Composite Environmental Vulnerability Index (EnvScore)

• Composite Environmental Vulnerability Index (EnvScore) is an integrated measure that aggregates food environment, pollution, and socioeconomic factors to quantify environmental risks related to public health.
• It employs normalized multi-domain indicators and both equal-weight and health-outcome weighted methods to ensure comparability across various spatial and temporal scales.
• EnvScore guides targeted policy interventions and epidemiological research by revealing vulnerability hotspots and directing resource allocation.

A Composite Environmental Vulnerability Index (EnvScore) is an aggregated quantitative measure designed to summarize and compare the relative environmental vulnerability of geographic units by integrating multi-domain indicators that affect public health outcomes, particularly in relation to structural determinants such as pollution, food access, socioeconomic disadvantage, and climate hazards. EnvScore methodologies, as seen in established frameworks for both the United States and Australia, provide interpretable metrics that support epidemiological research, inform policy interventions, and scale across spatial and temporal resolutions (Mamillapalli et al., 28 Dec 2025, Price et al., 2024).

1. Theoretical Motivation and Conceptual Framework

Composite environmental vulnerability indices arise from the recognition that public health risks, including childhood obesity and acute mortality, are strongly shaped by a confluence of individual, socioeconomic, and structural environmental factors. These indices operationalize the concept of "obesogenic" and hazardous environments by quantifying domains such as food access, pollution exposure, socioeconomic stressors, and adaptive capacity (Mamillapalli et al., 28 Dec 2025, Price et al., 2024). The structural aggregation supports research rooted in socio-ecological models (e.g., Swinburn et al. 2011; Kumanyika 2019), which posit that complex layering of disadvantage elevates risk at the population level.

2. Domains and Data Sources

EnvScore construction employs variables spanning three core domains:

1. Food Environment: Metrics include state-level proportions of low-income and low-access census tracts (LILATracts_1And10), poverty rate, median family income, and vehicle access in disadvantaged tracts, extracted from USDA Food Access Research Atlas (Mamillapalli et al., 28 Dec 2025).
2. Pollution Burden: Air quality indicators, including good/moderate/unhealthy AQI days, median and maximum AQI, and ozone thresholds, are drawn from EPA AQS Data Mart (Mamillapalli et al., 28 Dec 2025). In the Australian context, additional exposure variables such as NO, NO₂, O₃, PM₂.₅, excess heat/cold factors (EHF/ECF), and historical temperature percentiles are included for fine-scale spatio-temporal monitoring (Price et al., 2024).
3. Socioeconomic Sensitivity and Adaptive Capacity: Australian implementations add over forty socio-demographic and health-status variables—such as population density, education, age structure, chronic disease rates, and housing characteristics (Price et al., 2024). Adaptive capacity integrates hospital beds, greenspace percentage, water bodies, NDVI, vehicle and internet access.

These multi-domain indicators are temporally and spatially tagged, with granular aggregation possible at weekly, monthly, or annual resolutions and spatial scales such as US states, or Australian SA2/SA3/LGA regions.

Domain	Example Indicators	Data Source
Food Environment	PovertyRate, LILATracts_1And10	USDA Food Access Atlas
Pollution Burden	Median AQI, Days Ozone, PM₂.₅	EPA AQS Data Mart
Socioeconomic/Adaptive	Age, Income, Disease Rates, Hospitals, NDVI	AusEnHealth, Census

3. Preprocessing and Normalization

To enable coherent aggregation, all input indicators are normalized, typically utilizing min–max scaling: $$Z_{i,s} = \frac{x_{i,s} - \min_{s'} x_{i,s'}}{\max_{s'} x_{i,s'} - \min_{s'} x_{i,s'}}$$ This places all variables on the unit interval [0, 1], preserving relative rank and spacing and ensuring that variables with disparate units (e.g., dollars vs. proportions) contribute equitably (Mamillapalli et al., 28 Dec 2025).

Percentile-based normalization is also utilized, especially in high-dimensional implementations, where for region $i$ and time $t$: $$f_{i}(\underline{x}_{\,t}) = \frac{\mathrm{rank}(x_{i,t};\, x_{1,t},\dots,x_{N,t})}{N+1}$$ Variables are then grouped and averaged within themes to prepare for aggregation (Price et al., 2024).

4. Aggregation: Weighting and Composite Formulation

The aggregation method is central to the interpretability and robustness of EnvScore:

• Equal-Weight Averaging: US implementation aggregates $n=10$ normalized indicators with simple arithmetic mean: $EnvScore_{s} = \frac{1}{n} \sum_{i=1}^n Z_{i,s}$ No principal components analysis or optimization is applied; each domain contributes equally, enabling transparent apportionment and facile adjustment for policy analysis (Mamillapalli et al., 28 Dec 2025).
• Health-Outcome Weighted Index: Australian models utilize weighted averages based on the empirical association of each variable with an observed health outcome (e.g., age-standardized mortality), as indexed by Kendall's $\tau$:
  • Compute $$w_{n,k} = \text{Kendall’s } \tau(x_{n,k,\cdot}, Y_\cdot)$$.
  • Construct theme-level and sub-index scores, aggregate as: $$wVI_{i,t} = \frac{1}{3} \sum_{k=1}^3 f_i(WS_{\cdot,k,t})$$
  • Where $WS$ are weighted sub-indices across domains (Price et al., 2024).

Comparisons show that health-outcome weighting achieves higher correlation with target outcomes and mitigates overrepresentation risk from highly correlated variable clusters. Equal-weight approaches facilitate interpretability and policy modification but may obscure domain prioritization in health-targeted settings.

5. Temporal and Spatial Resolution

EnvScore methodologies support multi-scale analysis. Raw data (e.g., daily AQI, 3-hourly pollutant measures) is aggregated to the desired temporal resolution; demographic and built environment characteristics are handled as static or imputed for periods within a year if only annual data are available (Price et al., 2024). Percentile ranks, normalization, and aggregation are computed within each spatial stratum to enable comparability and cross-region analysis at the selected scale (state, LGA, SA2/SA3).

Time-series EnvScore outputs reveal acute spikes due to episodic hazards (e.g., heatwaves) not captured by annual averaging.

6. Validation, Alignment, and Example Outputs

Validation employs correlation structures, clustering analyses, and cross-scale alignment checks:

• Structural Robustness: Pairwise correlation heatmaps among indicators confirm absence of perfect collinearity; k-means clustering finds vulnerability clusters consistent with known public health regions (e.g., "Obesity Belt") (Mamillapalli et al., 28 Dec 2025).
• Outcome Alignment: EnvScore displays clear monotonic association with modeled obesity risk probabilities, with high-vulnerability states in the US South and Midwest matching elevated micro-level predicted risk (Mamillapalli et al., 28 Dec 2025).
• Face Validity: Weighted indices show higher concordance with mortality burdens than simple percentile sums (Price et al., 2024).

Example state-level outputs for highest US vulnerabilities: | State | EnvScore | PovertyRate (%) | LILATracts_1And10 | Median AQI | Days Ozone | |-------------|----------|------------------|--------------------|------------|------------| | Mississippi | 0.550 | 24.83 | 0.347 | 42.60 | 59.20 | | Arkansas | 0.503 | 20.09 | 0.292 | 41.73 | 71.73 | | Louisiana | 0.486 | 22.58 | 0.268 | 39.96 | 100.86 | | Alabama | 0.448 | 21.21 | 0.249 | 41.43 | 23.79 |

7. Policy Implications and Analytical Utility

Composite EnvScore provides a scalable, transparent screening instrument for identifying regions with convergent structural disadvantage and targeting health interventions. Its equal-weight design facilitates policy iteration and addition of new indicators, while health-outcome weighted versions optimize direct relevance for burden assessment (Mamillapalli et al., 28 Dec 2025, Price et al., 2024).

Overlaying EnvScore with micro-level predictions exposes latent geographic "hot spots" even where individual-level data lacks state identifiers, guiding multi-sector resource allocation and prioritization of intervention domains (e.g., food access, poverty reduction, air-quality strategies).

This suggests continued refinement may expand EnvScore utility to causal modeling, real-time analytics, and more nuanced adaptation to context-specific health outcome targeting. A plausible implication is that future developments should calibrate weighting models to both acute and chronic public health endpoints and explore spatio-temporal dynamics at fine geographical granularity.