Bayesian spatial modeling of childhood overweight and obesity prevalence in Costa Rica

**Background:** Childhood overweight and obesity are rising globally, with Costa Rica reporting a combined prevalence of 34.5% among children aged 6–12 in 2016. Understanding geographic patterns and socioeconomic drivers is critical for targeted public health interventions. This study aimed to identify district-level associations between socioeconomic factors and childhood overweight and obesity as separate conditions using Bayesian spatial modeling.

**Methods:** Data came from the 2016 Weight and Size Census (347,379 children aged 6–12 attending public and private schools) and the 2011 National Census, aggregated across 472 districts. Overweight was defined as BMI between 1 and 2 standard deviations above the WHO median; obesity as BMI >2 SD above median. A Bayesian multinomial logistic model with a BYM2 spatial random effect was used, treating normal weight as the reference category. Covariates included unemployment rate, percentage urban population, percentage of homes with at least one critical deprivation (poverty proxy), percentage of population under 14 years, percentage of single-mother homes, average occupants per home, and average years of schooling (with a quadratic term). Model selection used WAIC and CPO criteria. The final model included deprivation, population under 14, schooling, and schooling². Estimation used Integrated Nested Laplace Approximation (INLA).

**Key Results:** For obesity, a 10-percentage-point increase in homes with critical deprivation was associated with a 7.7% decrease in relative odds (OR 0.992, 95% CI 0.988–0.995). A 10-percentage-point increase in population under 14 was associated with a 14.0% decrease in relative odds of obesity (OR 0.985, 95% CI 0.976–0.994). Schooling showed a non-linear relationship: the odds ratio for the linear term was 1.189 (95% CI 1.029–1.373) and for the quadratic term 0.989 (95% CI 0.982–0.997), indicating an inverted U-shape peaking at approximately 8 years of education. For overweight, only schooling was statistically significant (linear OR 1.130, 95% CI 1.014–1.259; quadratic OR 0.994, 95% CI 0.988–0.999). The spatial random effect was substantial: for overweight, τ posterior mean 68.538 (SD 10.961) and ϕ 0.844 (SD 0.094); for obesity, τ 26.533 (SD 3.538) and ϕ 0.944 (SD 0.046), indicating strong structured spatial correlation. Exceedance probability maps showed that districts in central Costa Rica, the Nicoya Peninsula, Central Pacific coast, and border areas with Panama had high probabilities of exceeding national targets (20% overweight, 14% obesity).

**Clinical Implications:** The non-linear relationship between education and obesity suggests that as educational attainment increases beyond approximately 8 years, obesity risk declines—supporting education as a 'social vaccine' against obesogenic environments. The negative association between deprivation and obesity reflects a pattern common in developing countries where higher socioeconomic status paradoxically increases obesity risk, contrasting with patterns in high-income nations. Districts with higher proportions of older residents (lower % under 14) had higher obesity odds, suggesting peer presence may promote active lifestyles. These findings highlight the need for tailored interventions: promoting healthy habits among educated parents, ensuring safe spaces for physical activity, and targeting high-risk geographic areas (commercial/tourist hubs, administrative centers, border regions). The accompanying Shiny dashboard provides health authorities with interactive tools to explore prevalence estimates and exceedance probabilities.