What is cold-related mortality? A multi-disciplinary perspective to inform climate change impact assessments

Background: There is a growing discussion regarding the mortality burdens of hot and cold weather and how the balance between these may alter as a result of climate change. Net effects of climate change are often presented, and in some settings these may suggest that reductions in cold-related mortality will outweigh increases in heat related mortality. However, key to these discussions is that the magnitude of temperature-related mortality is wholly sensitive to the placement of the temperature threshold above or below which effects are modelled. For cold exposure especially, where threshold effects are often ill-defined, choices in threshold placement have varied widely between published studies, even within the same location. Despite this, there is little discussion around appropriate threshold selection and whether reported associations reflect true causal relationships - i.e. whether all deaths occurring below a given temperature threshold can be regarded as cold-related and are therefore likely to decrease as climate warms. Objectives: Our objectives are to initiate a discussion around the importance of threshold placement and examine evidence for causality across the full range of temperatures used to quantify cold-related mortality. We examine whether understanding causal mechanisms can inform threshold selection, the interpretation of current and future cold-related health burdens and their use in policy formation. Methods: Using Greater London data as an example, we first illustrate the sensitivity of cold related mortality to threshold selection. Using the Bradford Hill criteria as a framework, we then integrate knowledge and evidence from multiple disciplines and areas-including animal and human physiology, epidemiology, biomarker studies and population level studies. This allows for discussion of several possible direct and indirect causal mechanisms operating across the range of 'cold' temperatures and lag periods used in health impact studies, and whether this in turn can inform appropriate threshold placement. Results: Evidence from a range of disciplines appears to support a causal relationship for cold across a range of temperatures and lag periods, although there is more consistent evidence for a causal effect at more extreme temperatures. It is plausible that 'direct' mechanisms for cold mortality are likely to occur at lower temperatures and 'indirect' mechanisms (e.g. via increased spread of infection) may occur at milder temperatures. Conclusions: Separating the effects of 'extreme' and 'moderate' cold (e.g. temperatures between approximately 8-9 degrees C and 18 degrees C in the UK) could help the interpretation of studies quoting attributable mortality burdens. However there remains the general dilemma of whether it is better to use a lower cold threshold below which we are more certain of a causal relationship, but at the risk of under-estimating deaths attributable to cold.