Analysis of the Predictors of Mortality from Ischemic Heart Diseases in the Southern Region of Brazil: A Geographic Machine-Learning-Based Study.

BACKGROUND

Mortality due to ischemic heart disease (IHD) is heterogeneously distributed globally, and identifying the sites most affected by it is essential in developing strategies to mitigate the impact of the disease, despite the complexity resulting from the great diversity of variables involved.

OBJECTIVE

To analyze the predictability of IHD mortality using machine learning (ML) techniques in combination with geospatial analysis in southern Brazil.

METHODS

Ecological study using secondary and retrospective data on mortality due to ischemic heart disease (IHD) obtained from the Mortality Information Systems (SIM-DATASUS) de 2018 a 2022, covering 1,191 municipalities in the states of Paraná (399), Santa Catarina (295), and Rio Grande do Sul (497). Ordinary Least Squares Regression (OLS), Geographically Weighted Regression (GWR), Random Forest (RF), and Geographically Weighted Random Forest (GWRF) analyses were performed to verify the model with the best performance capable of identifying the most affected sites by the disease based on a set of predictors composed by variables of procedures and access to health.

RESULTS

In the analyzed period, there were 59,093 deaths, 65% of which were men, 82.7% were white, and 72.8% occurred between 60 and 70 years of age. Ischemic heart disease presented the highest mortality rates in the northwest and north regions of the state of Paraná, and in the central-east, southwest and southeast regions of Rio Grande do Sul, the latter state accounting for 41% of total deaths. The GWRF presented the best performance with R = 0.983 and AICc = 2298.4, RMSE: 3.494 and the most important variables of the model in descending order were electrocardiograph rate, cardiac catheterization rate, access index to hemodynamics, access index of pre-hospital mobile units, cardiologists rate, myocardial scintigraphy rate, stress test rate, and stress echocardiogram rate.

CONCLUSION

The GWRF identified spatial heterogeneity in the variation of geographic predictors, contrasting the limitation of linear regression models. The findings showed patterns of vulnerability in southern Brazil, suggesting the formulation of health policies to improve access to diagnostic and therapeutic resources, with the potential to reduce IHD mortality.