Comparing the Role of Relative and Absolute Humidity in Heat-Related Mortality: A Case Time-Series Study in South Korea

Abstract

Background: Epidemiology studies have reported inconsistent associations of humidity with heat-related health outcomes, despite strong plausibility of such physiological associations. In this regard, there has been a heated debate on which humidity metric to use in epidemiological research.

Objectives: This study aimed to compare the role of two common humidity metrics, relative and absolute humidity, in heat-related mortality in summer using a nationwide mortality dataset.

Methods: We applied a case time-series design for summer (June-September) mortality across the entire 229 districts of South Korea from 2011 to 2019. The temperature was fitted using a distributed lag nonlinear model (DLNM) with 10 lag days. A linear interaction between the cross-basis of temperature and humidity was included in each model to examine the different patterns of association between heat and mortality by humidity level (low and high humidity defined by fifth and 95th percentile of each humidity distribution).

Results: A total of 780,102 deaths were recorded in the summer from 2011 to 2019 in South Korea. The association between extreme heat (temperature approximately above the 99th percentile of the temperature distribution) and mortality was modified more by absolute humidity than by relative humidity, although the effect modification of both humidity indicators was not statistically significant. The relative risks at the 99.Ninth percentile temperature in comparison with the minimum mortality temperature were 1.21 [95% confidence interval (CI): 1.11, 1.31] and 1.22 (95% CI: 1.03, 1.44) for low and high relative humidity, respectively, and 1.11 (95% CI: 0.89, 1.37) and 1.25 (95% CI: 1.15, 1.34) for low and high absolute humidity, respectively.

Discussion: Our findings provide epidemiological evidence on the role of relative and absolute humidity in heat-related mortality and suggest that absolute humidity may be a more appropriate metric than relative humidity when assessing health impact. https://doi.org/10.1289/EHP15827.

Figure 1.

Geographical distribution of mean temperature (A), relative humidity (B), and absolute humidity (C) in the summer of South Korea, during 2011–2019. The shapefile of these maps was obtained from the GEOSERVICE-WEB (https://www.geoservice.co.kr/). Data on temperature and relative humidity were collected from the Korean Meteorological Administration, and absolute humidity was calculated from temperature and relative humidity using a conversion formula. Summer was defined as the four warmest months (June–September) based on the average monthly temperature. The summary data for this figure can be found in Excel Table S2.

Figure 2.

Cumulative association between temperature and mortality in the summer of South Korea, overall (A) and by relative (B) and absolute (C) humidity: RR and 95% CI. The shaded areas indicate 95% CIs. Summer was defined as the four warmest months (June–September) based on the average monthly temperature. The RRs were estimated by conducting a case time-series design with a fixed-effects model and a quasi-Poisson family, without (A) or with (B and C) humidity in the model, respectively. A total of 780,102 deaths was included in these analyses. Low and high humidity represent fifth and 95th percentiles of humidity distribution, respectively. The solid vertical line represents the minimum mortality temperature (20.7°C) estimated from the overall association without interaction, and the dotted vertical line presents the 99th percentile of temperature distribution (30.6°C). The summary data for this figure can be found in Excel Table S4. Note: CI, confidence interval; RR, relative risk.

Figure 3.

Cumulative association between temperature and mortality by sex and age by relative (A) and absolute (B) humidity in the summer of South Korea: RR and 95% CI. The shaded areas indicate 95% CIs. Summer was defined as the four warmest months (June–September) based on the average monthly temperature. The RRs were estimated by conducting a case time-series design with a fixed-effects model and a quasi-Poisson family. The number of males, females, people 0–64 y of age, and people $\ge 65$ y of age included in each analysis was 432,263; 347,839; 207,850; and 572,133, respectively. Low and high humidity represent fifth and 95th percentiles of humidity distribution, respectively. The solid vertical line represents the minimum mortality temperature (20.7°C) estimated from the overall association, and the dotted vertical line presents the 99th percentile of temperature distribution (30.6°C). The summary data for this figure can be found in Excel Table S11. Note: CI, confidence interval; RR, relative risk.

Figure 4.

Cumulative association between temperature and cause-specific mortality by relative (A) and absolute (B) humidity in the summer of South Korea: RR and 95% CI. The shaded areas indicate 95% CIs. Summer was defined as the four warmest months (June–September) based on the average monthly temperature. The RRs were estimated by conducting a case time-series design with a fixed-effects model and a quasi-Poisson family. The number of circulatory, respiratory, endocrine, and infectious-related mortality included in each analysis was 159,481; 74,221; 29,815; and 21,969, respectively. Low and high humidity represent fifth and 95th percentiles of humidity distribution, respectively. The solid vertical line represents the minimum mortality temperature (20.7°C) estimated from the overall association, and the dotted vertical line presents the 99th percentile of temperature distribution (30.6°C). The summary data for this figure can be found in Excel Table S13. Note: CI, confidence interval; Circulatory, diseases of circulatory system; Endocrine, endocrine, nutritional, and metabolic diseases; Infectious, certain infectious and parasitic diseases; Respiratory, diseases of respiratory system; RR, relative risk.