How Your Location Relates to Health - Variable Importance and Interpretable Machine Learning for Environmental and Sociodemographic Data

Sep 17, 2026

Uploaded by Skyler

Health outcomes depend on complex environmental and sociodemographic factors whose effects vary across location and time. Using the MEDSAT dataset for England, the study applies multiple variable-importance techniques to find the most informative predictors across several health outcomes. It then builds an interpretable machine-learning framework using GAMs and MGWR to model local and global spatial dependencies of each variable. Results highlight NO2 and other outcome-specific factors, reveal regional shifts in air pollution and solar radiation, and examine changes during COVID. The approach targets actionable insights for health disparities and supports interpretable models in public health.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Abstract

Introduction

How Your Location Relates to Health: Variable Importance and Interpretable Machine Learning for Environmental and Sociodemographic Data Ishaan Maitra 1 * , Raymond Lin 1 * , Eric Chen 1 , Jon Donnelly 1 , Sanja ˇSepanovi2 , Cynthia Rudin 1 1Duke University, North Carolina, United States 2Nokia Bell Labs, Cambridge, United Kingdom [ishaan.maitra@duke.edu](ishaan.maitra@duke.edu), [raymond.lin@duke.edu](raymond.lin@duke.edu), [eric.y.chen@duke.edu](eric.y.chen@duke.edu), [jon.donnelly@duke.edu](jon.donnelly@duke.edu), [sanja.scepanovic@nokia-bell-labs.com](sanja.scepanovic@nokia-bell-labs.com), [cynthia@cs.duke.edu](cynthia@cs.duke.edu) arXiv :2501 .02 1 1 1v 1 [ cs .LG] 3 Jan 2025 Abstract Health outcomes depend on complex environmental and sociodemographic factors whose effects change over location and time. Only recently has fine-grained spatial and temporal data become available to study these effects, namely the MEDSAT dataset of English health, environmental, and sociodemographic information. Leveraging this new resource, we use a variety of variable importance techniques to robustly identify the most informative predictors across multiple health outcomes. We then develop an interpretable machine learning framework based on Generalized Additive Models (GAMs) and Multiscale Geographically Weighted Regression (MGWR) to analyze both local and global spatial dependencies of each variable on various health outcomes. Our findings identify NO2 as a global predictor for asthma, hypertension, and anxiety, alongside other outcome-specific predictors related to occupation, marriage, and vegetation. Regional analyses reveal local variations with air pollution and solar radiation, with notable shifts during COVID. This comprehensive approach provides actionable insights for addressing health disparities, and advocates for the integration of interpretable machine learning in public health. Code—[github.com/maitraishaan/VI-and-IML-MEDSAT](github.com/maitraishaan/VI-and-IML-MEDSAT)[ ](github.com/maitraishaan/VI-and-IML-MEDSAT)Datasets—[mediatum.ub.tum.de/1714817](mediatum.ub.tum.de/1714817) Introduction Recent increases in the prevalence of chronic diseases such as diabetes and hypertension (Jia 2022) along with empirical evidence of the health risks associated with climate change (Obradovich et al. 2018) highlight the pressing need to understand what environmental and sociodemographic features impact health. Only recently has a very detailed dataset (MEDSAT, Scepanovic et al. 2023) become available that integrates health determinants, including sociodemographic, environmental, and geographic features, withoutcome-related prescription data as a proxy for health outcomes. It offers an unprecedented opportunity for exploring these relationships in close detail across England – if we have the trustworthy tools to do it. *These authors contributed equally. Copyright © 2025, Association for the Advancement of Artificial Intelligence ([www.aaai.org](www.aaai.org)). All rights reserved. Prior work has utilized other integrated datasets to uncover relationships between environmental or sociodemographic variables and health outcomes, but these studies tend to 1) rely on a single variable importance method, 2) fail to measure spatially local relationships, and 3) do not create tools that provide extremely interpretable and auditable models. A group of recent works have focused on demonstrating how air pollution triggers respiratory and cardiovascular diseases (Annesi-Maesano et al. 2021; RazaviTermeh, Sadeghi-Niaraki, and Choi 2021; Ravindra et al. 2023; Liu et al. 2024) . Other works have aimed to determine risk factors for depression and anxiety (Qasrawi et al. 2022; Byeon 2021; Elshawi, Al-Mallah, and Sakr 2019) . These studies have a primary limitation: the lack of a comprehensive dataset that would enable the disentangling of contributions from multiple factors to health outcomes. The lack of a large, complex datase

FAQs

What dataset supports the analysis in this work?

The study uses the MEDSAT dataset, which integrates health determinants, sociodemographic factors, environmental measurements, and geographic information across England.

How does the framework identify the most informative predictors?

It combines variable importance approaches using knockoffs for variable filtering and then selects the top 10 variables via an average of SHAP, LOCO, Permutation Importance, and Conditional Model Reliance rankings.

What interpretable models are used to study spatial dependencies?

The work uses Generalized Additive Models (GAMs) and Multiscale Geographically Weighted Regression (MGWR) to analyze both global and local spatial relationships between variables and health outcomes.

Sep 17, 2026

Uploaded by Skyler