Improving the mapping of leisure-time physical activity inequities: the use of artificial intelligence to advance estimates of small-areas in Brazil

Abstract

Objective

We estimated the prevalence of leisure-time physical activity (LTPA) in small areas of the city of Belo Horizonte and analyzed inequities across areas and between two time periods, 2009–2013 and 2014–2018.

Study design

Small area estimation using clustered data.

Methods

Data from the Surveillance of Risk and Protective Factors for Chronic Diseases (Vigitel) between 2009 and 2018 for the city of Belo Horizonte, Brazil, were used. Firstly, interviews were georeferenced (n = 16,019) in the census tracts (3,830) of the city. Secondly, the socioeconomic and sanitation components of a Health Vulnerability Index (IVH), indicator used to classify census tracts, served as input variables for a k-means clustering technique to group the tracts in smaller areas of higher homogeneity in relation to IVS components. Lastly, direct estimation of LTPA prevalence was obtained in each cluster by applying post-stratification weights to sample. Absolute and relative differences were calculated between periods and prevalence differences between clusters to analyze inequalities.

Results

Nine clusters were found. LTPA prevalence ranged from 23.70 % in a very high-risk cluster to 45.55 % in a low-risk cluster during the 2009–2013 period, and from 31.44 % in a high-risk cluster to 52.81 % in a low-risk cluster from 2014 to 2018. Reducing inequities were observed among the more disadvantaged clusters, but it remained persistently large between the lowest and highest vulnerability groups.

Conclusion

Small area estimates are an advantage for a more accurate level of health surveillance. Different machine learning methods are encouraged to provide information for more tailored interventions at local level.