Suicide disparities across metropolitan areas in the US: A comparative assessment of socio-environmental factors using a data-driven predictive approach

Abstract

Disparity in suicide rates across various metropolitan areas in the US is growing. Besides personal genomics and pre-existing mental health conditions affecting individual-level suicidal behaviors, contextual factors are also instrumental in determining region-/community-level suicide risk. However, there is a lack of quantitative approach to model the complex associations and interplays of the socio-environmental factors with the regional suicide rates. In this paper, we propose a holistic data-driven framework to model the associations of socio-environmental factors (demographic, socio-economic, and climate) with the suicide rates, and compare the key socio-environmental determinants of suicides across the large and medium/small metros of the vulnerable US states, leveraging a suite of advanced statistical learning algorithms. We found that random forest outperforms all the other models in terms of both in-sample goodness-of-fit and out-of-sample predictive accuracy, which is then used for statistical inferencing. Overall, our findings show that there is a significant difference in the relationships of socio-environmental factors with the suicide rates across the large and medium/small metropolitan areas of the vulnerable US states. Particularly, suicides in medium/small metros are more sensitive to socio-economic and demographic factors, while that in large metros are more sensitive to climatic factors. Our results also indicate that non-Hispanics, native Hawaiian or Pacific islanders, and adolescents aged 15-29 years, residing in the large metropolitan areas, are more vulnerable to suicides compared to those living in the medium/small metropolitan areas. We also observe that higher temperatures are positively associated with higher suicide rates, with large metros being more sensitive to such association compared to that of the medium/small metros. Our proposed data-driven framework underscores the future opportunities of using big data analytics in analyzing the complex associations of socio-environmental factors and inform policy actions accordingly.

Fig 1. Violin plot depicting normalized suicide mortality rates between large and medium/small metropolitan areas.

Violin plots are similar to box plots, with a rotated kernel density plot on each side showing the probability density of the data at different values.

Fig 2. Schematic of the proposed data-driven research framework.

Fig 3. Large metropolitan counties: Model diagnostics of final random forest model.

(A) Residuals QQ plot (the blue dashed lines represent 95% confidence intervals); (B) Predicted versus actual suicide counts, normalized per 100,000 of population.

Fig 4. Medium/Small metropolitan counties: Model diagnostics of final random forest model.

(A) Residuals QQ plot (the blue dashed lines represent 95% confidence intervals); (B) Predicted versus actual suicide counts, normalized per 100,000 of population.

Fig 5. Variable importance ranking of top 15 predictors.

Top 15 socio-environmental factors selected from random forest in relation to suicide rates are shown in (a) and (b) with respect to large metros and medium/small metros.

Fig 6. Suicide mortality rate and race: (A) Large metro areas; (B) Medium/small metro areas.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.

Fig 7. Suicide mortality rate and gender: (A) Large metro areas; (B) Medium/small metro areas.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.

Fig 8. Suicide mortality rate and age: (A) Large metro areas; (B) Medium/small metro areas.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.

Fig 9. Suicide mortality rate and education: (A) Large metro areas; (B) Medium/small metro areas.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.

Fig 10. Suicide mortality rate in economics for (B) Medium/small metro areas only.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.

Fig 11. Suicide mortality rate and climate: (A) Large metro areas; (B) Medium/small metro areas.

Rug lines on the x axis indicate prevalence of data points; black curve is the average marginal effect of the predictor variable; red lines indicate the 95% confidence intervals.