State-of-the-Art Risk Models for Diabetes, Hypertension, Visual Diminution, and COVID-19 Severity in Mexico

Heladio Amaya, Jennifer Enciso, Daniela Meizner, Alex Pentland, Alejandro NoriegamedRxiv 2021 · DOI

Abstract

Diabetes and hypertension are among the top public health priorities, particularly in low- and middle-income countries where their health and socioeconomic impact is exacerbated by limited healthcare access. This research developed and validated state-of-the-art risk models for undiagnosed diabetes, hypertension, associated visual complications, and severe COVID-19 risk in Mexico. Using gradient boosting trees trained on representative national datasets (ENSANUT 2018 and DGE COVID-19), the models achieved significant improvements in accuracy and algorithmic equity compared to international benchmarks. Specifically, the diabetes model reached an AUC of 0.83 with high sensitivity (90%) and specificity (56%), while drastically reducing performance disparities across gender, indigenous status, and region. These tools, deployed via a free online platform, aim to democratize early detection and enable large-scale preventive strategies.

Introduction: Addressing the Diagnosis Gap

In low- and middle-income countries (LMICs), chronic diseases like diabetes and hypertension often go undiagnosed until severe complications—such as blindness or kidney failure—occur. In Mexico, approximately 50% of patients are unaware they have these conditions. Furthermore, these comorbidities are primary drivers of severe COVID-19 outcomes.

Existing risk assessment tools, such as the American Diabetes Association (ADA) calculator, were largely developed using data from predominantly white, high-income populations. When applied to the Mexican population, these tools often suffer from "algorithmic inequity," performing significantly worse for certain subgroups, such as indigenous populations or women. This study sought to build a tool that is both more accurate and more equitable by design through localization.

Methods and Data

The researchers utilized Gradient Boosting Classifiers to build predictive models. To ensure practical utility, they focused on non-clinical features that a layperson could answer in a simple questionnaire (e.g., age, weight, family history, and household assets).

Data Sources

  1. ENSANUT 2018: The Mexican National Health and Nutrition Survey, providing representative health and socioeconomic data for over 43,000 adults, including biomarkers (HbA1c) as "ground truth."
  2. DGE COVID-19 Dataset: Data from the Mexican General Directorate of Epidemiology, covering over 1.6 million tests to model disease severity.

Optimization for Equity

Unlike standard models that only optimize for overall Area Under the Curve (AUC), the authors explicitly measured and minimized "disparity" ($\delta$), defined as the difference in model sensitivity across different demographic subgroups.

Performance Results

The "Prosperia" models outperformed established benchmarks across all categories. For diabetes, the model achieved an AUC of 0.83, significantly higher than the 0.74 achieved by the ADA calculator and 0.75 by LA-FINDRISC.

ROC curves for diabetes risk models comparing Prosperia (AUC=0.83) against ADA and LA-FINDRISC benchmarks.

Similar improvements were seen in other areas:

  • Hypertension: AUC of 0.79 (vs. 0.67 benchmark).
  • COVID-19 Severity: AUC of 0.84 (vs. 0.80 official IMSS calculator).

Algorithmic Equity and Localization

A key insight of this work is the "Localization Premium." By training on local data, the models successfully used socioeconomic indicators (like region of residence or household assets) as proxies for healthcare access. This approach proved that overall accuracy does not need to be sacrificed for equity.

In the diabetes model, the disparity in sensitivity between sex groups was reduced from 9.7% (ADA) to just 1.5%. Similarly, disparity between indigenous and non-indigenous groups dropped from 8.7% to 2.5%.

Comparison of sensitivity disparity by sex across different risk models, showing Prosperia's significantly lower disparity (1.5%) compared to the ADA benchmark (9.7%).

Conclusion

The use of large, representative databases coupled with modern statistical learning allowed for the development of risk models with state-of-the-art accuracy and equity. While limitations such as self-reporting bias and the digital divide remain, these tools represent a meaningful step toward democratizing early detection. The project has transitioned from research to practice with the deployment of a free online tool for mass risk assessment in Mexico.

Summary updated

Abstract

BACKGROUND Diabetes and hypertension are among top public health priorities, particularly in low and middle-income countries where their health and socioeconomic impact is exacerbated by the quality and accessibility of health care. Moreover, their connection with severe or deadly COVID-19 illness has further increased their societal relevance. Tools for early detection of these chronic diseases enable interventions to prevent high-impact complications, such as loss of sight and kidney failure. Similarly, prognostic tools for COVID-19 help stratify the population to prioritize protection and vaccination of high-risk groups, optimize medical resources and tests, and raise public awareness. METHODS We developed and validated state-of-the-art risk models for the presence of undiagnosed diabetes, hypertension, visual complications associated with diabetes and hypertension, and the risk of severe COVID-19 illness (if infected). The models were estimated using modern methods from the field of statistical learning (e.g., gradient boosting trees), and were trained on publicly available data containing health and socioeconomic information representative of the Mexican population. Lastly, we assembled a short integrated questionnaire and deployed a free online tool for massifying access to risk assessment. RESULTS Our results show substantial improvements in accuracy and algorithmic equity (balance of accuracy across population subgroups), compared to established benchmarks. In particular, the models: i) reached state-of-the-art sensitivity and specificity rates of 90% and 56% (0.83 AUC) for diabetes, 80% and 64% (0.79 AUC) for hypertension, 90% and 56% (0.84 AUC) for visual diminution as a complication, and 90% and 60% (0.84 AUC) for development of severe COVID disease; and ii) achieved substantially higher equity in sensitivity across gender, indigenous/non-indigenous, and regional populations. In addition, the most relevant features used by the models were in line with risk factors commonly identified by previous studies. Finally, the online platform was deployed and made accessible to the public on a massive scale. CONCLUSIONS The use of large databases representative of the Mexican population, coupled with modern statistical learning methods, allowed the development of risk models with state-of-the-art accuracy and equity for two of the most relevant chronic diseases, their eye complications, and COVID-19 severity. These tools can have a meaningful impact on democratizing early detection, enabling large-scale preventive strategies in low-resource health systems, increasing public awareness, and ultimately raising social well-being.
BibTeX
@misc{amaya2021state,
  author = {Heladio Amaya and Jennifer Enciso and Daniela Meizner and Alex Pentland and Alejandro Noriega},
  title = {{State-of-the-Art Risk Models for Diabetes, Hypertension, Visual Diminution, and COVID-19 Severity in Mexico}},
  year = {2021},
  doi = {10.1101/2021.01.18.21250034},
  howpublished = {medRxiv},
}
Figure