Navigating climate dynamics: Epidemiological modeling examines the mortality impact of variable temperatures in Europe

In a recent study published in The Lancet Regional Health, researchers evaluate the impact of temporal data aggregation on estimating the temperature-mortality relationship in Europe using weekly and monthly data as alternatives to daily health records.

Study: The effect of temporal data aggregation to assess the impact of changing temperatures in Europe: an epidemiological modelling study. Image Credit: Zyn Chakrapong / Shutterstock.com

Background 

Exposure to ambient temperatures significantly increases morbidity and mortality, particularly in vulnerable populations. Approximately 9.43% of global urban mortality is due to sub-optimal temperatures, 8.52% of which are due to cold temperatures and the remaining 0.91% due to heat. These figures are lower for Europe, where temperature accounts for 7.17% of all deaths.

Short-term health outcomes resulting from temperature are typically assessed in a distributed lag non-linear model incorporating lagged and non-linear temperature health. Nevertheless, obtaining daily health records for these analyses is often difficult due to legal and bureaucratic barriers, thus narrowing the scope and depth of research.

To overcome these challenges, weekly or monthly health data that might be readily available could be used. Thus, further research is needed to ascertain if aggregated weekly or monthly health data can accurately estimate temperature-related mortality and ultimately overcome the limitations of accessing daily health records.

About the study

The current study utilized a mortality database covering 27,444,314 daily all-cause mortality counts from 1998 to 2004 within 147 regions across 16 European countries, including both urban and rural populations totaling over 400 million people. The represented countries included Croatia, Austria, Belgium, Czech Republic, Denmark, Germany, France, Italy, Luxembourg, the Netherlands, Poland, Portugal, Slovenia, Spain, Switzerland, and the United Kingdom (UK) (specifically, regions in England and Wales).

This comprehensive dataset was free of missing values. Daily regional average temperatures were derived from gridded European Climate Assessment & Dataset (ECA&D) Observational (E-OBS) data.

The analysis period was confined to a specific seven-year stretch, thus ensuring a consistent weekly, two-weekly, and four-weekly Monday-to-Sunday cycle for data aggregation. Daily regional temperature and mortality data were averaged into these periods for separate epidemiological modeling.

In stage one, quasipoisson regression was performed for each region, thereby modeling the location-specific temperature-lag mortality relation. One such set of models differed in daily, weekly, two-weekly, and four-weekly temporal resolutions, including intercepts and a natural cubic spline for seasonal and long-term trends and cross-basis function for temperature-mortality relationships.

The daily model also controlled for day-of-week effects. Natural cubic splines were used as exposure- and lag-response functions for each temporal scale in these models.

In stage two, multilevel meta-regression analyses were used to pool location-specific coefficients derived from the initial stage. The analyses consisted of random effects by country and average temperatures as predictors.

This analysis generated zone-specific minimum mortality temperatures (MMTs), which were converted into local temperature and mortality statistics as fatal numbers and percentages. This approach was used to measure deaths due to heat and cold using the mean temperature deviation from the MMT.

Using empirically based Monte Carlo simulations, national and European burdens were aggregated from regional data. This allowed for a thorough assessment of temperature-related mortality across Europe that considered the challenges of accessing daily health records.

Study findings 

The exposure-lag-response association varied significantly under the four data models for daily, weekly, two-week, and four-week exposure. An asymmetrical V-shaped daily pattern was established, whereas relative risk (RR) increased with rising divergence from MMT.

The general trend was evident in the weekly and two-weekly models, whereas the four-weekly model demonstrated more symmetrical relationships. The weekly model's lag-response relationship was similar to the daily model pattern for extreme temperatures.

The comparison between the models showed that regional MMT and RR differences increased as the level of time data aggregation rose. However, mean temperature differences were minor for the moderate correlations between daily, weekly, and two-weekly models. This resulted in no spatial relations of daily and four weekly models despite flattening RR values in the central temperatures for the model for four weeks.

The study further estimated attributable numbers (AN) and fractions (AF) for each model. The AF for weekly, two-weekly, and four-weekly periods were compared across the models.

While there was a strong linear relationship between the daily, weekly, and two-weekly models, systematic underestimations were observed, particularly for extreme temperatures. The weekly and two-weekly models tended to underestimate the lowest AF values for cold temperatures and generally underestimated hot temperature AFs. The four-weekly model showed significant deviations in both cold and heat AF estimations due to underestimation of the MMT in many regions.

Seasonal and annual variations in temperature-related mortality were also analyzed. For cold temperatures, the weekly and two-weekly models accurately captured winter mortality but underestimated it in other seasons, especially summer.

Hot temperature AFs were consistently underestimated from May to September. In absolute terms, the daily, weekly, and two-weekly models estimated significant numbers of temperature-related deaths, with notable underestimations in the weekly and two-weekly models.

Year-to-year analyses highlighted that the weekly and two-weekly models consistently underestimated total, cold, and heat AF across all years. Despite this, linear trends remained largely preserved. The underestimation percentages for total, cold, and heat-attributable mortality in these models were quantified, thus revealing significant discrepancies, particularly in the four-weekly model.

An in-depth analysis examined the relationship between annual AF values in the daily model and the differences found in the weekly and two-weekly models. To this end, a stronger correlation was observed for relative differences than absolute differences, with this relationship linear for all temperatures and cold but non-linear for heat. The weekly model's underestimation was minimal during the extreme summer of 2003, thus suggesting a smaller margin of error for less severe summers.

Conclusions

While temporally aggregated models generally maintain linear trends and capture major patterns, they tend to underestimate the impact of extreme temperatures, with the degree of underestimation varying with the level of temporal aggregation.

Written by

Vijay Kumar Malesu

Vijay holds a Ph.D. in Biotechnology and possesses a deep passion for microbiology. His academic journey has allowed him to delve deeper into understanding the intricate world of microorganisms. Through his research and studies, he has gained expertise in various aspects of microbiology, which includes microbial genetics, microbial physiology, and microbial ecology. Vijay has six years of scientific research experience at renowned research institutes such as the Indian Council for Agricultural Research and KIIT University. He has worked on diverse projects in microbiology, biopolymers, and drug delivery. His contributions to these areas have provided him with a comprehensive understanding of the subject matter and the ability to tackle complex research challenges.