Anomalous temperature and seasonality of mortality in the United States

Abstract

Measuring and identifying the correlates and drivers of mortality are key functions of public health research. Information on seasonal mortality, and its inter-annual variations, enables prioritising and evaluating interventions during particular times of the year. It also provides the basis to study the association of death rates with seasonal temperature. My thesis aimed to provide and apply a statistical framework to characterise seasonality of death rates by cause of death, month, sex and age group, and to understand the role of temperature as a predictor, and possibly determinant, of how seasonal mortality changes from year to year and geographically. I generated death rates from geo-coded vital registration data on all 85.5 million deaths in the entire contiguous United States over a 37-year period (1980-2016). I grouped the underlying causes of death into mutually exclusive and collectively exhaustive broad causes (cardiorespiratory diseases, cancers, injuries, other) along with further sub-causes for cardiorespiratory diseases and injuries. I processed ERA-Interim reanalysis weather data to create population-weighted monthly temperature statistics. I applied wavelet, centre of gravity and circular statistics methods to analyse seasonal dynamics of mortality. I found that overall death rates for those 45 years and older were highest in the winter, mostly due to cardiorespiratory disease deaths. In contrast, death rates in adolescents and young adults peaked in the summer, mostly due to injury deaths. Seasonal differences in older age groups have changed little over time, whereas in young children they have largely disappeared. I then formulated a Bayesian spatio-temporal model to estimate how anomalous monthly temperature – defined as temperature deviation compared to long-term norm for each state and month – affects deaths from different causes. A 1°C anomalously warm year in the entire contiguous United States would be associated with an estimated 941 (95% credible interval (CrI) 831, 1,053) additional injury deaths (0.5% of total injury deaths in 2016), concentrated in adolescent to middle-aged males. There would be an estimated 4,369 (4,024, 4,706) fewer cardiorespiratory disease deaths (0.4% of total cardiorespiratory disease deaths in 2016), concentrated in those 55 years and older. There would be a decrease of cardiorespiratory disease deaths in all but summer months. I found no association between anomalous monthly temperature and cancer deaths. Continued efforts are necessary to address seasonal peaks in mortality, particularly in older age groups, across the United States, especially in a changing climate.