Background: Within major metropolitan areas there is spatial variability in heat-related mortality, but it is not known whether the places associated with high mortality are consistent across time. Temporal changes in geographic patterns of heat risk would have important implications for intervention strategies and projections of health outcomes in future climates. Aims: We sought to quantify the predictability and temporal variability of high-risk zones for heat-related mortality in seven climatically diverse cities in the United States. Our hypothesis is that high-risk zones can be reliably predicted from historical data. Methods: 25 years of daily postal code-level counts of mortality data were split into training and testing data sets. A two-stage modelling approach was used to identify zones within each city with significantly high mortality during the training period. These zones were labelled as hypothetical "targets" for intervention activities. We then evaluated model skill in predicting high-risk zones in advance. We also used cross-validation to quantify the overall temporal variability irrespective of chronological sequence. Confounding variables including seasonally and long-term trends were accounted for with a generalized additive model. The study was limited to days that were associated with both high temperatures and high mortality rates at the city-wide scale. Results: Heat-related mortality in the testing period was found to be significantly higher in target zones compared to non-target zones. Target zones were also more likely to be high mortality zones in the testing period than non-target zones. In most cities the number of zones of high heat-related mortality declined from the training to the testing period. Conclusions: Geographically-targeted long-term intervention measures aimed at reducing the public health burden associated with heat stress may be effective because the location of high-risk zones is relatively consistent over time.
展开▼
