A data-driven framework for the definition of heat in CV research for better comprehension of climate changes on human health

Abstract Background Recent literature has identified numerous combinations of meteorological indicators, methods and comparison thresholds to define heat in order to analyse its impact on adverse cardiovascular (CV) events. In fact, despite increasing attention to the issue due to ongoing climate change, there is no standardised definition of heat nor guidelines for its establishment, and in studies often little attention is given to the selected definition, despite its impact on the results. Purpose To fill this gap, our aim was to propose a data-driven framework for selecting the temperature threshold for defining heat in the context of studies on adverse CV outcomes, combining the most common approaches identified in the literature. Methods The following five-step framework is proposed: 1) Selection of temperature lags based on the contour plot obtained by the distributed lag non-linear model (DLNM); 2) Estimation of the relative risk (RR) of having a CV emergency in heat versus non-heat days by Poisson regression considering the 90th-99th percentile (%ile) of the temperature distribution for each lag; 3) Selection of the final lag based on the curvatures of the 10 included %iles; 4) Calculation of the % change in RR from %ile i to %ile i+1; 5) Selection of the heat threshold, defined as the %ile from which all the next %iles are associated with a higher % change, or, if not existing, the %ile preceding the one with the highest % change. We tested this framework, for each year, on retrospective CV emergencies in May-September 2017-2022 in Milan, Italy, using the apparent temperature data, accounting for air temperature, relative humidity and wind speed. Results The contour plot of the DLNM, controlled for air pollution, seasonality and long-term trend, showed that lags 0 and 1 were characterised by an increased RR for a CV emergency. The RR estimated for each %ile for each considered lag (i.e., heat on day i, heat on day i or on day i-1, and heat on day i and on day i-1), produced a similar curvature, so first lag option was arbitrarily selected. For Milan, the 95th %ile was the point from which all the subsequent %iles were associated to progressively higher % changes, resulting in a RR of having a CV emergency compared to non-heat days equal to 1.113 (95% CI: 1.087-1.139). The mean daily apparent temperature values associated to such %ile, varying from 27.7 °C to 29.9 °C depending on the year, should then be considered as the heat threshold in the examined context. Conclusion We proposed a flexible data-driven framework for the definition of heat that could easily be extended to similar scenarios. Although the framework still requires some arbitrary choices to be made, and the need of validation on other real-world data, it could provide a valuable guideline for relevant research, and could contribute to the standardization of heat definition for CV health research aiming at a better comprehension of climate changes on human health.