979/2019, the ministry switched to a registration measure: individuals actually registered by federally financed family health teams and primary care teams, divided by the IBGE population estimate [15]. The numerator now depended on whether teams had updated their patient registers, whether their funding was active, and which team types counted. The difference was visible immediately: the new series opened at 57.15% in January 2021 and reached 69.50% by December, a level the old series had never reached [17].

The ministry then created a third construction. Nota Técnica 301/2022 defines a "potential coverage" indicator explicitly designed to preserve a historical series, counting teams funded by states and municipalities as well as the federal government, and mixing capacity parameters with registered populations [16]. The document says what the problem was: the registration measure could not be compared with the pre-2021 series, and someone needed an indicator that could.

Two consequences follow. First, any claim that Brazilian coverage "fell from X to Y" across 2020-2021 is an artifact unless it names which construction it used. The break is larger than most year-to-year changes that commentators argue about. The research network Rede APS made the point in 2022: the longitudinal ESF series had become untrackable at municipal level because both numerator and denominator had changed [17].

Second, the honest measure of the programme's scale is the team count, which is definition-stable because a team is a team. That series runs from roughly 300 teams in 1994, to 21,400 in 2004, 39,300 in 2014, 43,223 in 2019, and 54,742 in 2025, with 277,341 community health agents and 34,950 oral health teams alongside them [12, 21]. Whatever else happened, the number of family health teams rose by about 27% between 2019 and 2025. A collapse in coverage that leaves 11,500 more teams standing is not a collapse in coverage. It is a change in what the word covered.

What the strategy actually is

The Family Health Strategy began in 1994 as the Programa Saúde da Família and became the organising model for Brazilian primary care through the 2006 and 2011 national policies. A team is normally a generalist physician, a nurse, a nursing technician, and four to six community health agents, responsible for a defined population of about 3,450 people in a defined territory. The community health agent is the distinctive component: a resident of the area employed to visit households monthly, register families, and bridge the clinic and the conditions that produce disease.

That design matters for everything that follows. The intervention is not "a doctor." It is a bundle: geographic empanelment, monthly household contact, a nursing-led workflow, referral coordination, and a generalist physician. Designs that isolate physician supply are testing one component of a system whose effects may depend on the others.

The financing history is a sequence of expansions and reversals. Federal transfers grew through the 2000s and were tied to team structure and, from 2011, to performance through PMAQ. The 2016 constitutional amendment froze federal spending in real terms for twenty years. The 2017 primary care policy loosened team composition, allowed teams with fewer community agents, and created a lighter "primary care team" modality. Portaria 2.979/2019 replaced structure-based transfers with capitation weighted by registration plus performance indicators, and counted both team types in the same numerator, weakening the financial advantage of the more expensive model [13, 26]. Portaria 3.493/2024 replaced most of that with a new model that restored per-team funding and added a quality component [14].

The expansion case, and why it is probably right

The evidence that the strategy reduced mortality is large, consistent in sign, and methodologically limited in ways that are well understood.

The earliest national study used 27 state-level panels from 1990 to 2002 and found that a ten percentage point increase in coverage was associated with a 4.5% decrease in infant mortality, controlling for water, sanitation, income, female literacy, fertility, and the supply of health professionals [1]. A municipality-level study of 771 municipalities from 1996 to 2004, using negative binomial fixed effects, found infant mortality 13%, 16%, and 22% lower at successively higher levels of coverage, with larger effects where baseline infant mortality and deprivation were worse [2].

The cardiovascular result is the cleanest. Across 1,622 municipalities from 2000 to 2009, consolidated coverage — at least 70% — was associated with a rate ratio of 0.82 for cerebrovascular mortality and 0.79 for heart disease mortality from ambulatory care sensitive causes. When coverage had been consolidated for all of the previous eight years, the ratios fell to 0.69 and 0.64. Coverage had no effect on accident mortality, used as a placebo, and the programme increased home visits, health education, and consultations while reducing hospital admissions for the same conditions [3]. A dose-response gradient, a plausible mechanism, a matching effect on utilization, and a null on an outcome primary care should not touch is about as good as ecological evidence gets.

For amenable mortality overall, the best-known estimate comes from 1,622 municipalities over 2000-2012: moving from zero to full coverage was associated with a 6.8% reduction. The effect was 11.0% in municipalities with the strongest health governance and 4.3% in the weakest [4]. That interaction matters out of proportion to its size. It says the programme's effect is not a property of the programme alone; it is a property of the programme plus the local state's capacity to run it.

For children, a four-country Latin American analysis covering 5,647 municipalities from 2000 to 2019 found that high coverage was associated with a rate ratio of 0.81 for under-five mortality, 0.72 for post-neonatal mortality, and 0.75 for deaths at ages one to four, with the strongest associations for nutritional deficiencies (0.55), anaemia (0.64), vaccine-preventable conditions (0.70), and infectious gastroenteritis (0.78). The authors attribute about 305,890 avoided under-five deaths over two decades to coverage [5]. At the individual level, a cohort of 75,339 live births to low-income mothers in Rio de Janeiro found that using family health services before or during the first two trimesters was associated with an adjusted odds ratio of 0.53 for neonatal death and 0.67 for infant death, with larger associations among poorer mothers [7]. A microsimulation for Brazil's fifteen largest cities found that even a 40 percentage-point coverage increase would cut premature non-communicable disease mortality by about 20%, against a 33% SDG target, while narrowing racial and income gaps in mortality [8].

Put together, the expansion literature supports a real effect concentrated in the outcomes primary care can plausibly affect: post-neonatal deaths from infection and nutrition, and ambulatory-care-sensitive deaths from hypertension and diabetes. The magnitudes cluster between roughly 5% and 20% depending on outcome, exposure contrast, and horizon. That is a substantial public health effect. It is not a miracle, and it is not uniform.

The confound that will not go away

The strategy expanded at the same time as almost everything else that improves child survival. Sanitation rose, vaccination was universal, fertility fell, incomes rose, and Bolsa Família put cash into the same poor households the teams were sent to serve. Any ecological estimate is an estimate of the strategy plus its era.

The best response to that objection is a 2017 systematic review of 31 studies, of which 25 were ecological and 21 used longitudinal quasi-experimental designs. Fourteen found higher coverage associated with lower post-neonatal and child mortality. The review's most interesting finding concerns the confound itself: when studies adjusted for Bolsa Família, the estimated effect of the strategy on child mortality became larger, not smaller [6]. That runs against the intuition that cash transfers were doing the work. One plausible explanation is that the programmes were not co-located as outsiders assume: transfers reached municipalities with established capacity, while the health strategy was pushed into places with less, so controlling for transfers removed a downward bias.

The review is also honest about where evidence is thin. Across thirteen studies of hospitalizations for ambulatory-care-sensitive conditions, no clear pattern emerged. Four studies found no effect on vaccination or low birth weight. Most included studies covered the pre-2017 period, so the entire policy churn since then — the 2017 loosening, the 2019 reform, the 2020-2022 squeeze, the 2023-2025 rebuild — is close to unevaluated in the peer-reviewed literature [19].

The correct summary of the expansion era is therefore not "primary care works." It is: primary care, delivered as a continuous team with community health agents in a poor municipality, reduces deaths from a specific and predictable set of causes, by a modest amount, with the largest benefits where baseline conditions are worst. Everything beyond that sentence is extrapolation.

Measuring the retreat without lying

The 2019-2022 squeeze was real, but it has to be measured in the right units. Four measures survive the definitional fog.

Federal money first. Under the 2016 spending cap, federal health spending was frozen in real terms while population and prices grew; the 2019 reform then redistributed within that ceiling. An analysis by the São Paulo state health secretariat, using ministry data, found that comparing 2022 with the model the new programme replaced, 3,312 municipalities — 60% of the country — received less. The same analysis reports that under the previous model about 98% of existing family health teams received federal co-financing, and that under the new one the average fell to 91.2% [18]. That is a quiet but consequential change: a team that exists but is not co-financed is a team its municipality is paying for in full, which is exactly the condition under which teams get dissolved.

The medical workforce next. Mais Médicos peaked at 18,240 positions, fell to about 13,700 by 2022, and had grown back to 27,300 by late 2025 — a 36% decline between January 2018 and January 2023, followed by a doubling [23]. The Cuban exit was the sharpest part of the decline.

The multidisciplinary layer third. The Núcleos de Apoio à Saúde da Família, which supplied nutritionists, psychologists, physiotherapists, and other specialists to family health teams, lost federal financing in 2020. Between 2018 and 2022 the number of NASF teams per 100,000 people covered by the strategy fell 17.2% [20]. This is the layer that handles the chronic disease, mental health, and rehabilitation work that generalist teams are least equipped to do alone.

And the teams themselves. They grew: from 43,223 in 2019 to 54,742 in 2025, with 6,288 multidisciplinary eMulti teams — more than the 4,487 NASF centres that existed in 2019 — and 277,341 community health agents [12, 21].

The pattern is coherent once the units are right. Between 2019 and 2022 Brazil did not close family health teams; it made each team cheaper in real terms, cut the specialist support around it, and lost a large share of the doctors staffing the poorest municipalities. By 2025 the team count, the doctor pipeline, and the multidisciplinary layer had all been rebuilt past 2019 levels. Whatever damage the squeeze did was concentrated in a four-year window and in the quality and staffing of care rather than in its nominal availability.

That distinction is the whole argument. A coverage percentage that counts teams shows a programme holding steady or growing. A percentage that counts registered individuals shows whatever the registration campaign did. Neither shows that the average team lost its nutritionist, its psychologist, and part of its medical hours.

The largest natural experiment in primary care

The Cuban withdrawal is the closest thing this literature has to a randomized shock. It was abrupt, unanticipated, large, and geographically concentrated in places that had no other doctors. Municipalities with only Cuban doctors lost their entire Mais Médicos medical workforce; municipalities with none lost nothing. That is a clean treatment-control structure, and at least three research groups have used it.

The findings converge on a pattern that is uncomfortable for the forecast. Utilization fell: appointments for chronic conditions, particularly hypertension and diabetes, declined where Cuban doctors left. But hospitalizations and mortality did not rise in any way that survives scrutiny, and the consolidated reading of the Mais Médicos evidence is that the programme improved access while having surprisingly limited effects on extreme outcomes such as death [10].

There are two ways to read that. The deflationary reading is that primary care's mortality effect has been overstated all along, and that expansion-era estimates pick up municipal development rather than medical care. The constructive reading is that the natural experiment tested the wrong component. The withdrawal removed physicians. It did not remove the team, the nurse, the community health agents, the building, the register, or the referral relationship. If the strategy's mortality effect runs substantially through those components — monthly household contact, early detection, vaccination, diarrhoeal disease management, nursing-led chronic follow-up — then removing the doctor is a partial treatment, and a partial treatment should produce a partial and possibly undetectable effect on deaths.

The constructive reading is better supported, for three reasons. The expansion literature's strongest and most consistent effects are on post-neonatal mortality and on infectious and nutritional causes, the outcomes least dependent on physician presence. The placebo tests behave correctly: coverage moves accident mortality not at all. And effect modification by governance and baseline deprivation implies the programme's value lies in execution quality, not in the mere presence of a physician.

None of that makes the forecast stupid. It makes it a forecast: it applied an elasticity estimated from expansion, in places that had little care, to a contraction, in places that had some. There is no law that the dose-response curve is symmetric.

Why the forecast missed

Five mechanisms explain most of the gap.

Substitution is the largest. When a doctor disappears from a clinic, patients do not stop needing care; they go somewhere — hospital outpatient departments, private clinics paid out of pocket, other public facilities nearby. Utilization data show the shift; mortality data do not, because the care was relocated rather than lost. Substitution is expensive for households and bad for continuity, but it breaks the causal chain between doctor supply and death.

The outcome was wrong for the question. Mortality is a rare, lagged, noisy event at municipal scale. A municipality of 8,000 people experiences a handful of infant deaths a year, so a 10% change is one or two events, indistinguishable from Poisson noise. Forecasts that multiply a small elasticity by a large population produce impressive absolute numbers while remaining statistically invisible where the shock occurred.

The window was short and contaminated. The withdrawal happened in late 2018. Plausible mortality consequences accrue over years, and the years available were 2019-2021 — the years in which COVID-19 reorganized health-seeking behaviour everywhere. Any national before-and-after analysis in that window is mostly measuring the pandemic.

The elasticity was borrowed from a different margin. Adding a doctor to a municipality that has never had one produces a large change in what is available. Removing one doctor from a team that still has a nurse, four community health agents, and a functioning referral path produces a much smaller change in the effective supply of care. The relevant marginal product is not constant, and the expansion-era estimates identify the high end of the curve.

And the policy response outran the model. The 2023-2025 rebuild added 13,600 doctors to Mais Médicos, 4,750 family health teams, and 6,288 multidisciplinary teams [21, 22, 23]. Whatever mortality cost the squeeze imposed, the counterfactual baseline for measuring it stopped being stable almost as soon as the measurement window opened.

Calibrated verdict

• Very high confidence (~95%) that the Family Health Strategy is associated with lower infant and ambulatory-care-sensitive mortality in Brazil, with a dose-response gradient and correct placebo behaviour [1, 2, 3, 4, 5].
• High confidence (~85%) that the causal effect is real rather than entirely confounded, given consistency across fixed-effects, cohort, and cross-country designs, while accepting that magnitudes are inflated by municipal development and coverage measurement error.
• Moderate confidence (~65%) that the true effect of full coverage is 5-12% for all-cause amenable mortality and 15-25% for post-neonatal mortality, with published extremes reflecting design and exposure differences rather than a single parameter.
• High confidence (~85%) that the Cuban withdrawal produced no mortality increase detectable in national data, and moderate confidence (~70%) that it produced little true mortality effect, because utilization fell while deaths did not [9, 10].
• Moderate confidence (~60%) that the 2019-2022 financing and workforce squeeze caused a real but small mortality cost, concentrated in poor municipalities and chronic disease, which existing studies are underpowered to detect.
• High confidence (~90%) that the headline Brazilian coverage percentage is not comparable across 2020-2021, and that claims of a coverage collapse based on that series are methodological artifacts [15, 16, 17].

Reversal conditions

This reading would change under any of the following.

A well-powered municipal event study around the 2019-2022 financing change, using definition-stable team counts and community health agent density as exposure, finding a clear increase in ambulatory-care-sensitive mortality where co-financing was lost. That would move the quality channel from plausible to established and raise the estimated cost of the squeeze substantially.

A re-analysis of the Cuban withdrawal using individual-level death records and a pre-registered exposure definition, finding mortality effects that administrative aggregates could not see. Small-area mortality data exist at a finer grain than the municipal studies used, and a null at that resolution would be much more informative.

Evidence that the 2023-2025 expansion produces no improvement in post-neonatal or ambulatory-care-sensitive mortality by 2030, in a period without pandemic disruption. That would substantially weaken the case that the strategy's mortality effect is causal rather than developmental.

A credible decomposition showing the expansion-era estimates are driven by municipalities simultaneously receiving large cash transfers, sanitation investment, and federal hospital spending. The Bolsa Família adjustment points the other way, but it is one specification in a literature with many.

The study that would settle it

The decisive design is feasible and, as far as this review found, unrun. It needs a definition-stable measure of primary care intensity, small-area or individual death records rather than municipal aggregates, and an exposure that varies the quality of teams rather than their existence.

The natural vehicle is the 2019-2022 financing shock. Federal co-financing of existing teams fell from about 98% to 91.2%, meaning thousands of teams moved from fully federally financed to partially or unfunded, at different times and to different degrees across municipalities. That variation is plausibly exogenous to local mortality trends, because it was generated by a formula, a spending cap, and a registration threshold rather than by local health conditions. Linking it to cause-specific mortality at small-area level, with the pandemic period handled explicitly rather than averaged in, would test the quality channel directly.

The complement is the 2023-2025 rebuild, a positive shock of similar structure: 4,750 new teams, 13,600 new doctors, and 6,288 multidisciplinary teams, allocated by vulnerability index and municipal application. A pre-registered analysis of post-neonatal mortality and ambulatory-care-sensitive admissions in the municipalities that received the largest allocations would be running by 2027 and readable by 2030.

Until then, the defensible summary is this. Brazil built the largest and best-studied primary care system in the developing world, and it probably works, modestly, on the outcomes it was designed to affect. The country then squeezed it for four years in ways that are hard to measure and easy to overstate, and is now rebuilding it past where it started. The most interesting fact in the literature is not that primary care saves lives. It is that removing eight thousand doctors from the poorest municipalities in Brazil produced barely a ripple in the death statistics — which says something important about what the doctors were doing, and something more important about what the teams around them were doing.

References

1. Macinko J, Guanais FC, de Fátima M, de Souza M. Evaluation of the impact of the Family Health Program on infant mortality in Brazil, 1990-2002. Journal of Epidemiology and Community Health. 2006;60(1):13-19. doi:10.1136/jech.2005.038323
2. Aquino R, de Oliveira NF, Barreto ML. Impact of the family health program on infant mortality in Brazilian municipalities. American Journal of Public Health. 2009;99(1):87-93. doi:10.2105/AJPH.2007.127480
3. Rasella D, Harhay MO, Pamponet ML, Aquino R, Barreto ML. Impact of primary health care on mortality from heart and cerebrovascular diseases in Brazil: a nationwide analysis of longitudinal data. BMJ. 2014;349:g4014. doi:10.1136/bmj.g4014
4. Hone T, Rasella D, Barreto M, Atun R. Large reductions in amenable mortality associated with Brazil's primary care expansion and strong health governance. Health Affairs. 2017;36(1):149-158. doi:10.1377/hlthaff.2016.0966
5. Moncayo AL, Cavalcanti DM, Ordoñez JA, Almeida C. Can primary health care mitigate the effects of economic crises on child health in Latin America? An integrated multicountry evaluation and forecasting analysis. The Lancet Global Health. 2024;12(6):e938-e946. doi:10.1016/S2214-109X(24)00094-9
6. Bastos ML, Menzies D, Hone T, Dehghani K, Trajman A. The impact of the Brazilian family health strategy on selected primary care sensitive conditions: a systematic review. PLoS ONE. 2017;12(8):e0182336. doi:10.1371/journal.pone.0182336
7. Hone T, Been JV, Saraceni V, Coeli CM, et al. Associations between primary healthcare and infant health outcomes: a cohort analysis of low-income mothers in Rio de Janeiro, Brazil. The Lancet Regional Health – Americas. 2023;22:100519. doi:10.1016/j.lana.2023.100519
8. Basu S, Hone T, Villela D, Saraceni V, et al. Contribution of primary care expansion to Sustainable Development Goal 3 for health: a microsimulation of the 15 largest cities in Brazil. BMJ Open. 2022;12(1):e049251. doi:10.1136/bmjopen-2021-049251
9. Hone T, Powell-Jackson T, Santos LMP, et al. Impact of the Programa Mais Médicos (More Doctors Programme) on primary care doctor supply and amenable mortality: quasi-experimental study of 5,565 Brazilian municipalities. BMC Health Services Research. 2020;20:873. doi:10.1186/s12913-020-05716-2
10. Rocha R. More doctors, better health? Consolidating evidence from Brazil's Mais Médicos program. Social Science & Medicine. 2025:117559. doi:10.1016/j.socscimed.2024.117559
11. Vivas Francesconi G, Tasca R, Basu S, Rocha TAH, Rasella D. Mortality associated with alternative policy options for primary care and the Mais Médicos (More Doctors) Program in Brazil: forecasting future scenarios. Revista Panamericana de Salud Pública. 2020;44:e31. doi:10.26633/RPSP.2020.31
12. Organisation for Economic Co-operation and Development. Primary Health Care in Brazil. OECD Reviews of Health Systems. Paris: OECD Publishing; 2022.
13. Brasil. Ministério da Saúde. Portaria nº 2.979, de 12 de novembro de 2019. Institui o Programa Previne Brasil. Diário Oficial da União; 2019.
14. Brasil. Ministério da Saúde. Portaria GM/MS nº 3.493, de 10 de abril de 2024. Altera o cofinanciamento federal da Atenção Primária à Saúde. Diário Oficial da União; 2024.
15. Brasil. Ministério da Saúde. Nota Técnica nº 418/2021-CGGAP/DESF/SAPS/MS: nota metodológica da cobertura da Atenção Primária à Saúde das equipes financiadas pelo Ministério da Saúde. Brasília; 5 November 2021.
16. Brasil. Ministério da Saúde. Nota Técnica nº 301/2022-CGESF/DESF/SAPS/MS: metodologia da Cobertura Potencial da Atenção Primária à Saúde. Brasília; 5 December 2022.
17. Rede de Pesquisa em Atenção Primária à Saúde. Invisibilidade do indicador de cobertura da ESF. Rede APS Bulletin. 18 April 2022. https://redeaps.org.br/informativos-e-noticias/noticias/invisibilidade-do-indicador-de-cobertura-da-esf/11713
18. Governo do Estado de São Paulo, Secretaria de Estado da Saúde. Portaria 3.493/2024: novo modelo de financiamento da Atenção Primária em Saúde. Technical presentation, July 2024. https://www.cosemssp.org.br/wp-content/uploads/2024/07/PT-3493-MARINA-MELO.pdf
19. Aquino R, Vilasbôas ALQ. Impacts of the Family Health Strategy on the health status of the Brazilian population: a literature review. Ciência & Saúde Coletiva. 2025;30:e13942025. doi:10.1590/1413-812320253012.13942025
20. Instituto de Estudos para Políticas de Saúde. Como evoluiu o número de vínculos profissionais NASF após o fim do incentivo financeiro ao NASF. Nota Técnica nº 31. São Paulo: IEPS; 2023.
21. Brasil. Ministério da Saúde. Política Nacional da Atenção Básica: 31 anos da Estratégia Saúde da Família. Technical presentation, December 2025. https://congresso.cosemspr.org.br/wp-content/uploads/2025/12/Politica-Nacional-da-Atencao-Basica.pdf
22. Brasil. Ministério da Saúde. Mais Saúde da Família: guia para ampliação e qualificação no seu município. Brasília: Ministério da Saúde; 2025.
23. Brasil. Ministério da Saúde. Programa Mais Médicos cresce 99% e amplia acesso à Atenção Primária em 4,5 mil municípios brasileiros. News release, 24 December 2025. https://www.gov.br/saude/pt-br/assuntos/noticias/2025/dezembro/programa-mais-medicos-cresce-99-e-amplia-acesso-a-atencao-primaria-em-4-5-mil-municipios-brasileiros
24. Brasil. Ministério da Saúde. Posicionamento do Ministério da Saúde sobre o programa Mais Médicos. News release, 14 November 2018. https://www.gov.br/saude/pt-br/assuntos/noticias/2018/novembro/posicionamento-do-ministerio-da-saude-sobre-o-programa-mais-medicos
25. Darlington S. Brazil fails to replace Cuban doctors, hurting health care of 28 million. The New York Times. 11 June 2019.
26. Melo MA, Silva ASD, Pimenta AL, Mendes A, Paiva JLL. PHC funding over time: inducement, expansion, and qualification of the Family Health Strategy model. Ciência & Saúde Coletiva. 2025;30:e12642025. doi:10.1590/1413-812320253012.12642025