Preplanned Studies: Effects of Cold Spells on Mortality — Ningbo City, Zhejiang Province, China, 2014–2018

The frequency of extremely cold events has gradually decreased due to globing warming around the world, but there are cold spells caused by the continuous transfer of the Arctic polar vortex in the mid-latitude regions (1-2). As most cities in China are located in mid-latitude areas, the frequency of cold waves is expected to increase, especially in southern China. A study showed that the cold spells in 2008 swept through south-central China, resulting in a sharp mortality increase with estimated losses exceeding US $22.3 billion (3). Additionally, the varying tolerance and adaptability of populations in different regions led to inconsistencies on the definition of cold spells and its health effects in different regions (4). This study applied time-stratified case-crossover analysis to explore the associations between cold spells and mortality during different lag periods and among different population groups during cold months of 2014–2018 in Ningbo City, China. The mortality burden attributed to cold spells was also estimated. It was found that an acute effective response period appeared within about a week. Circulatory and respiratory system diseases were sensitive diseases and the elderly over 65 years old were more vulnerable. About 21.6% deaths could be attributed to cold spells.

The daily death data, meteorological data (e.g., temperature), and air pollution data (eg. PM_2.5, O₃) during 2014–2018 came from Ningbo CDC, China Meteorological Administration, and Ningbo Environmental Protection Bureau. The research period was defined as the November–March of each year from 2014 to 2018 to exclude the impact of heatwave events. We defined cold spells as days when the daily mean temperature was at or below the P₁₀ (5.5 ℃) or P₅ (3.9 ℃) percentile for at least 2, 3, or 4 consecutive days of the study period (1=cold spell days, 2=non-cold spell days). We compared the results of different definitions and selected the optimal one for stratified analysis to identify potential vulnerable populations and sensitive diseases.

The associations between cold spells and mortality were investigated in a two-stage analysis. First, the associations between cold spells and mortality were estimated by using a time-stratified case-crossover design combined with a distributed lag non-linear model (DLNM) (5), controlling for relative humidity and air pollutants. In order to identify the temporal characteristics of cold spells, we assessed the associations during different lag periods. Stratified analysis by sex, age, and cause of mortality were also conducted to identify sensitive diseases and vulnerable populations. Second, the attributable fractions (AFs) were estimated according to the associations between cold spells and mortality to evaluate the attributable mortality burden of cold spells (6). In addition, the stability of the model was validated by conducting sensitivity analysis (Supplementary Table S1). All analyses were implemented by using R statistical software (version 4.0.2, The R Foundation for Statistical Computing, Vienna, Austria).

During the study period, the total number of deaths was 83,532. The average daily mean temperature was 9.1 ℃ and the daily mean deaths of all causes were 120±19 per day (Table 1). This study shows the summary information of cold spells under different definitions (Supplementary Table S2,). Under the same temperature threshold, the shorter the duration, the more cold spells and days would occur.

Figure 1 depicted the lag responses of associations between cold spells under 6 different definitions and the mortality. All cold spells had a non-linear effect on the risk of total death. The death risk showed a trend of decreasing first and then increasing, and it was the largest on the day of exposure (lag0).

According to the lag effects, the cumulative relative risks (CRR) of total mortality associated to cold spells under different definitions were obtained for different lag periods (lag0, lag0–7, lag0–14, and lag0–21) (Supplementary Table S3). Based on the value of CRR, we selected “cold spell A” (temperature threshold ≤P₅, duration ≥2d) as the best cold spell definition to conduct the stratified analysis by sex, age, and cause of death. Under the “cold spell A”, the value of CRR increased with the increasing of lag periods. While a sharp increment of CRR value on lag0–7 was detected (Table 2), which indicated that there was an acute effect period about 7 days after the cold spell appeared.

The CRR for males and females were 1.322 [95% confidence interval (CI): 1.171, 1.493] and 1.220 (95% CI: 1.068, 1.394), respectively, within 21 days after cold spell happened compared with non-cold spell periods. The people above 65 years old increased the most when cold spell appeared, the CRR was 1.325 (95% CI: 1.186, 1.481). No statistically significant association was found for the people of 0–14 years, and the CRR was 1.772 (95% CI: 0.613, 5.120).

The CRR of death from respiratory diseases and circulatory diseases were 1.444 (95% CI: 1.173, 1.777) and 1.465 (95% CI: 1.261, 1.702) on lag0–21, respectively. No statistically significant associations were found between cold spell and mortality of genitourinary and endocrine system diseases.

It was shown in Table 2 that 21.6% (95% CI: 13.3%, 29.1%) of deaths could be attributed to cold spells during cold season in Ningbo. The attributable fraction (AF) for males and females was 24.4% (95% CI: 14.6%, 33.0%) and 18.0% (95% CI: 6.4%, 28.3%), respectively. Among different age groups, the population over 65 years old had the highest death burden ascribed to cold spell, with the AF of 24.5% (95% CI: 15.7%, 32.5%). The AF value of respiratory system diseases and circulatory system diseases was 30.7% (95% CI: 14.7%, 43.7%) and 31.7% (95% CI: 20.7%, 41.2%), respectively.