Numerous epidemiological studies have demonstrated a significant association between ambient temperature and population health, but evidence is limited for the health impact of temperature variability between neighboring days.

The study compares exposure-response associations of years of life lost (YLL) with different indicators of temperature variability between neighboring days, including temperature change between neighboring days (TCN, difference of mean temperature between neighboring days), temperature variability (TV, the standard deviation of maximum and minimum temperatures between neighboring days), and total temperature variability between neighboring days (TTV) according to the directions and effects of temperature variability between neighboring days which was developed in the present study. The study aims to explore which measure can better assess the impact of temperature variability between neighboring days on mortality.

Death registration data and meteorological data during 2013-2017 were collected from 40 districts/counties in Guangdong, China. The exposure-response association of diurnal temperature range with YLL rate (YLL per 100 000 population) and the association of nocturnal temperature range with YLL rate were investigated using a two-stage approach including distributed lag non-linear model (DLNM) and multivariable meta-analysis. Then TTV was weighted by attributable YLL rate of diurnal temperature range and nocturnal temperature range. The correlations of the three indicators of temperature variability between neighboring days were examined by Pearson correlation analysis. The exposure-response associations of YLL rate with TCN, TV, and TTV were evaluated by DLNM model and multivariable meta-analysis. The effects of different indicators of temperature variability between neighboring days on mortality were compared.

The daily average YLL rate of the 40 study locations in Guangdong Province was 22.3 per 10⁵ inhabitants during the study period. The means of TCN, TV, and TTV were (0.0±1.8)℃, 4.6±1.5, and (8.1±2.7)℃, respectively. These indicators all approximated a normal distribution. TCN had a weak correlation with TV and TTV (r=0.097 9, r=0.088 0), and TV had a strong correlation with TTV (r=0.889 1). The exposure-response association between TV and YLL rate was insignificant after controlling the overall lag effect of temperature, but TV and TTV were statistically associated with YLL rate. Both the TV-YLL and the TTV-YLL exposure-response relationships were U-shaped, suggesting that both ends of TV and TTV increased YLL rate in the study population. The attributable YLL rates of extremely low TV (P₅, TV=2.2) and extremely high TV (P₉₅, TV=7.2) were 1.0/10⁵ (95% CI:0.1/10⁵-1.9/10⁵) and 3.1/10⁵ (95% CI:1.2/10⁵-5.1/10⁵) respectively, and both were lower than the attributable YLL rates of extremely low TTV (P₅, TTV=2.1℃) (2.1/10⁵, 95% CI:0.2/10⁵-4.0/10⁵) and extremely high TTV (P₉₅, TTV=12.1℃) (4.1/10⁵, 95%CI:2.3/10⁵-5.8/10⁵). The effects of moderately low or high TV and TTV were similar.

Associations of YLL with TCN, TV, and TTV are inconsistent. TTV takes both degree and direction of temperature variability into account, and is a better predictor of the impact of temperature variability between neighboring days on human health.