Using Data Mining Techniques to Understand Patterns of Suicide and

Reattempt Rates in Southern Brazil

Caibe A. Pereira

, Rômulo C. R. Peixoto

, Manuella P. Kaster

, Mateus Grellert

and Jônata T. Carvalho

Biochemistry Department, Federal University of Santa Catarina, Florianopolis, Brazil

Informatics and Statistics Department, Federal University of Santa Catarina, Florianopolis, Brazil

Institute of Informatics, Federal University of Rio Grande do Sul, Porto Alegre, Brazil

Keywords: SUS, Health Infrastructure, Machine Learning.

Abstract: Suicide is a multifactorial, complex condition and one of the leading global causes of death, with suicide

attempt as the main risk factor. To this day, studies have shown relevant indicators that help identify people

with risk of committing suicide, but the literature still lacks comprehensive studies that evaluate how different

risk factors interact and ultimately affects the suicide risk. In this paper, we aimed to identify patterns in data

from the Brazilian Unified Health System – SUS, from 2009 to 2020, of individual reports of suicide attempts

and suicide deaths in the Brazilian Southern States, integrating those with a database of the healthcare

infrastructure. We framed the problem as a classification task for each micro-region to predict suicide and

reattempt rate as low, moderate, or high. We developed a pipeline for integrating, cleaning, and selecting the

data, and trained and compared three machine learning models: Decision Tree, Random Forest, and XGBoost,

with approximately 97% accuracy. The most important features for predicting suicide rates were the number

of mental health units and clinics, and for both suicide and reattempts were the number of physicians and

nurses available. This novel result brings valuable knowledge on possible directions for governmental

investments in order to reduce suicide rates.

1 INTRODUCTION

Suicide is one of the leading causes of death, with

approximately 700,000 deaths annually. Suicides are

preventable, and prevention programs can mediate

this problem. To be more effective, the programs

must be directed to the risk populations (WHO,

2014). Suicide attempts are one of the main risk

factors, and each additional attempt increases the risk.

Moreover, detecting the risk of suicide is an open

challenge in mental health research (Gao et al., 2015).

The stress-diathesis model for suicidal behavior

associates biological and psychiatric traits with

environmental stressors (van Heeringen & Mann,

2014), increasing the complexity of this condition.

The environment has a significant contribution to

https://orcid.org/0000-0002-9719-2832

https://orcid.org/0000-0001-6319-9668

https://orcid.org/0000-0003-0258-6204

https://orcid.org/0000-0003-0600-7054

https://orcid.org/0000-0001-9020-2076

suicidal behavior during the spectrum from suicide

ideation to completion (Turecki et al., 2019).

Environmental factors, such as biological (e.g., noise

and pollution) and social stressors (e.g., crime and

harassment fear) and infrastructure (e.g., green areas,

traffic) influence mental health (Johnson et al., 2023).

The studies that analyze suicide behavior or

deaths usually use sociodemographic or clinical

features, and not the environment context of the

individual. Thus, in this study, we propose a data

mining and machine learning workflow to identify

patterns of healthcare infrastructure to predict suicide

reattempt rate (SAR) and suicide rate (SR). We used

aggregated data from individuals who attempted or

committed suicide and healthcare infrastructure in the

Southern region of Brazil to predict rates.

Pereira, C., Peixoto, R., Kaster, M., Grellert, M. and Carvalho, J.

Using Data Mining Techniques to Understand Patterns of Suicide and Reattempt Rates in Southern Brazil.

DOI: 10.5220/0012350500003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 2, pages 385-392

ISBN: 978-989-758-688-0; ISSN: 2184-4305

385

Data were collected from the Brazilian Mortality

Information System (SIM) and Notifiable Diseases

Information System (SINAN), part of the Unified

Health System platform (DATASUS). In this paper,

we selected a case study in the southern states of

Brazil: Paraná (PR), Santa Catarina (SC), and Rio

Grande do Sul (RS). The SIM form is performed post-

mortem; thus, it is more prone to missing values,

especially if the deceased relatives are not present.

This paper is organized as follows: Section 2

presents an overview of suicide, with a focus on

Brazilian data and machine learning studies of

suicide. Section 3 and 4 specifies the data collection,

treatment, and analysis for this study, and the

machine learning models used. Section 5 presents a

discussion of the results and concludes the study. To

the best of our knowledge, this is the first work that

proposes a learning-based solution for suicide or

attempted suicide in Brazil using large-scale public

health data. We show here how important nurses’ and

physicians’ availability and basic and mental health

facilities are to classify suicide risk.

2 BACKGROUND

In Brazil, suicide attempt is a compulsory notification

and every attempt is registered in the SINAN. The

notification started to be compulsory in 2011, but in a

gradual manner; thus, we chose more recent years to

analyze. Suicide attempt notification occurs in the

healthcare unit where the victim was admitted.

SINAN form is used for this purpose. It is an

individual form divided into ten domains, comprising

sociodemographic data about the individual with

details about the attempt (Ministry of Health, 2021).

All Brazilian healthcare-related infrastructure is

sent to the National Registry of Healthcare

Establishments (CNES) database, with different

categories. CNES is monthly updated and we

collected data from the southern states between 2009

and 2020. We gathered information on types of

establishments, professional specialization, mental

health services.

2.1 Suicide Data in Brazil

Brazil is a continental country divided into five

macro-regions (North, Northeast, Central-West,

Southeast, and South). SR and SAR differ between

regions, with the highest rates in the southern region.

Suicide in Brazil follow global trends, such as sex

ratio or methods (Ministry of Health, 2021). Registry

of suicide deaths in Brazil is compulsory and

classified according to the International

Classification of Diseases (ICD)-10 since 1996. Data

are publicly available in the DATASUS. The system

provides information on the death circumstance,

location, and personal information.

2.2 Related Work

2.2.1 Brazilian Studies

The studies that evaluated Brazilian SR and SAR data

were exploratory, analyzing the profile of the

individuals but with no predictive power. For SR,

data was collected from DATASUS; SAR data was

collected from different databases. In the Ministry of

Health (2021) report, from 2010 to 2019 SR showed

an increase of 42% in Brazil. The Southern was the

region with the highest SR. Moreover, attempt at

younger ages was more prevalent and mainly

occurred at home by self-poisoning.

Other studies focus on different aspects of suicide

in Brazil. From 2000 to 2017, Brazilian SR increased

significantly but differently by sex: male SR

increased by 75% and female by 85% (Palma et al.,

2021). Male-to-female ratio is even higher in older

people, reaching 8.2-fold more male deaths (Martini

et al., 2019). In adolescents, SR also increased over

the years and was associated with social inequality

and unemployment (Jaen-Varas et al., 2019).

In SC, a comparison between suicide attempters

and non-suicidal self-injured adolescents and adults

from 2014 to 2018 shows common traits between

groups, such as the majority of females, White

people, people with psychiatric disorders, and

poisoning as the method used. Comparing the

recurrence of the attempt, adults attempt more than

adolescents and with a higher percentage of alcohol

intoxication during the attempt (Pinheiro et al., 2021).

2.2.2 Machine Learning and Suicide

Electronic records were used to assess medical

databases and predict suicide attempts from self-

injured patients. The prediction was performed at

different time points using the Random Forest (RF)

model, with prediction accuracy decreasing with time

(Walsh et al., 2017). RF was used to predict attempts

in males and females with cocaine abuse; for females,

psychiatric issues showed more importance, whereas

for males, it was drug use (Roglio et al., 2020).

Different machine learning tests were used to

predict suicide in patients with depressive disorder.

Attempt, ideation, race, religion, and depressive

severity were the main factors for an attempt (Nordin

HEALTHINF 2024 - 17th International Conference on Health Informatics

386

et al., 2021). In addition, socioeconomic variables

(e.g., income, age, drinking, education) were the main

variables to predict suicide attempt and ideation by

XGBoost (XGB), support vector machine, or logistic

regression models (Choi et al., 2021).

Men and women differed regarding risk factors

with physical health more important for men. For

both sexes, psychiatric disorders and medication, 48

months previous to death, were important risk factors

(Gradus et al., 2020). Deep neural networks were

used to stratify the risk of suicide attempt with clinical

and demographic indicators. The model was able to

separate the sample into four categories, from low to

very high risk of attempt (Zheng et al., 2020).

McCullough et al. (2023) constructed an interaction

network of patients admitted to the hospital for

suicide or self-harm risk, indicating the trajectories

that led to referral to the emergency department.

Different types of data (sociodemographic,

clinical) and populations (countries/cultures or

diagnoses), can affect suicide; thus, analyzing data

from various conditions can increase our knowledge

about the risk factors and assist preventive measures.

Table 1 presents the studies that analyzed suicide

data. The studies focused on demographic or clinical

data. Our work uses demographic data to profile the

individuals attempting and committing suicide and

we aimed to identify patterns within the city's

healthcare infrastructure data to predict suicide and

suicide reattempt rates in Southern Brazil.

3 METHODS

The next sections describe the methods for data

collection, the analysis of the suicide rates dataset,

features and the models used for the predictions, and

our results. Figure 1 shows the workflow of the study.

The codes for the methods and results are deposited

in the following Github repository:

https://github.com/anonimo-SBBD/SuicideRates.git

3.1 Data Collection and Pre-Processing

DATASUS provides curated data for SIM, SINAN,

and CNES data sets. To access the data from

DATASUS, we used the 'download' function of the

PySUS library (Coelho et al., 2021). We downloaded

the data from the SIM from 2009 to 2020. For

SINAN, we used data from 2016 to 2020. The files

were filtered according to the southern states and the

cause of death (as suicide) for SIM. For SINAN, we

filtered the observations in which the ‘self-inflicted’

and the type of violence ‘other’ columns

corresponded to ‘yes’ (Pinheiro et al., 2021). We

filtered the data based on the ‘violence specification’

column, using different string values that indicated

suicide attempt (“suic/ts/t.s/intox/enve/enfor”

referring to Portuguese spellings).

We chose the southern region as a case study

because the region has the highest SR and SAR and

similar socioeconomic and cultural aspects between

states. We performed an initial separated analysis of

the three states but observed that the analyses were

similar and decided on a single, unified analysis to

increase sample size.

We removed columns not relevant, redundant,

with more than 30% of missing values, or with an

unidentifiable meaning. The following variables

remained: age, sex, race/color, marital status,

schooling, city, and date of occurrence for suicide

completion plus the presence of disabilities or

disorders and the method used in the attempt. We

excluded columns with specific values or columns

with values equivalent to the outcome.

We add new variables based on the date and age:

month, year, season, day of the week, and range of

age (in 10 years, starting with 10 years old). Some

variables with missing data were converted to

"ignored value" (i.e., a value was not imputed while

filling the form) because this an available option in

SIM and SINAN forms. The entries with missing

values in the "age" or "date" variables were excluded.

We used these data sets in the exploratory analysis. In

the CNES data set, we use the columns for public

professional, category of the professional (43

categories), type of health establishment (33 types),

and mental health specialization.

Since in SIM, SINAN, and CNES data sets

variables were mostly categorical, we separated each

category using the One-hot encoding method. Date

and city code were used to group and add the values

with information on the number of entries for each

category presented and merged SIM or SINAN to

CNES data sets into a single dataframe (DF). For the

machine learning predictions, we calculated the SR of

each city, in the respective year and month, by

100,000 inhabitants and added a column with a

classification of the SR rate based on WHO

recommendation: low SR ≤ 5; 5 ≤ moderate SR ≤ 10;

high SR > 10 (WHO, 2014). For SAR, we classified

based on the percentage of reattempt rate: low SAR ≤

33%; 33% < moderate SAR ≤ 66%; high SAR > 66%

of reattempt considering all the attempts. We also add

a column with the city's estimated population and

demographic density for the models not to compute

the rates based on the population number.

Using Data Mining Techniques to Understand Patterns of Suicide and Reattempt Rates in Southern Brazil

387

Table 1: Comparison between studies analyzing suicide data with machine learning approaches. Model results show the

representative of the best model the study performed (if more than one model). AUC, area under the curve; Acc, accuracy;

Sens, sensitivity; Spec, specificity; PPV, positive predictive value; NPV, negative predictive value; SIM, Mortality

Information System; SINAN, Notification of Interpersonal and Self-Inflicted Violence; CNES, National Registry of

Healthcare Establishments.

Study Database Time range Analysis Strategy Models results Country

McCullough

et al., 2023

Clinical interactions of

patients in the Hospital

Emergency Department

Classify the interaction

that led to referral to the

Emergency Department

Bernoulli naive Bayes

classifier

Acc.= 0.82

Australia

Gradus et

al., 2019

Clinical and

demographic data

1995 - 2015

Suicide prediction and

risk factor analysis

Random Forest

Men – AUC=0.80,

Spec.=0.97

Women – AUC=0.88,

Spec.=0.96

Denmark

Zheng et al.,

2020

Clinical and

demographic records

2015 - 2017

Attempt prediction and

risk factor analysis

Deep neural network

AUC=0.77, PPV=0.10,

relative risk=59.02

United

States of

America

Walsh et al.,

2017

Synthetic derivative -

Clinical and

demographic data of

revious attem

1 week to 2

years

Recurrence prediction

and risk factor analysis

Random Forests

AUC=0.84, precision=0.79,

recall=0.95

United

States of

America

Roglio et

al., 2020

Clinical and

demographic hospital

data - cocaine abuse

patients

2012 - 2018

Attempt prediction and

risk factor analysis

Random Forest

Men - AUC=0.68,

Acc.=0.66, Sens.=0.82,

Spec.=0.50, PPV=0.47,

NPV=0.84

Women - AUC=0.73,

other=0.71

Brazil

Nordin et

al., 2021

Clinical and

demographic hospital

data - depressive

disorder patients

Attempt prediction and

risk factor analysis

Bagging Decision tree

AUC=0.87, Acc.=0.92,

Sens.=0.92, Spec.=0.53,

PPV=0.89, NPV=0.76

Malasya

Choi et al.,

2021

KNHANES dataset -

suicide attempts and

ideation

2007 - 2019

Attempt prediction and

risk factor analysis

XGBoost

AUC=0.99, Acc.=0,99,

precision=0.98, recall=0.99,

F1-score=0.98

South Korea

This study

Demographic and

healthcare

infrastructure

2009 - 2020

Recurrence attempt rate

and suicide rate

prediction

XGBoost

Acc.=0.98, precision=0.97,

Sens./Spec.=0.94

Brazil

Figure 1: Workflow from data collection to outcome classification.

HEALTHINF 2024 - 17th International Conference on Health Informatics

388

3.2 Machine Learning Models and

Training Process

To classify SR or SAR, we used the SIM/CNES and

SINAN/CNES merged DF. We divided the DF into

two: train set (80%, SR - 17,007, SAR - 8,292 entries)

and test set (20%); we set the same random state and

divided the main outcome (SR or SAR). The models

were tested with 10-fold cross-validation.

We tested three tree-based machine learning

models: Decision Tree (DT), RF, and XGB. The

models were chosen based on simplicity to execute,

and analyze the results, do not require many data

treatments, such as scaling, and perform well with

categorical and numerical data. Feature importance

was computed with the mean accumulation of

impurity decrease within each tree.

4 RESULTS

4.1 Exploratory Data Analysis

4.1.1 Suicide Attempt

The mean age of the attempters was 30.7 years, and

the absolute values decreased with the increase of age

(27% from 10 to 19 decreasing to 4% from

individuals over 60 years, approximately). For

schooling, middle and high school education (26.5%

and 32.0%) were the most represented (25.30% of the

data were missing). More than 1/3 of the attempters

had a disorder or disability (37.4%), and the most

common were psychiatric (23.6%) and mood (12.9%)

disorders. The main methods chosen for the attempt

were intoxication (67.8%), cutting/stabbing (9.5%),

and hanging (7.8%). Moreover, 18% of males and 8%

of females were intoxicated with alcohol during the

attempt. Interestingly, 38.5% of the individuals had

already attempted suicide, whereas 41.7% were first-

time attempters (19.8% of data was missing).

4.1.2 Suicide Completion

The three states had a cumulative rate of 6.02 (PR),

8.31 (SC), and 10.32 (RS) per 100,000 inhabitants.

From 2009 to 2020 SR increased: PR 41.00%, SC

51%, and RS 27.67%. Males died 3.85-fold more than

females and the mean age was 45 years old; most

deaths occurred between ages 20 to 59 years, and

most were White (87.38%). Schooling was divided by

years studied: from four to seven years corresponded

to 26.76%, and from eight to eleven years

corresponded to 24.20% (24.42% of the data was

missing). Most deaths occurred in the person's

residence (66.45%) and were distributed similarly

throughout the year.

4.2 Machine Learning Classification

Generalization of the models presented high

accuracy, approximately 0.95. Table 2 presents the

best parameters of the models, the evaluation metrics,

and the variables most important to the models to

predict SR and SAR. Low, moderate, and high classes

were imbalanced: SR - 58,67%, 17,87%, and 23,46%;

SAR - 15,67%, 76,60%, and 7,73%.

5 DISCUSSION

Combat of suicide deaths is a challenge to countries

around the globe. Low- and middle-income countries

are at particular risk, with almost 80% of global

deaths. In some of those countries, death rate has been

increasing in the past years, such as Brazil (WHO,

2021). Suicide deaths are preventable, and the World

Health Organization general guidelines suggest

restricting means (firearm and medications control)

and increasing awareness for mental health (WHO,

2014). Moreover, understanding regional specificity

helps developing more effective public policies.

From an economic perspective, suicide deaths

may cost up to 3 million dollars to the government in

direct and indirect costs per death (Kinchin & Doran,

2018; Shepard et al., 2016). Moreover, in Brazil, the

average cost of a suicide attempt is approximately

US$ 7,000 (Sgobin et al., 2015). Many costs occur

after the attempt and include years lost due to trauma,

post-treatment, and impact on work and life of family

and acquaintances (Kinchin & Doran, 2018).

Observing that many attempts occur between ages

10 and 20, we recommend the development of more

programs of mental health and suicide prevention

presented during school years. Indeed, school-based

preventive programs targeted at children and

adolescents can be cost-effective (Ahern et al., 2018).

Our analysis of the profile of individuals

committing suicide corroborates previous findings,

suggesting that preventive programs must target

specific populations. Suicide attempts are the main

risk factors, and many suicide completers do not die

on the first attempt. Non-pharmacological approaches

after the attempt, such as active contact and follow-

up and cognitive behavioral therapy, may be used

along with prevention programs to increase cost-

effectiveness and decrease SR and SAR (Martínez-

Alés et al., 2021; Ross et al., 2021).

Using Data Mining Techniques to Understand Patterns of Suicide and Reattempt Rates in Southern Brazil

389

Table 2: Comparison between machine learning models results and most important features used by the models to classify

reattempt and suicide rates in Southern Brazil. Important features specific to reattempt (*) or completion (

) rates (features

are in no particular order).

Model Parameters Results Important Features

Decision Tree ccp alpha = 0.005

Accuracy = 0.95

Precision = 0.94

Sensitivity = 0.95

Specificity = 0.91

Number of physicians

Demographic density

Number of nurses

Number of psychosocial centers

Number of basic units*

Random Forest max_depth = 15

Accuracy = 0.96

Precision = 0.98

Sensitivity = 0.93

Specificity = 0.94

Demographic density

Number of physicians

Number of nurses

Number of clinics

Public professionals

Number of dentists*

XGBoost

learning rate = 0.1

n estimators = 3000

subsample = 0.5

colsample bytree= 0.5

Accuracy = 0.98

Precision = 0.97

Sensitivity = 0.94

Specificity = 0.94

Number of physicians

Number of nurses

Number of clinics

Public professionals

Number of neonatal units*

Number of short sta

hos

ital*

Infrastructure of the environment in which the

individual is inserted is also important. For example,

green areas within the pathways used by the

population and spatial inaccessibility to psychiatrists

or psychotherapists was shown to interfere in suicide

rates in the region (Shen & Lung, 2018; Tadmon &

Bearman, 2023). In addition, since a substantial

proportion of suicide completers use the healthcare

system the year before their death (Ahmedani et al.,

2014), we sought to understand the impact that the

city's healthcare infrastructure has on SR and SAR in

a case study of the Brazilian Southern region.

The Brazilian health care system is decentralized,

with more focus on primary care and outpatient

specialized services than in hospitalization. It has

private and public domains, and more than 70% of the

population relies on public services. The services are

distributed based on the demographical density and

economic indicators. Thus, smaller cities have

primary but may lack specialized services. Larger

cities usually work as an epicenter, providing

specialized health services (Paim et al., 2011).

We used three models to classify SR and SAR:

DT, RF, and XGB. The models had similar results,

with predictive values of approximately 95%. As the

data are unbalanced, other metrics are also important

to be analyzed, and as presented in Table 2, F1-score

averaged approximately 0.94. As expected, XGB,

having a correction of errors during training, had a

better performance, although all models showed

satisfactory results in classifying SR and SAR.

Understanding which factors are decisive may

help focus on specific targets to change or improve.

Our data shows that Psychosocial Care Centers

(CAPS), a type of healthcare facility focused on

treatment and improvement of psychiatric conditions

in the population, are one of the main features.

Additionally, since the majority of the Brazilian

population relies on the National Health System,

increasing the system with more clinicians and

nurses, especially in cities with high SR and SAR,

may help decrease the rates. Interestingly, some

models also considered the dental professionals,

showing that general well-being is important.

CAPS are divided into six classes and for people

suffering psychological conditions, and focused on

adults, or children and adolescents, or substance

abusers, and may be open 24 hours per day (with

hospital beds available). The simplest CAPS require

15,000 inhabitants, and the most completes up to

150,000 (Ministry of Health, 2023). Brazil has 5 565

cities, from which only 677 have more than 50 000

inhabitants. Thus, the majority of the cities only have

CAPS I available. Considering that psychiatric

conditions and substance abuse are some of the main

factors of suicide, and that children and adolescents

showed high rates of suicide attempt, rates would

probably decrease with more centers in more cities.

A limitation of our study is that in each question

of the SIM/SINAN forms, the evaluator may mark

‘ignored’. The correct filing of the form with answers

that have semantic values can increase the specificity

HEALTHINF 2024 - 17th International Conference on Health Informatics

390

of the analysis, and if the forms were more similar,

we could group the forms and analyze

simultaneously. Also, for the SINAN form, the

presence of an anonymous identifier would help

distinguish the profile of the individual after the first

attempt. In addition, we were not able to distinguish

the different specializations of the healthcare

professionals, such as psychiatrists from general

physicians, or the teams present at the different

facilities, which could improve the models'

performance and point to more direct improvements.

Lastly, underreporting plays a crucial role, especially

in smaller regions, where suicide is more stigmatized.

6 CONCLUSIONS AND

PERSPECTIVES

In this study, we focused on extracting and

interpreting patterns from suicide completion and

reattempt rates in Brazil. This is the first study using

the Brazilian healthcare infrastructure to classify

rates. Our models achieved a high predictive

performance of up to 97% accuracy in predicting

suicide death or reattempt. Compared to other studies,

we focused on the environment in which the

population is inserted, trying to use the model in a

descriptive manner, to identify and better understand

the patterns arising from models’ application. This

approach showed the importance of Psychosocial

Care Centers and the number of physicians and nurses

in impacting deaths and suicide reattempts. Future

studies could use a similar approach with other city

infrastructures, such as those related to

industrialization, employment, education, and

sanitation to decrease these preventable deaths.

ACKNOWLEDGEMENTS

We would like to thank CAPES and FAPESC for the

financial support.

REFERENCES

Ahern, S., Burke, L.-A., McElroy, B., Corcoran, P.,

McMahon, E. M., Keeley, H., Carli, V., Wasserman, C.,

Hoven, C. W., Sarchiapone, M., Apter, A., Balazs, J.,

Banzer, R., Bobes, J., Brunner, R., Cosman, D., Haring,

C., Kaess, M., Kahn, J.-P., … Wasserman, D. (2018).

A cost-effectiveness analysis of school-based suicide

prevention programmes. European Child & Adolescent

Psychiatry, 27(10), 1295–1304. https://doi.org/

10.1007/s00787-018-1120-5

Ahmedani, B. K., Simon, G. E., Stewart, C., Beck, A.,

Waitzfelder, B. E., Rossom, R., Lynch, F., Owen-

Smith, A., Hunkeler, E. M., Whiteside, U., Operskalski,

B. H., Coffey, M. J., & Solberg, L. I. (2014). Health

care contacts in the year before suicide death. Journal

of General Internal Medicine, 29(6), 870–877.

https://doi.org/10.1007/s11606-014-2767-3

Choi, J., Cho, S., Ko, I., & Han, S. (2021). Identification of

risk factors for suicidal ideation and attempt based on

machine learning algorithms: a longitudinal survey in

Korea (2007–2019). International Journal of

Environmental Research and Public Health, 18(23),

12772. https://doi.org/10.3390/ijerph182312772

Coelho, F. C., Baron, B. C., de Castro Fonseca, G. M.,

Reck, P., & Palumbo, D. (2021). AlertaDengue/PySUS:

Vaccine. Zenodo. https://doi.org/10.5281/zenodo.488

3502

Gao, K., Wu, R., Wang, Z., Ren, M., Kemp, D. E., Chan, P.

K., Conroy, C. M., Serrano, M. B., Ganocy, S. J., &

Calabrese, J. R. (2015). Disagreement between self-

reported and clinician-ascertained suicidal ideation and

its correlation with depression and anxiety severity in

patients with major depressive disorder or bipolar

disorder. Journal of Psychiatric Research, 60, 117–

124. https://doi.org/10.1016/j.jpsychires.2014.09.011

Gradus, J. L., Rosellini, A. J., Horváth-Puhó, E., Street, A.

E., Galatzer-Levy, I., Jiang, T., Lash, T. L., & Sørensen,

H. T. (2020). Prediction of sex-specific suicide risk

using machine learning and single-payer health care

registry data from Denmark. JAMA Psychiatry, 77(1),

25. https://doi.org/10.1001/jamapsychiatry.2019.2905

Jaen-Varas, D., Mari, J. J., Asevedo, E., Borschmann, R.,

Diniz, E., Ziebold, C., & Gadelha, A. (2019). The

association between adolescent suicide rates and

socioeconomic indicators in Brazil: a 10-year

retrospective ecological study. Brazilian Journal of

Psychiatry, 41(5), 389–395. https://doi.org/10.1590/

1516-4446-2018-0223

Johnson, T., Kanjo, E., & Woodward, K. (2023).

DigitalExposome: quantifying impact of urban

environment on wellbeing using sensor fusion and deep

learning. Computational Urban Science, 3(1), 14.

https://doi.org/10.1007/s43762-023-00088-9

Kinchin, I., & Doran, C. (2018). The cost of youth suicide

in Australia. International Journal of Environmental

Research and Public Health, 15(4), 672.

https://doi.org/10.3390/ijerph15040672

Martínez-Alés, G., Cruz Rodríguez, J. B., Lázaro, P.,

Domingo-Relloso, A., Barrigón, M. L., Angora, R.,

Rodríguez-Vega, B., Jiménez-Sola, E., Sánchez-Castro,

P., Román-Mazuecos, E., Villoria, L., Ortega, A. J.,

Navío, M., Stanley, B., Rosenheck, R., Baca-García, E.,

& Bravo-Ortiz, M. F. (2021). Cost-effectiveness of a

contact intervention and a psychotherapeutic program

for post-discharge suicide prevention. The Canadian

Journal of Psychiatry, 66(8), 737–746. https://doi.org/

10.1177/0706743720980135

Using Data Mining Techniques to Understand Patterns of Suicide and Reattempt Rates in Southern Brazil

391

Martini, M., da Fonseca, R. C., de Sousa, M. H., de

Azambuja Farias, C., Cardoso, T. de A., Kunz, M.,

Longaray, V. K., & Magalhães, P. V. da S. (2019). Age

and sex trends for suicide in Brazil between 2000 and

2016. Social Psychiatry and Psychiatric Epidemiology,

54(7), 857–860. https://doi.org/10.1007/s00127-019-

01689-8

Mccullough, M. H., Small, M., Jayawardena, B., & Hood,

S. (2023). Mapping patient interactions in psychiatric

presentations to a tertiary emergency department.

MedRxiv Preprint, 1–23. https://doi.org/10.1101/

2023.05.17.23290083

Ministry of Health. (2021). Mortalidade por suicídio e

notificações de lesões autoprovocadas no Brasil.

Boletim Epidemiológico Secretaria de Vigilância Em

Saúde − Ministério Da Saúde, 52(33), 1–10.

https://www.gov.br/saude/pt-br/centrais-de-

conteudo/publicacoes/boletins/epidemiologicos/edicoe

s/2021/boletim_epidemiologico_svs_33_final.pdf

Ministry of Health. (2023). Centros de Atenção

Psicosocial. https://www.gov.br/saude/pt-br/composi

cao/saes/desme/raps/caps

Nordin, N., Zainol, Z., Mohd Noor, M. H., & Lai Fong, C.

(2021). A comparative study of machine learning

techniques for suicide attempts predictive model.

Health Informatics Journal, 27(1), 146045822198939.

https://doi.org/10.1177/1460458221989395

Paim, J., Travassos, C., Almeida, C., Bahia, L., & Macinko,

J. (2011). The Brazilian health system: history,

advances, and challenges. The Lancet, 377(9779),

1778–1797. https://doi.org/10.1016/S0140-6736(11)

60054-8

Palma, D. C. de A., Oliveira, B. F. A. de, & Ignotti, E.

(2021). Suicide rates between men and women in

Brazil, 2000-2017. Cadernos de Saúde Pública, 37(12),

e00281020. https://doi.org/10.1590/0102-311x00281

020

Pinheiro, T. de P., Warmling, D., & Coelho, E. B. S. (2021).

Caracterização das tentativas de suicídio e

automutilações por adolescentes e adultos notificadas

em Santa Catarina, 2014-2018. Epidemiologia e

Serviços de Saúde, 30(4), e2021337.

https://doi.org/10.1590/s1679-49742021000400026

Roglio, V. S., Borges, E. N., Rabelo-da-Ponte, F. D.,

Ornell, F., Scherer, J. N., Schuch, J. B., Passos, I. C.,

Sanvicente-Vieira, B., Grassi-Oliveira, R., von

Diemen, L., Pechansky, F., & Kessler, F. H. P. (2020).

Prediction of attempted suicide in men and women with

crack-cocaine use disorder in Brazil. PLOS ONE, 15(5),

e0232242. https://doi.org/10.1371/journal.pone.0232242

Ross, E. L., Zuromski, K. L., Reis, B. Y., Nock, M. K.,

Kessler, R. C., & Smoller, J. W. (2021). Accuracy

requirements for cost-effective suicide risk prediction

among primary care patients in the US. JAMA

Psychiatry, 78(6), 642.

https://doi.org/10.1001/jamapsychiatry.2021.0089

Sgobin, S. M. T., Traballi, A. L. M., Botega, N. J., &

Coelho, O. R. (2015). Direct and indirect cost of

attempted suicide in a general hospital: cost-of-illness

study. Sao Paulo Medical Journal,

133(3), 218–226.

https://doi.org/10.1590/1516-3180.2014.8491808

Shen, Y.-S., & Lung, S.-C. C. (2018). Identifying critical

green structure characteristics for reducing the suicide

rate. Urban Forestry & Urban Greening, 34, 147–153.

https://doi.org/10.1016/j.ufug.2018.06.005

Shepard, D. S., Gurewich, D., Lwin, A. K., Reed, G. A., &

Silverman, M. M. (2016). Suicide and suicidal attempts

in the United States: costs and policy implications.

Suicide and Life-Threatening Behavior, 46(3), 352–

362. https://doi.org/10.1111/sltb.12225

Tadmon, D., & Bearman, P. S. (2023). Differential spatial-

social accessibility to mental health care and suicide.

Proceedings of the National Academy of Sciences,

120(19). https://doi.org/10.1073/pnas.2301304120

Turecki, G., Brent, D. A., Gunnell, D., O’Connor, R. C.,

Oquendo, M. A., Pirkis, J., & Stanley, B. H. (2019).

Suicide and suicide risk. Nature Reviews Disease

Primers, 5(1), 74. https://doi.org/10.1038/s41572-019-

0121-0

van Heeringen, K., & Mann, J. J. (2014). The neurobiology

of suicide. The Lancet Psychiatry, 1(1), 63–72.

https://doi.org/10.1016/S2215-0366(14)70220-2

Walsh, C. G., Ribeiro, J. D., & Franklin, J. C. (2017).

Predicting Risk of Suicide Attempts Over Time

Through Machine Learning. Clinical Psychological

Science, 5(3), 457–469. https://doi.org/10.1177/

2167702617691560

WHO (2014). Preventing suicide: A global imperative.

World Health Organization.

WHO (2021). Suicide worldwide in 2019: global health

estimates.

Zheng, L., Wang, O., Hao, S., Ye, C., Liu, M., Xia, M.,

Sabo, A. N., Markovic, L., Stearns, F., Kanov, L.,

Sylvester, K. G., Widen, E., McElhinney, D. B., Zhang,

W., Liao, J., & Ling, X. B. (2020). Development of an

early-warning system for high-risk patients for suicide

attempt using deep learning and electronic health

records. Translational Psychiatry, 10(1), 72.

https://doi.org/10.1038/s41398-020-0684-2

HEALTHINF 2024 - 17th International Conference on Health Informatics

392