Small-area inequalities in weight excess and obesity among adults in Belo Horizonte, Brazil: a repeated cross-sectional ecological study using Vigitel and synthetic populations

Technical Summary: Small-Area Inequalities in Weight Excess and Obesity in Belo Horizonte

Problem Statement
Weight excess and obesity are escalating public health challenges in Brazil, driven by complex interactions between food systems, urban environments, and social inequalities. While national surveillance systems like Vigitel (Surveillance of Risk and Protective Factors for Chronic Diseases by Telephone Survey) provide robust city-level estimates, these aggregates often mask critical intra-urban heterogeneity. In large, socially stratified cities like Belo Horizonte, municipal averages can obscure priority areas and flatten social gradients, limiting the utility of data for targeted local public health planning. Furthermore, conventional small-area analyses using survey data face a methodological hurdle: socially vulnerable territories are frequently underrepresented in telephone-based samples, leading to sparse observations and unstable prevalence estimates where detailed evidence is most needed.

Methodology
This study employed a repeated cross-sectional ecological design to analyze data from Vigitel for adult residents (≥18 years) of Belo Horizonte between 2006 and 2018.

• Data Source & Stratification: Survey years were aggregated into three periods (2006–2009, 2010–2013, and 2014–2018) to enhance statistical stability. All analyses accounted for the complex survey design, including post-stratification weights based on sex, age, and education.
• Outcomes: The study analyzed binary outcomes for weight excess (BMI ≥25.0 kg/m²) and obesity (BMI ≥30.0 kg/m²), derived from self-reported height and weight.
• Territorial Clustering: Instead of using standard administrative units, the study utilized the Health Vulnerability Index (HVI) to define nine socially coherent microarea clusters. These clusters were derived via hierarchical clustering of census tract data regarding sanitation, housing, and socioeconomic status (income and education). The clusters ranged from Low Vulnerability (LA-0, LA-1) to Very High Vulnerability (VH-0, VH-1).
• Synthetic Population Augmentation: To address data sparsity in vulnerable clusters, the study incorporated synthetic populations generated using Conditional Generative Adversarial Networks (cGANs). Synthetic data were introduced only in clusters with fewer than 1,000 interviews over the aggregated period. The cGANs modeled complex joint distributions of key variables (age, education, BMI, HVI cluster) while preserving privacy. The fidelity of the synthetic data was validated using the KSComplement metric and marginal distribution comparisons.
• Statistical Analysis: Prevalence estimates and 95% confidence intervals (CI) were calculated for each sex, time period, and HVI cluster. Temporal trends were assessed by comparing non-overlapping CIs, and sex differences were evaluated using weighted p-values.

Key Results

• Temporal Trends: Both weight excess and obesity increased over time across most microareas in Belo Horizonte.
• Sex and Vulnerability Interactions:
  • Weight Excess: Generally more prevalent among men, particularly in low and medium vulnerability clusters. However, this gap narrowed in high and very high vulnerability clusters, where women showed similar or higher prevalence.
  • Obesity: Consistently more prevalent among women, especially in intermediate, high, and very high vulnerability clusters. A clear social gradient was observed for obesity among women, with prevalence increasing alongside vulnerability. In contrast, obesity increases among men were less uniform and largely confined to more vulnerable areas.
• Methodological Impact: The inclusion of synthetic populations successfully increased the effective sample size in data-sparse microareas, thereby improving the stability of local prevalence estimates without replacing the empirical basis of the survey data.

Significance and Claims
The paper argues that weight excess and obesity in Belo Horizonte are not uniformly distributed but are structured by the intersection of sex, gendered social processes, and territorial vulnerability.

• Epidemiological Insight: The study demonstrates that city-wide averages conceal marked territorial gradients. Specifically, the burden of obesity is progressively concentrated in microareas marked by greater vulnerability, particularly among women. This suggests that weight excess and obesity are shaped by related but distinct social mechanisms: men in less vulnerable areas may reflect earlier stages of the nutritional transition (sedentary work, eating out), while women in highly vulnerable areas face cumulative pressures including food insecurity, unpaid care work, and restricted access to safe physical activity spaces.
• Methodological Contribution: The research validates the utility of combining HVI-based territorial clusters with synthetic population strategies. This approach makes intra-urban inequalities more visible and analytically tractable, offering a robust method for small-area surveillance where conventional sampling fails.
• Policy Implications: The findings support the need for interventions that are both territorialized and gender-responsive. The authors suggest that effective obesity prevention requires moving beyond municipal averages to address the specific mechanisms producing obesogenic conditions in vulnerable territories, including improvements in food environments, transport, urban safety, and social protection policies.

Limitations
The authors acknowledge limitations inherent to the study design, including the use of self-reported anthropometric data (though validated for surveillance), the ecological nature of the study which precludes individual causal inference, and the reliance on model assumptions for synthetic data generation. They emphasize that synthetic populations improve estimate stability but do not correct structural biases in the original data source.