Background: Studies on high temperatures and mortality have not focused on underdeveloped tropical regions and have reported the associations of different temperature metrics without conducting model selection.
Methods: We collected daily mortality and meteorological data including ambient temperatures and humidity in Ahmedabad during summer, 1987-2017. We proposed two cross-validation (CV) approaches to compare semiparametric quasi-Poisson models with different temperature metrics and heat wave definitions. Using the fittest model, we estimated heat-mortality associations among general population and subpopulations. We also conducted separate analyses for 1987-2002 and 2003-2017 to evaluate temporal heterogeneity.
Findings: The model with maximum and minimum temperatures and without heat wave indicator gave the best performance. With this model, we found a substantial and significant increase in mortality rate starting from maximum temperature at 42 °C and from minimum temperature at 28 °C: 1 °C increase in maximum and minimum temperatures at lag 0 were associated with 9.56% (95% confidence interval [CI]: 6.64%, 12.56%) and 9.82% (95% CI: 6.33%, 13.42%) increase in mortality risk, respectively. People aged ≥65 years and lived in South residential zone where most slums were located, were more vulnerable. We observed flatter increases in mortality risk associated with high temperatures comparing the period of 2003-2017 to 1987-2002.
Interpretation: The analyses provided better understanding of the relationship of high temperatures with mortality in underdeveloped tropical regions and important implications in developing heat warning system for local government. The proposed CV approaches will benefit future scientific work.
Keywords: Climate change; Cross-validation; Extreme heat; India; Mortality.
Copyright © 2021 Elsevier Inc. All rights reserved.