Artigos

Introduction
Morbidity and mortality due to violence is a global concern, being one of the public health priorities in many countries.1 Global estimates show a 30% prevalence of women victims of physical and/or sexual violence,2 a 50% prevalence of children and adolescents subjected to violence,3 and a 17.5% prevalence of elderly victims of general abuse.4 In Latin America, violence against children and adolescents exceeds 30%.3 In Brazil, according to the National Health Survey (NHS) in 2019, 19.4% of women suffered some violence, and 16.9% sought out health care.4 In Brazil, underreporting represents a challenge in interpreting records of interpersonal violence.4,5
An intense mobilization to raise awareness of violence against children, adolescents, women and the elderly as a public health problem,6 starting in 2011, shaped a scenario that led to the legal requirement of universal and continuous reporting in the health care network as a surveillance strategy in Brazil.6-9 A representative act on the national public policy agenda, the record of violence led to the expansion of an interpretative framework concerning the weight of the events and the understanding of the phenomenon from a health perspective.9-12
One dimension of the line of care and protection for victims, the institutionalization of the reporting of violence in the Notifiable Diseases Information System (Sistema de Informação de Notificação de Agravos – SINAN) began to reveal cases that had previously been hidden because they did not result in serious physical injuries or deaths, such as verbal and psychological aggression. However, the data represent the portion of cases reported by health professionals and services that adhere to the reporting of violence process,4,5 and are commonly the least socially tolerated injuries.6,10
A comparative study of data from the NHS with SINAN identified a significant underreporting of physical and sexual violence against women, respectively 24.1% and 89.4%.4 Although the current trend is towards an increase in records concerning the reporting of violence, with increasing adherence to health services, an underreporting of events in municipalities can also be observed.6,11,13,14 Between 2011 and 2018, municipalities that carried out at least one notification increased from 38% (n=2,114) to 78.7% (n=4,381), however, 1,189 municipalities did not report cases of violence.5
Evidence shows an association between violent events and deaths in Brazil.12,15 The increase in notifications of violence and homicide are related to firearm events involving adolescents in large cities.12 In turn, the higher the level of violence recorded by the police in the municipality, the greater the risk of homicide.15
Among the dimensions of underreporting of violent injuries, the degree of the implementation of the surveillance of violence articulated in a network of support and care for victims in municipalities is crucial.4,5,7 Spatial approaches in public health make it possible to locate areas with insufficient structure and inequality, along with gaps in surveillance and health care for people who suffer violence,16 thus helping to offer a better understanding of the extent of the problem and provide support for actions to improve reporting. This study aims to identify municipalities that suffer from an underreporting of interpersonal violence based on homicide estimates in Brazil from 2016 to 2018.

Methods
Study design
This work was an ecological study of data focused on the reporting of interpersonal violence and homicide records in Brazilian municipalities in the 2016-2018 triennium. The unit of analysis of the study is the municipality, which is the lowest level of geographic-administrative disaggregation with data available for the production of violence notification rates and homicide mortality rates nationally.
To circumvent limitations resulting from the difference in the quality of homicide death data in the Mortality Information System (Sistema de Informações sobre Mortalidade – SIM) observed in the disaggregation of records by municipalities and regions of Brazil, this study uses correction estimates for under-registration and garbage codes.15,17

Data source
The homicide data is unprecedented homicide estimates from the Global Burden of Disease Study (GBD) for 5,570 Brazilian municipalities, published for the period 2000 to 2018. The unpublished estimates from the Global Burden of Disease Study (GBD) for the 5,570 Brazilian municipalities. The SIM is the main source of mortality estimates published on the GBD-Brasil website: https://gbdbr.com.br/painel-de-dados/, while SINAN's records of interpersonal violence are available on the MS website: http://tabnet.datasus.gov.br/cgi/deftohtm.exe?sinannet/cnv/violebr.def, In addition to population estimates from the Ministry of Health (MS).
The GBD applies complex methods to process previsouly detailed mortality data. The GBD replicated update adjustment methods applied in Brazilian municipalities at the national and state levels. GBD estimates apply missing data redistribution processes for sex, age, and ignored municipalities, considering place and year. Procedures for under-registration corrections and garbage codes redistribution follow the sequence described below.1,20 In this GBD correction process, important steps stand out. The first is the mapping of the basic causes of death according to the GBD list by age (target causes and garbage codes). An assessment is then made of inconsistent causes (e.g. prostate cancer in women) and missing data (no age or sex). The next step redistributes all garbage codes to their respective target causes, the basic root causes. The GBD study applies Cause of Death Ensemble modelling (CODEm) to estimate deaths. Finally, techniques are applied to smooth the data, taking into account any stochastic variation over time, in a process called noise reduction, and underreporting correction factors are applied by sex and age.20

Data, and selection of variables
The object of this study refers to the interpersonal and intentional violent act of physical force or intra- and extra-family power.7 SINAN registers reports of interpersonal and self-inflicted violence. Thus, the selection of events of interpersonal violence occurred by including cases registered with “no” for two variables: self-inflicted injury (no) and injury committed by the individual (no). Homicides are defined in the GBD study according to International Classification of Diseases (ICD-10) codes X85 to Y09 and Y871.
Because this study uses homicide death records as a proxy for notifications of interpersonal violence, the analysis focuses on variables that are common to the SINAN and SIM databases. Although SINAN captures multiple aspects of data on domestic/intra-family and extra-family/community interpersonal violence, some of these aspects are not common and captured in SIM, such as those related to LGBTQI+ populations.7 Since the variables age and sex are present both in the notification of violence and in the death certificate, four analysis subgroups were defined in this study: I. Individuals of both sexes under 20 years of age (children and adolescents); II. Individuals of both sexes aged 60 or over (elderly); III. Women aged 20 to 59; and IV. The total of these subgroups. The data are presented according to the victims’ municipality of residence.
Data analysis
Reporting rates of interpersonal violence are calculated per 10,000 inhabitants, while homicide rates are per 100,000 inhabitants and were standardized using the direct method in each population subgroup analyzed in this study. Both are averages for the 2016-2018 trienniumto minimize fluctuations in small numbers, since 68.5% of all municipalities have less than 20,000 inhabitants.
The criterion defined in this study for underreporting of interpersonal violence in SINAN results from bivariate spatial analysis, whose values correspond to quadrant areas (Q1 to Q4) of spatial autocorrelation of interpersonal violence reporting rates with homicide rates. Thus, underreporting of violence in SINAN refers to low violence reporting values close to neighbors with high homicide values (Q3).
Bivariate spatial analysis
The spatial analysis is preceded by a description of the spatial distribution of reporting rates of interpersonal violence and homicide mortality from 2016 to 2018, as well as the frequency of municipalities that remain silent in reporting violence (zero reports) among those with higher homicide rates (?75% percentile in each subgroup), which therefore represents a greater chance of the underreporting of violence in SINAN.
The presence of spatial clustering was tested using the global Moran Index, with a significance level of 0.05. The bivariate Local Moran Index (LISA) presents the spatial autocorrelation (2nd-order Queen's contiguity) between normalized reporting rates of interpersonal violence and homicide rates for each of the analyzed subgroups from 2016 to 2018.21
According to the results, the municipalities were classified into values from 0 to 4, where 0 is insignificant, and the other values correspond to areas of quadrants Q1 to Q4 of the spatial autocorrelation of reporting rates of interpersonal violence with homicide rates:
1. Q1 (high-high): high values of violence reporting rates in the neighborhood of municipalities with high average homicide rates (red);
2. Q2 (low-low): low values of violence reporting rate in the neighborhood of municipalities with low average homicide rates (dark blue);
3. Q3 (Q3, low-high): low violence reporting rate in the neighborhood of municipalities with high average homicide rates (light blue); and
4. Q4 (Q4, high-low): high reporting of violence rates in the neighborhood of municipalities with low homicide rates (light red).
Q1 and Q2 are areas with positive spatial association (homogeneous), while Q3 and Q4 are areas with negative spatial association (heterogeneous). Municipalities belonging to quadrant Q3 (low reporting with high mortality) were considered critical and priority areas, with a strong indication of underreporting of violence in SINAN, as the magnitude of the incidence of reports is quite inconsistent in relation to the homicide rate. By contrast, Q4 includes municipalities that are considered to have more reliable data concering the reporting of violence.
Spatial analyses were performed using GeoDa 1.2 and QGis 3.16. This study used secondary data, approved by the Research Ethics Committee of the Federal University of Minas Gerais (registered under protocol no. 3,258,076).
Results
The North, Northeast, and Midwest regions of Brazil concentrated 29% of the 591.800 reports of violence and 58% of the 55.900 homicides in the 2016-2018 triennium. Among the elderly (?60 years), the distribution of these events between regions was similar. In children and adolescents (?20 years of age), the Northeast reported 44.6% of the homicides and 16.4% of the reports of violence. among adult women (20 to 59 years of age), the North, Northeast, and Midwest accounted for 53.6% of the homicides and 24.6% of reports of violence (Table 1).
In the 2016-2018 triennium, 17.9% of the municipalities with the lowest violence reporting rates (?0.8/10 thousand) were found mainly in states in the Northeast (17.4%), Midwest (11.6%), and North (11.3%) of the country, as well as the highest homicide rates (?13.7/100 thousand) of 27.9%, 27.4%, and 24.9%, respectively, particularly in the states of Bahia, Pará, Maranhão, and Ceará. Conversely, 22.7% of the municipalities showed the highest reporting rates of interpersonal violence (?17.8/10 thousand), especially in the states of the Southeast (44.7%) and South (27.1%), which also reported the lowest homicide rates (?3.1/100 thousand), with greater emphasis in the states of São Paulo, Minas Gerais, and the South region. This scenario was repeated in the population subgroups analyzed in this study, and proved to be more intense among elderly victims (Figures 1 and 2). In municipalities with higher homicide rates (percentile ? 75%), 31.4% were silent regarding the reporting of interpersonal violence, especially in the Midwest, Northeast, and North regions. The same was repeated in each of the three population subgroups (Tables S1 and S2 in the supplementary material).
The spatial correlation between the reporting rate of interpersonal violence and the homicide rate was negative in all analyzed subgroups (I<20 = -0.083; Iwomen20-59 = -0.023; I?60= -0.086; Itotal = -0.085)). Areas of reports of low violence with high homicide rates stand out (light blue), likely due to an underreporting of violence in SINAN, thus showing that these municipalities are both critical and of high priority (Figure 3).
Among children and adolescents, 899 municipalities were identified in this critical situation, mainly concentrated in the states of the Northeast (71.1%), with the exception of Piauí and Maranhão; the Midwest (58.7%), and the North (57.9%), mainly in areas in Pará and Goiás, in addition to Espírito Santo and Rio de Janeiro. Among the elderly, critical areas ad up to 1,284 municipalities, concentrated in the states of the North (80.5%), Midwest (75.0%), and Northeast (64.1%) regions. Another 670 municipalities make up critical areas in the reporting of female victims of violence, mainly found in the Midwest (82.9%), North (75.7%), and Northeast (43.9%) regions, with emphasis on the states of Goiás, Mato Grosso, Pará, and Roraima, in addition to Ceará, Pernambuco, the coast of Bahia, and the north of Espírito Santo. In total, low reportings in very violent areas occurred in 998 (33.0%) municipalities in all of the clusters, concentrated mostly in the Midwest region (77.8%), especially in the states of Goiás, Mato Grosso, and the Federal District; followed by the North region (75.0%), mainly in the state of Pará; and in the states of the Northeast region (65.8%), with the exception of Maranhão, Piauí, and Paraíba (Figure 3 and Tables 2 and S3).
Areas of a high level of reports of violence with low homicide rates (light red) indicate municipalities with better data consistency, which are located mainly in states in the Southeast and South regions, as well as in the state of Piauí (Figure 3).

Discussion
The reporting of violent incidents and the registration of homicides followed distinct and incompatible spatial distribution patterns in specific areas of the country. Municipalities in the North, Northeast, and Midwest regions simultaneously presented a lower incidence or absence of violence with a higher incidence of homicides, while the opposite occurred in the Southeast and South regions of Brazil. This inverse relationship between the incidence of violence reporting and the incidence of homicide means a negative spatial correlation, where clusters of nearby municipalities presented values in different directions, both in the population of children and adolescents, as well as women and the elderly. The differences between the lower probability of being reported as a victim of violence and the higher probability of dying from homicide were more significant among the elderly.
The characterization of conglomerates, which contradicts the assumption of a positive correlation between the incidences of violence and death,12,15 makes it possible to identify critical municipalities and regions for reporting interpersonal violence committed against children, adolescents, women, and the elderly. The adherence of municipalities in the North, Northeast, and Midwest regions is, therefore, a priority for the management of violence surveillance, especially given the need to implement services in places without notifications or expand actions to reorganize care and record violence in locations with underreporting.5
Victims of violence overcome a series of difficulties to seek help and interrupt the chain of events and circumstances to which they are subjected. After overcoming fear and seeking care in health services, a portion do not have their cases notified, as almost two-thirds of professionals do not adhere to reporting standards.4,22 As injuries are not reported, the purpose of surveilling and reporting violence is not adequately fulfilled, nor do healthcare services in a protection and support network serve as rights guaranteed to people in scenarios of violence.9 Among the violence reported by health services, physical violence was the most commonly reported in the country, while psychological violence was extremely underreported, closely followed by sexual violence.4
Critical areas of municipalities with low reporting rates of violence and high homicide rates were more expressive in the states of the Northeast region, not including Maranhão, Piauí, and Paraíba; Pará in the North region; and Goiás, Mato Grosso, and the Federal District in the Midwest region, as well as among the elderly population. In children and adolescents, these areas can be found in nearly all states in the Northeast, in areas in the states of Pará and Goiás, as well as in Espírito Santo and Rio de Janeiro. For women, critical areas are mainly in Goiás, Mato Grosso, Pará, and Roraima, in addition to Ceará, Pernambuco, coastal Bahia, and northern Espírito Santo. Data on the subnational coverage of the violence reporting system in Brazil point in the same direction.4,5 Despite the solid increase in records since the institutionalization of the surveillance of violence,6,11,13,14 states in these regions presented a lower percentage of reporting municipalities in SINAN.5 As a result, there was a greater underreporting of psychological, sexual, and physical violence. States with the worst reporting rates were Sergipe and Rio Grande do North in psychological violence; Sergipe, Piauí, and Amapá in sexual violence; and Pará, Sergipe, and Ceará in physical violence.
The service's inability to capture cases involves a series of factors. Difficulty identifying cases, especially domestic violence, which are not always addressed openly. Among more serious events, underreporting arises from weaknesses in the institutional support of service organization and professional training,22 such as a fear of legal involvement or retaliation, difficulties in the reporting and case handling processes, and the volume of activities. Furthermore, there is a low level of trust in the government agencies that are supposed to protect the people who report violence, as well as in the failure to recognize violence as a health issue.8,11,12 The health professional’s lack of adherence to set standards also affects the quality of the reported data.13,22,23 The poor recording of other health problems in SINAN suggests common factors of underreporting present in the system,24,25 such as the degree of adequacy of primary health care coverage and the performance of the care network in the country.26,27
In turn, results from this study indicated more robust records of reports of violence in areas of the Southeast and South regions. High rates of reports of violence in areas with low homicide rates were more evident in the states of São Paulo, Minas Gerais, and Piauí, as well as from the South region. Population-based data estimated better reporting rates in physical violence for the states of Rondônia, São Paulo, and Minas Gerais, in sexual violence for the Federal District, Paraíba, and Pará.4
The generation of epidemiological data allows information on the magnitude of violence and supports the development of public policies to combat the problem. Furthermore, the reporting of violent injuries is the initial epidemiological fact that seeks to trigger actions to stop violence, and that goes beyond the limits of the health service by requiring the examination of the network of family and social relationships of the person in a situation of violence in the territory.9
The ecological study gives rise to limitations of a singular ecological bias in the method itself. The analysis considered population subgroups of victims but did not examine the types of violence. The use of reporting and mortality data for the 2016-2018 triennium circumvented the problem of reducing the record of violence in the years of the COVID-19 pandemic in Brazil, and the variability of low numbers in municipalities.28 In addition, the estimation of homicides through the redistribution of garbage codes mitigates the unequal quality of recording of causes of death in the country.15
To other countries with a greater course in compulsory reporting of violence, Brazil has the advantage of adopting a national and standardized reporting system. However, it still presents difficulties in the degree of adherence of municipalities and health professionals to the surveillance of reportable violence, as well as in the organization of health services, which are challenges to the coordinated implementation of reporting.6,8,26,27
This study characterizes regional inequalities, and indicates critical and priority municipalities for the structuring of surveillance and healthcare services. The empirical findings on the relationship between reporting and homicide records can also be used in the development of minimum reporting prediction models based on data clusters from municipalities with information that is considered satisfactory. The estimates would enable the formulation of a specific indicator of the identification and coverage of reports in states and regions of the country. The most reliable portrayal of the degree of institutionalization of the reporting of violence enables management to organize initiatives in order to improve care and promote equity in access to the health system.
Acknowledgments
We authors would like to thank the Institutions Vital Strategies of Brazil and the Brazilian Ministry of Health for funding the studies presented in this article.

References
1. Abbafati C, Abbas KM, Abbasi-Kangevari M, Abd-Allah F, Abdelalim A, Abdollahi M, et al. Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet/Global Health Metrics. 2020; 396 (10258): 1204-22.
3. Violence against women Prevalence Estimates, 2018. Global, regional and national prevalence estimates for intimate partner violence against women and global and regional prevalence estimates for non-partner sexual violence against women. WHO: Geneva, 2021 [cited 2024 April 09]. Available from: https://www.who.int/publications/i/item/9789240022256.
3. Hillis S, Mercy J, Amobi A, Kress H. Global Prevalence of Past-year Violence Against Children: A Systematic Review and Minimum Estimates. Pediatrics. 2016; 137 (3): e20154079. https://doi.org/10.1542/peds.2015-4079.
4. Yon Y, Mikton CR, Gassoumis ZD, Wilber KH. Elder abuse prevalence in community settings: a systematic review and meta-analysis. Lancet Glob Health. 2017; 5(2): e147-e156. https://doi.org/10.1093/eurpub/cky093.
4. Vasconcelos NM, Bernal RTI, Souza J, Bordoni PHC, Stein C, Coll CVN, Murray J, Malta DC. Subnotificação de violência contra as mulheres: uma análise de duas fontes de dados. Cien Saude Colet (2023/Set) [cited 2024 March 24]. Available from: http://cienciaesaudecoletiva.com.br/artigos/subnotificacao-de-violencia-contra-as-mulheres-uma-analise-de-duas-fontes-de-dados/18899;
5. Brasil. Ministério da Saúde. Vigilância de violências e acidentes no Brasil: análise da cobertura da notificação compulsória de violência interpessoal/autoprovocada nos municípios brasileiros. Bol Epidemiol. 2020 [cited 2024 March 24]; 51 (4): 11-7. Available from: https://www.gov.br/saude/pt-br/centrais-de-conteudo/publicacoes/boletins/epidemiologicos/edicoes/2020/boletim-epidemiologico-vol-51-no-04.pdf/view.
6. Minayo MC de S, Souza ER de, Silva MMA da, Assis SG de. Institutionalizing the theme of violence within Brazil’s national health system: progress and challenges. Ciênc saúde coletiva. 2018; 23(6):2007–16.
7. Brasil. Ministério da Saúde. Viva: instrutivo notificação de violência interpessoal e autoprovocada [Internet]. Brasília: Ministério da Saúde; 2016 [cited 2024 April 06]; 92 p. Available from: https://bvsms.saude.gov.br/bvs/publicacoes/viva_instrutivo_violencia_interpessoal_autoprovocada_2ed.pdf.
8. Lima JS, Deslandes SF. Olhar da gestão sobre a implantação da ficha de notificação da violência doméstica, sexual e/outras violências em uma metrópole do Brasil. Saude soc. 2015; 24 (2): 661–73.
9. Brasil. Ministério da Saúde. Vigilância de Violências e Acidentes (Viva): 2013 e 2014. Brasília: Ministério da Saúde, 2017 [cited 2024 April 06 ]. Available from: https://bvsms.saude.gov.br/bvs/publicacoes/viva_vigilancia_violencia_acidentes_2013_2014.pdf.
10. Veloso MMX, Magalhães CMC, Dell'Aglio DD, Cabral IR, Gomes MM. Notificação da violência como estratégia de vigilância em saúde: perfil de uma metrópole do Brasil. Ciênc saúde coletiva. 2013;18 (5): 1263–72.
11. Marinho Neto KRE, Girianelli VR. Evolução da notificação de violência contra mulher no município de São Paulo, 2008-2015. Cad Saúde Colet, 2020; 28(4), 488–99.
12. Pinto IV, Ribeiro AP, Santos AP, Bevilacqua P, Lachtim SAF, Pereira VOM, Malta DC. Wounded adolescences: a portrait of firearm violence reported in Brazil. Rev Bras Epidemiol, 2020; 23, e200002.SUPL.1.
13. Lima VMF, Stochero L, Azeredo CM, Moraes CL, Hasselmann MH, Marques ES. Characterization and completeness of notification sheet of violence against the older adults in Niterói-RJ, 2011-2020. Epidemiol. Serv. Saúde. 2023; 32 (1): e2022451.
14. Arruda GT, Kocourek S, Oliveira JL. Violência contra o idoso no Rio Grande do Sul, Brasil: análise das notificações de 2009 a 2016. Revista Kairós: Gerontologia. 2018; 21 (3): 181-92.
15. Soares Filho AM, Vasconcelos CH, Cunningham M, Ribeiro ALP, Naghavi M, Malta DC. Spatial association of homicide rate with violence, sociodemographic, and public security factors: global burden of disease study 2018 for municipalities in Brazil. Public Health. 2024; 227:16-23.
16. Nucci LB, Souccar PT, Castilho SD. Spatial data analysis and the use of maps in scientific health articles. Rev. Assoc. Med. Bras. 2016; 62 (4): 336-41.
17. Soares Filho AM, Vasconcelos CH, Nóbrega AA da, Pinto IV, Merchan-Hamann E, Ishitani LH, et al. Improvement of the unspecified external causes classification based on the investigation of death in Brazil in 2017. Rev bras epidemiol. 2019;22:e190011.supl.3. https://doi.org/10.1590/1980-549720190011.supl.3
20. Johnson, S.C., Cunningham, M., Dippenaar, I.N. et al. Public health utility of cause of death data: applying empirical algorithms to improve data quality. BMC Med Inform Decis Mak 21, 175 (2021).
21. Anselin L. Bivariate, Differential and EB Rate Moran Scatter Plot. Global Spatial Autocorrelation [cited 2024 April 09]. Available from: https://geodacenter.github.io/.
22. Souza EG de, Tavares R, Lopes JG, Magalhães MAN, Melo EM de. Attitudes and opinions of professionals involved in the care to women in violence situation in 10 Brazilian cities. Saúde debate. 2018; 42:13–29.
23. Girianelli VR, Ferreira AP, Vianna MB, Teles N, Erthal RM de C, Oliveira MHB de. Qualidade das notificações de violências interpessoal e autoprovocada no Estado do Rio de Janeiro, Brasil, 2009-2016. Cad saúde colet. 2018; 26 (3): 318–26.
24. Bartholomay P, Pinheiro RS, Johansen FDC, Oliveira SB de, Rocha MS, Pelissari DM, et al.. Lacunas na vigilância da tuberculose drogarresistente: relacionando sistemas de informação do Brasil. Cad Saúde Pública. 2020; 36 (5): e00082219.
25. Rodrigues AB, Santana VS. Acidentes de trabalho fatais em Palmas, Tocantins, Brasil: oportunidades perdidas de informação. Rev bras saúde ocup. 2019; 44:e8.
26. Soares Filho AM, Vasconcelos CH, Dias AC, Souza ACC, Merchan-Hamann E, Silva MRF. Primary Health Care in Northern and Northeastern Brazil: mapping team distribution disparities. Ciênc saúde coletiva. 2022; 27 (1): 377–86.
27. Kashiwakura HK, Gonçalves AO, Azevedo RR, Nunes A, Silva CAT. A portrait of Brazilian primary care: municipal expenditure and infrastructure in Brazilian municipalities. Ciênc saúde coletiva. 2021; 26 (suppl 2): 3397-3408.
28. Platt VB, Guedert JM, Coelho EBS. Violence against children and adolescents: notification and alert in times of pandemic. Rev paul pediatr. 2021; 39:e2020267.