Update and extension of the Brazil SimSmoke model to estimate the health impact of cigarette smoking by pregnant women in Brazil

A previous application of the Brazil SimSmoke tobacco control policy simulation model was used to show the effect of policies implemented between 1989 and 2010 on smoking-attributable deaths (SADs). In this study, we updated and further validated the Brazil SimSmoke model to incorporate policies implemented since 2011 (e.g., a new tax structure with the purpose of increasing revenues/real prices). In addition, we extended the model to estimate smoking-attributable maternal and child health outcomes (MCHOs), such as placenta praevia, placental abruption, preterm birth, low birth weight, and sudden infant death syndrome, to show the role of tobacco control in achieving the Millennium Development Goals. Using data on population, births, smoking, policies, and prevalence of MCHOs, the model is used to assess the effect on both premature deaths and MCHOs of tobacco control policies implemented in Brazil in the last 25 years relative to a counterfactual of policies kept at 1989 levels. Smoking prevalence in Brazil has fallen by an additional 17% for males (16%-19%) and 19% for females (14%-24%) between 2011 and 2015. As a result of the policies implemented since 1989, 7.5 million (6.4-8.5) deaths among adults aged 18 years or older are projected to be averted by 2050. Current policies are also estimated to reduce a cumulative total of 0.9 million (0.4-2.4) adverse MCHOs by 2050. Our findings show the benefits of tobacco control in reducing both SADs and smoking-attributable MCHOs at population level. These benefits may be used to better inform policy makers in low and middle income countries about allocating resources towards tobacco control policies in this important area.


Introduction
Brazil has experienced a large decline in smoking prevalence in the last 25 years 1,2,3,4 .A previously developed SimSmoke simulation model by Levy et al. 1 estimated nearly a 50% decline in Brazil's smoking prevalence among adults aged 18 years or older, due to the implementation of interventions recommended by the World Health Organization Framework Convention on Tobacco Control (WHO-FCTC) between 1989 and 2010 5 .Since 2011, several states, covering 50% of the population, implemented 100% smoke-free air restrictions, followed by a federal law in 2014 6,7 .Moreover, in 2012, a new tax structure went into effect to further increase cigarette tax revenues, which led to an increase of 23% in inflation-adjusted prices between 2012 and 2014 8,9 .
Tobacco control is not only vital for reducing smoking prevalence and, as consequence, smokingattributable deaths among adults, but it also contributes towards the attainment of the Millennium Development Goals proposed by the United Nations' Secretary-General in 2000, in particular goal #4 (reducing child mortality) 10,11 .Indeed, studies have found a causal relationship between prenatal smoking and adverse maternal and child health outcomes (MCHOs), such as placenta praevia, placental abruption, preterm birth, low birth weight, and sudden infant death syndrome 12,13,14 .Although prenatal smoking is a particular concern in low and middle income nations 15 , the impact of prenatal smoking and the potential role of tobacco control policies in improving child health have received little attention 14,16 .
With strong tobacco control policies and a comprehensive surveillance network for MCHOs based on three health information systems -Information System on Live Births (SINASC), Mortality Information System (SIM), and Hospital Information System (SIH) 17,18 -, Brazil is well-suited to evaluate the effect of implementing tobacco control policies on smoking-attributable adverse MCHOs.We updated the Brazil SimSmoke to incorporate the role of new policies and extended it to assess the overall impact of tobacco policies on the reduction of cigarette smoking by pregnant women and, as a consequence, on MCHOs.To our knowledge, our study represents the first one that considers the effect of tobacco control policies on MCHOs for low and middle income countries, including Brazil.A large scale survey in 2013 also made it possible to further validate the model 4,19 .

Basic model and update in 2015
Brazil SimSmoke estimated the effect of Brazil's tobacco control policies on smoking prevalence and smoking-attributable deaths (SADs) among adults aged 18 years or older between 1989 and 2010 1 .A discrete-time, first-order Markov process was employed to project population growth by births and deaths, and smoking prevalence by smoking initiation, cessation, and relapse rates.Changes in tobacco control policies shifted smoking prevalence through initiation and cessation.SADs were calculated using relative risks, smoking prevalence, and death rates.Projected smoking prevalence rates were validated using data from the Brazilian module of the 2003 World Health Survey (WHS) 2 and from the 2008 Brazilian Global Adult Tobacco Survey (GATS) 20 .
Since 2011, stricter smoke-free air laws were implemented and cigarettes taxes were increased 6,7,8,9 .We incorporated those policies and validated the model using results from the 2013 Brazilian GATS, a large-scale population-wide survey 19 .

Extension of the model: incorporating smoking-attributable birth outcomes
Brazil SimSmoke was extended to estimate smoking-attributable MCHOs , including cases of low birth weight (LBW), preterm births (PTB), sudden infant death syndrome (SIDS), placenta praevia, and placental abruption.The number of smoking-attributable MCHOs is estimated as a product of the total number of cases for each outcome and the smoking-attributable fraction (SAF) for that outcome.The SAF is defined as: where p = active prenatal smoking prevalence and RR o = relative risk of MCHO for active prenatal smokers compared to prenatal nonsmokers 21 .Figure 1 shows the inputs, intermediate calculations, and outputs.Due to uncertainty in each of the three parameters, we conducted sensitivity analyses with credible upper and lower bounds.
To obtain the number of MCHOs, births by age and gender of the women of child-bearing potential were multiplied by the corresponding prevalence of MCHOs.Data on MCHOs from 2000 to 2013 were obtained from the SINASC, the SIM, and the SIH 17,18,22 , aggregated by ages 18-19, 20-29,  30-39, and 40-49.Because placenta praevia and placental abruption were included as one category since 2009, rates were combined in prior years.To avoid overlap, we excluded PTB from LBW cases.Since reliable data for adverse MCHOs were not available before 2000, we estimated trends in years prior to 2000 from the published literature on an outcome-by-outcome basis 23 , yielding an increasing trend for PTB and a decreasing trend for SIDS between 1989 and 2000, but no trend for LBW and placenta praevia/placental abruption.
Our literature review 23 found that PTB tend to be underdiagnosed and underreported.However, their measurement improved in 2011, when gestational age began to be collected in exact weeks and not grouped into categories 24 .With rates 50% higher in 2011, but flat in prior years, we adjusted pre-

Figure 1
Relationship of the components for each maternal and child health outcome (MCHO).
Cad. Saúde Pública 2017; 33 (12):e00207416 2011 PTB estimates upward by 50%, which served as our lower bound.Since 2011 estimates were underestimated by at least 15% 24 , we adjusted all years by an additional 15% as our midrange estimate and by 30% as our upper bound.We also adjusted SIDS estimates upward by 100% for midrange and by 200% for the upper bound 23 .Rates after 2013 for all MCHOs were assumed to remain at their 2013 levels (cf.Supplemental Table 1; http://www2.inca.gov.br/wps/wcm/connect/observatorio_controle_tabaco/site/home/biblioteca).
Our review 23 also indicated that the relative risks in Brazil were consistent with estimates for the United States, although at the low end in some cases.Consequently, the relative risks and the lower/ upper bounds of smoking-attributable MCHOs for Brazil are based on US risks, which are based on reviews and recent studies with large samples for each outcome (cf.Supplemental Table 2; http:// www2.inca.gov.br/wps/wcm/connect/observatorio_controle_tabaco/site/home/biblioteca) 14,23,25.
To estimate prenatal smoking prevalence, we assumed that the prevalence of smoking among pregnant women mirrors that of women of the same age in the general population, and that a fixed percentage of pregnant women who smoke quit during their pregnancy.Based on data from two earlier studies 26,27 , and previous analyses 23 , we estimate that 40% of women quit smoking during pregnancy.We also considered a lower bound where 30% quit at all ages, since the higher rate may reflect policies targeted at pregnant smokers implemented only after 2000 3 .Based on evidence from the 2013 Brazilian GATS 19 , we also estimated an upper bound where 50% of women smokers quit during pregnancy.These data were also used to distinguish prenatal smoking rates by age.Based on a study for Spain 28 , we assumed a 15% underreport of prenatal smoking, within the range of other countries 29 .

The effect of policies
Policies are assumed to have the same effect sizes on prenatal smoking as on the female population, based on previous studies.For example, demand studies 30,31,32,33,34 have obtained prenatal prevalence price elasticities between -0.13 and -0.7, which are consistent with the elasticities used in SimSmoke, and a recent study 34 found higher elasticities for those of low socioeconomic status.Studies have found that other tobacco control policies, including cessation treatment 35,36 , smoke-free air laws 26 , and media campaigns 37 , are effective at reducing prenatal smoking.
To estimate the effect of the tobacco control policies implemented since 1989, all 2015 policies were set to their 1989 levels for all years.This counterfactual represents predicted smoking rates if there were no changes in tobacco control policies in Brazil, i.e., the long-term trend in the absence of policy change.For smoking prevalence, we considered the effect of all policies regarding the counterfactual rate in the same year.For SADs and each adverse MCHO, we calculated the net gain by subtracting the number of outcomes with policies implemented from the number of outcomes under the counterfactual in the same year.
Since tax policies have been found to be particularly effective in reducing initiation among young people and increasing cessation among underprivileged ones 38 , we assessed the independent effect of price increases.This point is particularly important for Brazil, because, between 1986 and 2015, the country had multiple tobacco excise tax policies, going from a single rate ad valorem system in the 1990s to a mixed system composed of two specific rates and one small ad valorem component since 2012 39 .It is worth noting that, between 2008 and 2014, consumer prices increased 146%, and much of this price expansion was possible due to an increase in the specific rate over the accumulated inflation rate in 2009 and the new tax structure implemented in 2012 9,39 .

Validation and update of SimSmoke from 2011 to 2015
As shown in Table 1, SimSmoke predicted the overall relative decrease in smoking prevalence rates between 1989 and 2013 within 1% and 4% accuracy for males and females, respectively, but it was less accurate for the subperiod 2008-2013.The model made worse predictions for females by age   group in the 2008-2013 subperiod, underpredicting the decline for the age groups 18-44 and 65+ and overpredicting the decline for women aged 45-64 years old.
In Table 2, Brazil SimSmoke uses the counterfactual of no policies implemented since 1989 to show the effect of policies implemented between 1989 and 2015, including stronger smoke-free air laws and higher cigarette taxes implemented in the last five years.In this counterfactual scenario, smoking prevalence is projected to have fallen from 43.3% to 31.3% among males and from 27.1% to 19.2% among females between 1989 and 2050.Compared to policies set to their 1989 levels, the policies implemented between 1989 and 2015 indicate a net 62% (56.9%-67.1%)and 66.7% (51.0%-78.7%)relative reduction in the smoking prevalence in Brazil by 2050 among males and females, respectively.If only the tax policy was implemented in Brazil since 1989, the net relative reduction in the smoking prevalence by 2050 would have been of 40.3% (32.0%-47.6%)for males and 40.6% (32.3%-47.9%)for females.
With current policies in place, cumulative smoking-attributable deaths are estimated at 594,017 (535,194-650,881) among men and 239,048 (160,549-321,842) among women by 2015 (Table 3).This number increases to 5.0 million (4.6-5.4)among men and 2.5 million (1.8-3.1)among women by 2050.Compared to the counterfactual of no policies, Brazil SimSmoke estimates a cumulative total of 7.5 million (6.4-8.5)fewer deaths (male and female) by 2050, as a result of the policies implemented between 1989 and 2050.Tax policy alone would have accounted for about 4.2 million (3.3-5.0)fewer deaths between 1989 and 2050 (male and female).

Discussion
In previous study 1 , we showed that tobacco control policies were responsible for a 46% decrease in smoking prevalence between 1989 and 2010.Smoking prevalence in Brazil has fallen by almost 20% between 2011 and 2015.SimSmoke projected that 7.5 million deaths will be averted by 2050, including 500,000 cumulative deaths from policies implemented between 2011 and 2015 1 .This translates to a 46% reduction in smoking-attributable deaths between 2016 and 2050.
Brazil SimSmoke was also extended to consider the effect of prenatal smoking on smokingattributable MCHOs.The model projected that 0.9 million (0.4-2.4) MCHOs will be averted by 2050.Brazil was chosen for this analysis because it had high rates of smoking at population level, but has witnessed marked reductions in smoking rates 1,2,3,4 .Our results indicate that, between 2016 and 2050, the already implemented policies will reduce the total number (net absolute reduction) of placental abruption/placenta praevia by 7.4%; PTB by 3.1%; LBW births by 8.3%; and SIDS by 9.2% (data not shown in a table).
Brazil has achieved the Millennium Development Goal of reducing under-five child mortality by two thirds between 1990 and 2015 40 .Our results indicate that tobacco control policies may have played a major role.The case for tobacco control becomes more compelling upon extending the analyses to consider the effects of second-hand smoke and/or other MCHOs, such as ectopic pregnancy and cleft palate 12 .In addition, the costs of tobacco consumption pose a heavy burden on governments and society 10 .In 2011, smoking-attributable direct medical expenditures for cardiovascular, cancer, chronic obstructive pulmonary and perinatal diseases cost USD 15 billion in Brazil 41 .Continuing to increase cigarette prices will increase tax revenues and reduce smoking prevalence, particularly among underprivileged people, thus saving funds for other health care challenges 38 .

Limitations
The previously validated model 1 was updated in 2015 and made good predictions for the overall adult female and male smoking prevalence.The updated model made fewer good predictions for females by age group for the subperiod 2008-2013, which may reflect limitations in the model or in the data used to validate SimSmoke.Specifically, the model underpredicted the smoking rate reduction for those aged 18-44 yearsold, i.e., an age group closely linked to child-bearing potential.Consistent with the literature 1 , SimSmoke assigns a small effect size to the impact of warning labels on young people, who are often thought to ignore warnings.However, mass media campaigns and the new set of stronger health warnings related to both miscarriage and passive smoking exposure, implemented since 2009 6,7,8,9,42 , may have produced greater effects than predicted.Price and tax policy may also have had greater effects on younger females than predicted by the model, due to their lower income than males 8,9,38,39,40,41,42,43 , thus suggesting that our estimates probably underestimated the decrease in smoking prevalence from implementing chosen policies among women of child-bearing potential and, as a consequence, probably underestimated the cumulative number of averted MCHOs.
The results of the model are based on the assumptions inherent to it, the policy effect sizes, and the data quality 1,44 .Additional limitations are related to the estimates of the smoking-attributable MCHOs: (i) while the estimates of the relative risks of smoking for LBW babies, PTB, and SIDS for Brazil are generally consistent with estimates for the US 14,23 , we found no Brazilian studies that considered placental abruption or placenta praevia; (ii) while smoking behaviors of Brazilian mothers appear to be similar to those of high income countries, Brazilian pregnant women may smoke fewer cigarettes and be less nicotine dependent 4,14 , which may lower relative risks and the SAF measure used, thus suggesting that we may have overestimated the effect of policies on the cumulative number of averted MCHOs; (iii) in examining prenatal smoking over time, differences in smoking behaviors by socioeconomic status were not considered, which have become increasingly important in Brazil 1,2,3,4 ; for instance, smoking prevalence is usually higher among less educated women (versus high Cad.Saúde Pública 2017; 33(12):e00207416 educated), and they quit less during pregnancy 4,45 .As a consequence, although we also considered a lower quitting rate during pregnancy to increase the SAF measure used and to estimate the upper bounds for averted MCHOs, we still may have underestimated the number of averted MCHOs; (iv) although we could not consider the frequency of maternal smoking in successive pregnancies and its association with repetition of the selected MCHOs, smoking persistence may be an important risk factor to further increase the respective relative risks and, as a result, the cumulative number of averted outcomes 46 ; (v) although estimates of prenatal smoking prevalence were corrected for underreporting 28 , underreporting may also reduce the strength of the SAF measure used 29 and, therefore, underestimate the total number of averted MCHOs; (vi) wide variations in MCHOs exist by regions of Brazil, and they were not considered 23,24,47 ; (vii) Brazil SimSmoke does not directly incorporate changes in policies that target pregnant smokers; (viii) we did not consider the fact that pregnancy interruptions -by either induction or caesarean section -have become much more frequent in Brazil in the last years 17,48,49 .Although the relative risks of adverse MCHOs from Brazilian studies were within the range of the lower bounds of relative risks used to estimate smoking-attributable MCHOs for Brazil 23 , Brazilian pregnant women who do not smoke may be increasingly at higher risk for MCHOs than those from high income countries, which may further lower respective relative risks, thus suggesting that we may have overestimated the effect of policies on MCHOs.

Conclusion
The WHO-FCTC is a clear translation of the goals expressed in the Millennium Declaration regarding maternal and child health 5,10,11 .Our findings show the benefits of tobacco control in reducing both SADs and smoking-attributable MCHOs at population level.Thus, our analysis may better inform policy makers in low and middle income countries (LMICs) about allocating resources toward tobacco control policies in this important area.The analyses of this study are applied to Brazil, but similar results may be expected in other LMICs with strong tobacco control policies.

Contributors
A. S. Szklo participated in the data analysis and as lead author in the elaboration/preparation of the article.Z. Yuan participated in the data processing, data analysis, and elaboration/preparation of the article as co-author, and approved the final manuscript as submitted.D. Levy participated in the data analysis and elaboration/preparation of the article as co-author, and approved the final manuscript as submitted.(6,4-8,5 milhões) de mortes entre adultos com 18 anos ou mais.Estima-se também que as políticas atuais reduzirão um total acumulado de 0,9 milhão (0,4-2,4 milhões) de desfechos adversos de saúde materna e infantil até 2050.Nossos achados demonstram os benefícios do controle do tabagismo na redução da MAT e de danos à saúde materna e infantil atribuíveis ao tabagismo em nível populacional.Esses benefícios podem ser utilizados para informar os gestores nos países de renda baixa e média sobre a alocação de recursos para as políticas de controle do tabagismo.

Table 1
Validation of the Brazil SimSmoke: predictions versus survey estimates, by gender and age group, 2008-2013.

Table 2
Smoking prevalence from ages 18 to 85, Brazil.