Developing a Vulnerability Index Model in Mexico to Forecast Negative Health Outcomes

Background: Equitable allocation of public health resources relies on aggregated data to visualize demand. In the United States, the Centers for Disease Control and Prevention (CDC) incorporated 15 socio-demographic census indicators into the Social Vulnerability Index, which determines risk quantiles across varying geographic resolutions. However, the CDC methodology may be oversimplified and unsuitable beyond the US, such as in Mexico. We described the development and validity of new vulnerability index models (VIM) by using data from Mexico and by incorporating a more diverse set of indicators. Methods: We applied CDC methodology to Mexico with indicators that capture demographic, health, economic, and novel environmental themes (e.g., air quality). We assessed whether environmental conditions contribute to communicable disease, cancer, and other non-communicable disease (NCD) burden via a correlation analysis of VIM vulnerability rankings and state-level disease burden rankings. This analysis compared the reference VIM to an experimental VIM that included environmental indicators to assess whether environmental indicators improved VIM predictability. We also expected the model to best predict NCD-related outcomes, as politicization confounds outcomes related to infectious diseases, such as COVID-19 and HIV. Findings: Demographic indicators best correlated with 2020 COVID-19 death rates (p=0.01), and the environmental indicators best correlated with 2019 NCD death rates (p=0.01). Across all five disease outcomes, the experimental model with environmental indicators performed equal to or better than the reference model. Interpretation: Our novel VIM predicted burden across relevant metrics in Mexico and may apply to more vulnerable settings, thus better informing long-term resource allocation for global health. Future work will improve the experimental VIM’s predictive power via robust statistical approaches, such as principal component analysis.