Spatial Clusters of Cancer Mortality in Brazil - A Machine Learning Modeling Approach

Aug 4, 2026

Upload by Himbo

The study evaluates whether machine learning can predict cancer mortality (CM) at an ecological level and then uses the predictions to locate statistically significant spatial clusters of excess cancer mortality (eCM) across Brazilian municipalities. Age-standardized CM is derived from official national databases, using sociodemographic and health coverage variables as predictive features. Models are trained on 70% of data and validated on 30%. Spatial clusters are detected with SatScan, including separate analyses for the 10 most frequent cancer types.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

INTRODUCTION

Geographic variation in cancer and Brazil’s context

Scan statistics and machine learning in spatial epidemiology

METHODS

Data collection and case definition

Model training and validation

Spatial cluster detection and cancer-type analyses

RESULTS

Predictive performance

Cluster findings by cancer site

Key explanatory variables

CONCLUSION

Edited by: Nino Kuenzli, Swiss School of Public Health (SSPH+), Switzerland *Correspondence: Bruno Casaes Teixeira [bcasteix@gmail.com](bcasteix@gmail.com) Received: 23 January 2022 Accepted: 26 June 2023 Published: 20 July 2023 Citation: Casaes Teixeira B, Toporcov TN, Chiaravalloti-Neto F and Chiavegatto Filho ADP (2023) Spatial Clusters of Cancer Mortality in Brazil: A Machine Learning Modeling Approach. Int J Public Health 68:1604789. doi: 10.3389/ijph.2023.1604789 Spatial Clusters of Cancer Mortality in Brazil: A Machine Learning Modeling Approach Bruno Casaes Teixeira*, Tatiana Natasha Toporcov, Francisco Chiaravalloti-Neto and Alexandre Dias Porto Chiavegatto Filho Department of Epidemiology, Faculty of Public Health, University of São Paulo, São Paulo, Brazil Objectives: Our aim was to test if machine learning algorithms can predict cancer mortality (CM) at an ecological level and use these results to identify statistically signiﬁcant spatial clusters of excess cancer mortality (eCM) . Methods: Age-standardized CM was extracted from the ofﬁcial databases of Brazil. Predictive features included sociodemographic and health coverage variables. Machine learning algorithms were selected and trained with 70% of the data, and the performance was tested with the remaining 30% . Clusters of eCM were identiﬁed using SatScan. Additionally, separate analyses were performed for the 10 most frequent cancer types. Results: The gradient boosting trees algorithm presented the highest coefﬁcient of determination (R2 = 0 .66) . For total cancer, all algorithms overlapped in the region of Bagé (27% eCM) . For esophageal cancer, all algorithms overlapped in west Rio Grande do Sul (48%–96% eCM) . The most signiﬁcant cluster for stomach cancer was in Macapá(82% eCM) . The most important variables were the percentage of the white population and residents with computers. Conclusion: We found consistent and well-deﬁned geographic regions in Brazil with signiﬁcantly higher than expected cancer mortality. Keywords: Brazil, cancer, machine-learning, spatial-clusters, socioeconomic INTRODUCTION Cancer occurrence signiﬁcantly varies among different geographical locations and types of cancer. A comprehensive analysis of age-adjusted incidence rates on a global scale found that, in 2018, the incidence was 419 per 100,000 inhabitants in Oceania, 350 in North America, 217 in Latin America and the Caribbean, and 130 in Africa, as per the report by [1] . In Brazil, cancer is ranked as the second leading cause of death, resulting in 227,920 deaths in 2018, as estimated by the World Health Organization [2] . It also reports that the age-adjusted death rate due to cancer is 111 per 100,000 for men and 95 per 100,000 for women. Within Brazil, these rates ﬂuctuate considerably, with higher rates observed in the country’s more developed southern and southeastern regions [3] . Brazil, with its vast territory and stark socioeconomic disparities, is home to a multitude of ethnic groups. Yet, its healthcare system is relatively uniform nationwide, which makes it a potentially promising setting for eco-epidemiologic modeling. Machine learning models have been utilized in diverse healthcare ﬁelds, primarily for creating individualized prediction Int J Public Health | Owned by SSPH+ | Published by Frontiers 1 July 2023 | Volume 68 | Article 1604789 | | | --- | | | | FIGURE 1 | Study area and schematic diagram of methods (Spatial clusters of cancer mortality in Brazil: a machine learning modeling approach, Brazil, 2008–2016) . 1) 2010 Census [24], 2) Brazilian Territory 2020 [30], 3) Gross Domestic Product (GDP) by Purchasing Power Parity (PPP) and 2017 International Dollars [47], 4) Human Development Index (HDI) [48], 5) Gini Index (World Bank Estimate) [49], and 6) Poverty headcount ratio at $5 .50 a day (2011 PPP) (% of population) [50] . | algorithms. These include predicting the mortality risk during chemothera

FAQs

What is the primary objective of the study?

To test whether machine learning algorithms can predict cancer mortality at an ecological level and use the results to identify statistically significant spatial clusters of excess cancer mortality in Brazil.

How are predictive features and cancer mortality data constructed?

Age-standardized cancer mortality is extracted from official Brazilian databases. Predictive features include sociodemographic and health coverage variables, while cancer mortality is defined using ICD-10 malignant tumor coding aggregated to the municipality level.

Which method is used to detect spatial clusters and are cancer types analyzed separately?

Spatial clusters of excess mortality are identified using SatScan. Separate analyses are also performed for the 10 most frequent cancer types.

Aug 4, 2026

Upload by Himbo