Using artificial intelligence for estimates in small-area tobacco surveys
Abstract
Objectives: To analyze the effect of using ARTIFICIAL INTELLIGENCE and synthetic populations in population surveys to improve the accuracy of estimates for small areas.
Methods: Ecological study with data from Vigitel from 2006 to 2018 for Belo Horizonte, State Capital in Minas Gerais, Brazil. The outcome was the prevalence of current smokers. The interviews were georeferenced to the census tracts of the 2010 Census and grouped into nine vulnerability clusters derived from the Social Vulnerability Index (SVI). To address the lack of interviews in more vulnerable areas, synthetic populations were generated using conditional adversarial generative networks (cGANs), incorporated into the clusters with the fewest observations. The quality of the synthetic data was assessed for structure and statistical similarity, and the estimates were weighted by post-stratification (rake) based on the 2010 Census. Prevalences and their respective 95% confidence intervals were estimated for three periods, comparing real and combined data.
Results: The clusters with greater vulnerability had fewer interviews and a higher standard error. The incorporation of 700 synthetic records per cluster substantially reduced the standard error (to 1.7%) and improved the accuracy of the estimates, especially in the most vulnerable areas. The similarity between real and synthetic data was high (KSComplement = 0.968). The highest smoking prevalences were concentrated in the clusters with greater vulnerability, with a tendency to decrease over the analyzed period.
Conclusions: This strategy represents advances in the monitoring of NCDs and their risk factors, allowing for more accurate and representative estimates in contexts of high vulnerability, and supporting public policies and actions to identify tobacco use and other risk factors.
Key messages: • The study represents advances in the monitoring of NCDs and risk factors, by using artificial intelligence techniques, such as synthetic populations, to reduce variability in small area estimates.
• The study showed Tobbaco prevelence in areas of greater vulnerability. This can support health promotion policies and programs that encourage reduce Tobbaco use.
Related
- The Influence of Large Language Model-Generated Artificial Intelligence Risk Estimates on Clinical Decision-Making in Breast Imaging: A Simulation-Based Study
- Mapping the adoption and implementation of artificial intelligence (AI)-based conversational interface for tobacco cessation: a scoping review protocol.
- Artificial intelligence and radiomics in small cell lung cancer: a narrative review of imaging and multisource data integration
- Clinical Artificial Intelligence in Non-Small Cell Lung Cancer: From Early Detection and Computational Pathology to Precision Oncology
- Artificial Intelligence Versus Conventional Scoring for Assessing Small Bowel Capsule Endoscopy Cleanliness in Crohn's Disease
BibTeX
@article{malta2026using,
title = {{Using artificial intelligence for estimates in small-area tobacco surveys}},
author = {D Malta and C Saars and R Tomie Ivata Bernal and S Alves and J Bottoni and B Carrato and LA Tonnaco and E Ribeiro and M Gonçalvez and J Marques},
journal = {The European Journal of Public Health},
year = {2026},
month = oct,
doi = {10.1093/eurpub/ckag169.1196},
url = {https://doi.org/10.1093/eurpub/ckag169.1196},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Using artificial intelligence for estimates in small-area tobacco surveys
AU - D Malta
AU - C Saars
AU - R Tomie Ivata Bernal
AU - S Alves
AU - J Bottoni
AU - B Carrato
AU - LA Tonnaco
AU - E Ribeiro
AU - M Gonçalvez
AU - J Marques
PY - 2026
DA - 2026/10/01
JO - The European Journal of Public Health
DO - 10.1093/eurpub/ckag169.1196
UR - https://doi.org/10.1093/eurpub/ckag169.1196
AB - Objectives: To analyze the effect of using ARTIFICIAL INTELLIGENCE and synthetic populations in population surveys to improve the accuracy of estimates for small areas. Methods: Ecological study with data from Vigitel from 2006 to 2018 for Belo Horizonte, State Capital in Minas Gerais, Brazil. The outcome was the prevalence of current smokers. The interviews were georeferenced to the census tracts of the 2010 Census and grouped into nine vulnerability clusters derived from the Social Vulnerability Index (SVI). To address the lack of interviews in more vulnerable areas, synthetic populations were generated using conditional adversarial generative networks (cGANs), incorporated into the clusters with the fewest observations. The quality of the synthetic data was assessed for structure and statistical similarity, and the estimates were weighted by post-stratification (rake) based on the 2010 Census. Prevalences and their respective 95% confidence intervals were estimated for three periods, comparing real and combined data. Results: The clusters with greater vulnerability had fewer interviews and a higher standard error. The incorporation of 700 synthetic records per cluster substantially reduced the standard error (to 1.7%) and improved the accuracy of the estimates, especially in the most vulnerable areas. The similarity between real and synthetic data was high (KSComplement = 0.968). The highest smoking prevalences were concentrated in the clusters with greater vulnerability, with a tendency to decrease over the analyzed period. Conclusions: This strategy represents advances in the monitoring of NCDs and their risk factors, allowing for more accurate and representative estimates in contexts of high vulnerability, and supporting public policies and actions to identify tobacco use and other risk factors. Key messages: • The study represents advances in the monitoring of NCDs and risk factors, by using artificial intelligence techniques, such as synthetic populations, to reduce variability in small area estimates. • The study showed Tobbaco prevelence in areas of greater vulnerability. This can support health promotion policies and programs that encourage reduce Tobbaco use.
SN - 1101-1262
SN - 1464-360X
ER -