Table 3.
Estimates of association between greenness, measured using the normalized difference vegetation index (NDVI), and heat excess death fractions, stratified by climate zone and level of green space clustering.1
| Low Green Space Clustering | High Green Space Clustering | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Arid climate zone cities (N = 79) | |||||||||
| Greenness | Beta | 95% CI | Beta | 95% CI | AIC | p for LRT2 | Cochran’s Q | ||
| Moderate | −0.40 | −1.53 | 0.73 | −0.33 | −1.11 | 0.46 | 269.8 | 0.13 | 161.7 |
| High | −0.67 | −1.66 | 0.33 | 0.27 | −0.74 | 1.27 | |||
| Temperate and tropical climate zone cities (N = 244) | |||||||||
| Greenness | Beta | 95% CI | Beta | 95% CI | AIC | p for LRT2 | Cochran’s Q | ||
| Moderate | −0.12 | −0.46 | 0.23 | 0.00 | −0.27 | 0.27 | 775.5 | 0.34 | 550.1 (<0.01) |
| High | 0.02 | −0.37 | 0.40 | 0.02 | −0.28 | 0.33 | |||
Abbreviations: AIC, Akaike Information criterion; CI; confidence interval’; NDVI, normalized difference vegetation index.
The reference category is cities in the lowest tertile of the climate-zone specific distribution of greenness, as measured by NDVI. Moderate and high levels of greenness refer to the second and third tertiles of the climate zone specific distribution of greenness, measured by the NDVI. The results were derived from random effects meta-regression models that were adjusted for particulate matter < 2.5 µg/m3, social environment index, country group, and a measure of green space clustering. Models were run separately for arid and non-arid climate zones. Effect modification by green space clustering was assessed by including an interaction term between the three-level categorical greenness variable (NDVI) and a term representing clustering, dichotomized at the median of the climate-zone specific distribution. Clustering was measured using the FRAGSTATS 4.2 program, (McGarigal and Marks, 1995) and computed based on the spatial adjacency matrix of green patches. The clustering metric ranges between −1 and 1, where −1 indicates maximum disaggregation, 0 indicates random distribution, and 1 indicates maximum clustering.
The p-value for a likelihood ratio test (LRT) derives from a two-degree of freedom chi-square distributed test of improvement in model fit following inclusion of an interaction term between the three-level NDVI term and the two-level clustering term.