The effect of increasing Women's autonomy on primary and repeated caesarean sections in Brazil

Abstract Caesarean section (C‐section) rates continue to rise globally. Yet, there is little consensus about the key determinants of rising C‐section rates and the sources of variation in C‐section rates across the world. While C‐sections can save lives when medically justified, unnecessary surgical procedures can be harmful for women and babies. We show that a state‐wide law passed in São Paulo (Brazil), which increased women's autonomy to choose to deliver via C‐section even when not medically necessary, is associated with a 3% increase in overall C‐section rates. This association was driven by a 5% increase in primary C‐sections, rather than repeated C‐sections. Since the law emphasizes women's autonomy, these results are consistent with mothers' demand being an important contributor to high C‐section rates in this context.

• Act 1: It is the right of the pregnant woman, in elective situations, to opt for a C-section, guaranteed by her autonomy, if she has received all the detailed information about the vaginal and C-section deliveries, their respective benefits and risks. The decision must be registered in a free and informed consent form, elaborated in an easily understood language, respecting the pregnant woman's sociocultural characteristics. • Act 2: To guarantee the safety of the foetus, the C-section at the request of the pregnant woman, in situations of usual risk, can only be performed after the 39 th week of pregnancy and must be registered in the medical record. • Act 3: It is ethical for a doctor to perform a C-section on request, and if there is a disagreement between the medical decision and the pregnant woman's wishes, the doctor may claim her right to professional autonomy and, in these cases, refer the pregnant woman to another professional.
• In public healthcare systems, pregnant women can be subject to obstetric violence during vaginal delivery (e.g., the long length of delivery and labour pain). Caesarean deliveries may help overcome these issues. • There is insufficient scientific evidence on the relationship between caesarean deliveries and maternal and infant deaths.

D. Additional Information on the Unconstitutionality of Law 17,317
In July 2020, Law 17,137 was declared unconstitutional by the São Paulo's Court due to the perceived conflict with the Constitutional Law. The Constitution of the Federative Republic of Brazil establishes that women can deliver their babies by caesarean section when medically justified. However, Law 17,137 allowed women to deliver babies via caesarean regardless of medical reasons. Moreover, the cost of caesarean sections in public hospitals is covered by the federal government, as all the costs of public health hospitals in Brazil.

E. Healthcare system in Brazil and overall C-section rates in 2018
The The reference value for a medical procedure is divided into hospital services and professional services. The value of hospital services includes per diems, room rates, food, hygiene, patient support staff in bed, materials, medicines and Auxiliary Services for Diagnosis and Therapy (ASDT), except special medicines and special ASDT. The values of professional services correspond to the fraction of professionals (doctors, dentists and obstetric nurses) who worked during the hospitalization. The outpatient service includes outpatient permanence fee, professional services, materials, medication, and support. This system does not differentiate reference values by state or geographical region. Therefore, the financial repercussion of a Csection or a vaginal delivery in the Unified Health System should be similar across states and municipalities, given the same involved inputs. Using cost data from three public maternity hospitals (two in Rio de Janeiro and one in Belo Horizonte), Entringer et al. (2018)  Despite the lack of studies about regional disparities regarding cost-effectiveness of caesarean and vaginal deliveries, the public health literature has documented regional disparities in the provision of healthcare services across Brazilian regions (Massuda et al. 2018). Such disparities have been observed in the coverage and quality of birth care service. The coverage of health facilities is much larger in the Southeast and South regions, which is associated with high Csection rates in the corresponding regions compared to the rest of the country (Barros et al., 2015). The literature also shows that most maternity wards across the country have a low rate of adequacy that can affect the quality of labour and birth care, especially in the North and Northeast regions ( Table S2 for the list of states. Table S1 reports the overall CS (C-section) rate in the state of São Paulo (SP) and the mean of overall CS rate for the rest of Brazil (RB). Using the standard 2-by-2 difference-in-differences (DiD) method, we compute an increase of 0.9 percentage point (pp) in the overall CS rate due to the introduction of the law in September-October 2019 and an increase of 1.1 pp in November-December 2019. The standard 2-by-2 DiD point estimates are comparable to those using the synthetic control method (SCM). However, we use the SCM approach in the main analysis for two main reasons:

F. Standard 2-by-2 difference-in-differences point estimates
• First, as highlighted by Abadie (2021), the SCM is appropriate to estimate the effects of interventions that are implemented at an aggregate level affecting a small number of large units (such as a cities, regions, or countries), on some aggregate outcome of interest. In these setups with large control groups and small treatment groups, there is a large imbalance in sample size between groups which may affect statistical inference. As our setting involves an intervention affecting only one aggregate unit (the state of São Paulo), this makes SCM the most suitable approach for our setup. • Second, sometimes the standard DiD is not enough and needs to be extended to include higher-order differences (de Oliveira et al. 2021). Indeed, there are many degrees of freedom when trying to find a control state and using difference-in-differences methods (including how to control for group-specific trends). SCM methods help us restrict the number of degrees of freedom and limit the scope for cherry-picking, at least ex ante (Ferman et al. 2020). Table S2 below provides the weights used to construct the synthetic control for each of our main outcomes of interest.  It is worth noting that states with higher weights in the synthetic control group are, in general, states in the Southeast, South and Centre-West regions. This means that the donor pool used to generate the Synthetic São Paulo captures the regional disparities discussed in Appendix E.

G. Weights to construct the synthetic control group
A7 Figure G1 displays the geographical distribution of weights to construct the synthetic control when studying overall C-section rates. Figure G1: Map of the states that contribute to the synthetic control when studying the effect on the overall C-section rate. See Table S2 for the list of states.

H. SCM with more than one validation period
The existing literature on SCM suggests that the lengths of the training and validation periods depend on application-specific factors such as the extent of data availability on outcomes in the pre-and post-intervention period (Abadie 2021). In the research letter, we chose one validation period given the short time period that we are looking at (2 years at a bi-monthly frequency).
There is a trade-off between using data points for validation and for prediction. By increasing the number of periods for validation, we lose information that can be used to predict the outcome of synthetic São Paulo.
In Table S3, we demonstrate that our results are robust to increasing the number of validation period by varying the number of validation periods from 2 to 4. The results are very similar to those in the research letter where 1 validation period is used.  Figure S1 shows that despite the seasonality in CS rates, there is a sharp increase in the overall CS rate after the passing of the law in São Paulo (vertical dashed red line). We do not see the same pattern in the CS rate for the rest of Brazil. Figures S2 and S3 show that this increase is driven by increases in primary CS rates rather than repeated CS rates.

J. Impact of the law by mother's sociodemographic characteristics
In Figures S4 to S15 below, we present raw trends (left column) and SCM estimates (right column) of mothers' characteristics. The figures show no evidence of compositional changes in the proportion of high-educated (12 years or more) vs. low-educated (11 years of education or fewer) mothers, young (26 years or less) vs. non-young (27 years or more) mothers and married vs. unmarried mothers in São Paulo relative to the rest of Brazil.
However, Figures S5 and S7 suggest that the overall CS rate among low-educated mothers was more affected than that among high-educated mothers. Similarly, Figures S9 and S11 show that the overall CS among young mothers was more responsive to the introduction of the law than that among non-young mothers. Finally, Figures S13 and S15 show that the overall CS rate among unmarried mothers was more responsive to the law than for married mothers. Thus, the SCM estimates for different sociodemographic groups of mothers suggest heterogeneous effects of the law on the overall CS rate.
While the epidemiology literature has documented that the upward trend in CS rates in Brazil during the 2000s has been mainly driven by high-educated, young, and primiparous mothers (Barros et al., 2015), our results suggest that the increase in CS rate with the introduction of the law was driven by low-educated, young, and unmarried mothers. One interpretation of these heterogeneous effects is that while the secular increase in CS rates in Brazil has been driven by high-educated mothers, the effect of the law is likely to be concentrated among a group of mothers where CS rates are comparatively low. In other words, the "compliers" with the law are more likely to be low-educated, young, and unmarried mothers.

K. Impact of the law on fertility
In Figure S18, we plot the bimonthly time-trend of live births in São-Paulo and the rest of Brazil.
The figure shows no change in the trend of live births in the 4 months after the law was implemented. In Figure S19, we use the SCM to estimate changes in live births. The figure shows that there are no differences in the number of live births between São Paulo and synthetic São Paulo. While our sample ends in December 2019, future work could investigate fertility responses 9 months after the implementation of the law.

L. Impact of the law by availability of obstetricians and surgical obstetric beds
To investigate the role of supply factors, we compare the effects of the law across different health regions in São Paulo. We measure the availability of obstetricians and surgical obstetric beds at the health region level within São Paulo state. We use the number of obstetricians and surgical obstetric beds (Source: Cadastro Nacional de Estabelecimentos de Saúde/Datasus) as proxies for the availability of human and physical resources to perform C-sections. These variables are available for the 17 health regions of São Paulo state, clusters of municipalities by which the healthcare service is organized and coordinated by the State Government.
First, for each health region in São Paulo state, we normalize the number of obstetricians and surgical obstetric beds by the total number of live births (per 1,000 live births) and compute the median for both measures in 2019. Second, we split the São Paulo sample into two groups: newborns from areas with high (above 9.5) vs. low (9.5 or below) availability of obstetricians, as well as those from areas with a high (above 8.9) vs. low (8.9 or below) availability of surgical obstetric beds. Third, we employ the SCM and estimate the effect of the introduction of the Law 17,137 for these two groups.
Figures S20 to S23 below demonstrate that the effects of the law on CS rates are similar in areas with below-and above-median availability of obstetricians and surgical obstetric beds. This suggests a limited role of supply-side factors on the effect of the law. While not reported here, it is worth mentioning that the law did not affect the availability of obstetricians or surgical obstetric beds (results available from the authors upon request).

M. Impact of the law on health at birth
In the research letter we did not present results on the effect of the law on health outcomes at birth because these estimates are likely to be confounded by compositional bias. Since the law stipulates that C-sections are permitted after 39 weeks of gestation, the law is expected to mechanically increase gestational length amongst those deliveries via C-section. Therefore, birth outcomes can only be examined conditional on gestational length, resulting in a selected sample of babies being analysed. Indeed, gestational length is one of the outcomes used in studies looking at impacts on health at birth.
Despite these limitations, in Figures S24 to S27 below, we present both raw trends (left column) and SCM estimates (right column) of a commonly used measure of health at birth, birth weight (Clarke et al., 2021). We are unable to detect any effects of the law on average birth weight or the prevalence of low birth weight (<2500 g) births.

O. Excluding neighbouring states
To assess the validity of our analysis to potential migration biases from neighbouring states, Figures S31 to S33 replicate the SCM analysis displayed in Figure 2 (a, b, c) after excluding the four states that share borders with São Paulo: Mato Grosso do Sul, Minas Gerais, Paraná, and Rio de Janeiro. Our findings are virtually the same. Figure S31. Estimated placebo effect on Overall Csection rates Figure S32. Estimated placebo effect on Primary Csection rates A17 Figure S33. Estimated placebo effect on Primary Csection rates

P. Summary of quasi-experimental studies
US Supply-side financial incentives: Change in the price differential for C-sections vs vaginal deliveries paid by commercial insurers to Increasing the physician price differential by 1 standard deviation ($420) leads to a 12% increase in the odds ratio for C-A18 hospitals and physicians. section delivery. Increasing the hospital price differential by one standard deviation ($5,805) for births delivered by hospital-exclusive physician groups yields a 31 percent increase in the odds ratio. Borra et al.

Spain
Demand-side financial incentives: Removal of €2,500 baby-bonus for babies born starting after 1 Jan 2011.
Increased in number of daily Csections in late December 2010 (about 120 per day extra when focusing on the one-week window). de Elejalde & Giolito (2021) Chile Supply-side & demand-side incentives: a policy change that increased delivery at private hospitals by reducing the out-ofpocket cost for women with public insurance (private hospitals receive the same price for a vaginal or Csection delivery).
Increased the probability of a Csection by 8.6pp.

Currie & MacLeod
US Supply-side accountability: adoption of tort reforms by states in the US during the 1980s and 1990s.
The adoption of the rule of joint and several liability (the so-called deep pockets rule), in which doctors are held more accountable for their own actions, reduced the probability of Csection by 7%. Amaral-Garcia et al. (2021) England Effect of internet diffusion on Csection deliveries.
First-time mothers living in areas with better internet access are 6% more likely to obtain a C-section. No corresponding effects on healthcare outcomes.