Abstract

Climate change is increasing the frequency, severity, and length of extreme events such as drought, heatwave, and wildfire. Drought is one of the more complex climate phenomena that changes regionally and temporally around the world and is likely to have a detrimental influence on human health. However, there is still a lack of research on the assessment of health risks associated with drought, particularly in the United States. The purpose of this study is to assess the risk ratio of monthly drought exposure to all-cause mortality in the Northern Rockies and Plains of the United States from 2000 to 2018. We applied two-stage time-series modeling approach to estimate the location-specific and overall risk ratio (RR) of all-cause mortality associated with the United States Drought Monitor (USDM) drought. In a general population, the all-cause mortality was positively associated with the severe drought (RR: 1.050, 95% Cr: 1.034 to 1.068). Our findings suggested that the elderly, both male and female were the most affected subgroups along with both metropolitan (urban) and non-metropolitan (rural) counties. These findings highlight the need for policymakers and communities to adopt more effective drought mitigation strategies in this region.

Key words: All-cause mortality; drought; risk ratio; human health; climate change;

Introduction

Extreme weather-related events such as heat waves, droughts, and wildfire are becoming more apparent and severe due to climate change, and they are expected to have a negatively impact on human health directly or indirectly through numerous pathways. These climatic risk events are associated with an increased risk of hospitalizations and mortality including cardiovascular, respiratory, mental health, renal disease [1-3], and continue to adversely affect economic and social development worldwide [4]. Moreover, it occurs concurrently and alters environmental risk factor such as air pollution level, increasing the risk of disease and mortality [5].

Drought is one of these climate risk occurrences that has an impact on a variety of sectors, including human health, human communities, natural resources, and ecosystems [6,7]. The severity and frequency have been risen dramatically as a result of climate change [8], and lead to less precipitation, higher temperature, and an increase in the risk of multiple diseases [9-15]. Despite the enormous public health concerns, the assessment of health risk associated with drought is frequently disregarded in the United States, particularly in regional contexts, to the best of our knowledge. Berman et al. showed that mortality risk in the western United States increased with less exposure to drought during worsening drought period. A subgroup, white population between the age 25-64 in the United States, had a positive association with short-term severe drought, but not in the general population [16]. A statistically significant association was not found in the general population of Nebraska, however, longer-term drought increased all-cause mortality in white individuals with aged 25 to 34 in metro counties and aged 45 to 54 in non-metro counties [17]. Recently, a large-scale climate region-level comprehensive assessment was conducted to evaluate the risk ratio of respiratory disease associated with drought exposure [9]. They discovered that drought has a considerable impact on the risk of respiratory death, although the magnitude and direction of the risk ratio varies between climate regions in the United States. A substantial amount of research has been undertaken outside of the United States by many researchers, and we link to a comprehensive review study that assesses the danger of extreme and climate exposure in public health consequences [18].

The majority of available research has shown that not all demographics (age, race, and sex) or geographics (rural and urban) are similarly affected negatively by drought exposure. Identifying the most vulnerable subpopulations is critical for public health officials and policymakers in order to establish an effective mitigation strategy to protect communities from such hazards. Multiple characteristics, such as socioeconomic status, education level, and healthcare facilities, may not be immediately quantifiable. Geographically, research have shown that rural and farming groups are more vulnerable to the negative effects of drought exposure [19]. According to the National Center for Health Statistics (NCHS) 2013 binary rural/urban categorization [21], approximately 88.3% of the counties in the Northern Rockies and Plains region were classified as rural. Because agriculture is the primary source of income in this region, residents are more exposed to outside activity and are expected to be more vulnerable to extreme climate occurrences, such as drought.

In this article, we aim to use two-stage time-series model to investigate the overall risk ratio of all-cause mortality associated with drought exposure in the Northern Rockies and Plains region of the United States. Our primary focus is to assess the health risks by the United States Drought Monitor (USDM) with the moderate and severe drought in this region. We assess risk in the entire population as well as subgroups stratified by age, sex, and urbanity. The findings of this study will be used to build effective drought mitigation methods for future management.

Materials and Methods Health Data and Study Area

The mortality data can be requested from the Center for Disease Control and Prevention (CDC) from https://www.cdc.gov/ nchs/nvss/nvss-restricted-data.htm. To access the health data, Data Use Agreement (DUA) and its related supporting materials should be submitted by all members for research purpose.

From January 2000 to December 2018, we obtained allcause mortality data from the National Center for Health Statistics (NCHS). Using the data, we calculated monthly county-level death counts for all-cause mortality in the United States using International Classification of Diseases 10th revision (ICD-10 codes). These counts were stratified into four age groups (0-19, 2039, 40-64, and above 65 years), three race groups (White, Black, and other than white and black), and sex (male and female). There are no other demographic variables available. The surveillance, epidemiology, and end results (SEER) program provided annual county-level population estimates by demographic variables (age, race, and sex), which were used in all months of each calendar year. For our study, we extracted all-cause mortality data for the Northern Rockies and Plains region (five states: Montana, Nebraska, North Dakota, South Dakota, and Wyoming) from the complete dataset. This region includes 291 counties (257 rural and 34 urban), with rural counties accounting for 88.3% of the total.

Drought Exposure Data

The USDM has been popularly used due to the robustness compared with other drought metrics [20, 24]. The USDM is a collaborative effort between the National Oceanic Atmospheric Administration (NOAA), the U.S. Department of Agriculture (USDA), and the National Drought Mitigation Center (NDMC) that has been providing weekly updates of drought conditions since 2000 [24]. Using a convergence of evidence approach, drought authors blend moisture deficits from across the hydrological cycle (i.e., precipitation, soil moisture, evaporation, etc) with drought reports from local experts to categorize drought conditions into one of six categories: wet to normal conditions (None), abnormally dry (D0), moderate (D1), severe (D2), extreme (D3), and exceptional (D4). We then reclassified USDM measures into monthly binary and three-level categories for this assessment.

A binary measure was estimated based on the frequencies of the drought status within a given month and county. ‘No drought’ was defined if the frequencies of no drought and D₀ condition in the week within a given month and county has more than the frequencies for all D₁ to D₄ conditions. Otherwise, it is labeled as a drought condition. The three-level categorical status is determined as (i) no drought in the binary measure; (ii) moderate drought (with binary drought condition and the sum of frequencies of D₁ and D₂ are greater than that of D₃ and D₄); and (iii) severe drought (with binary condition and the sum of frequencies of D₃ and D₄ are greater than that of D₁ and D₂).

Environmental and Other Data

The NCHS 2013 binary urban/rural classification codes were used to classify counties into metropolitan and non-metropolitan [21]. Metropolitan counties were divided into four categories of Large central metro, Large fringe metro, medium, and small metro based on population size. Non-metropolitan counties were further differentiated into two subcategories: micropolitan or noncore with population less than 50,000. In this article, we use the terms urban and rural interchangeably with metropolitan and nonmetropolitan. Furthermore, monthly mean temperature anomaly was included from the NOAA’s Nclimgrid product at a 5km grid cell resolution [20]. We then calculated a monthly county level temperature anomaly using zonal averages of all grid cells falling inside a county boundary.

Statistical Modeling Two-stage Model

We used a two-stage model to estimate county-level and overall risk ratio of all-cause mortality associated with drought exposure. In the first stage, a quasi-Poisson regression model was used to estimate location-specific risk ratio. We included cubic B-spline of months with 7 degrees of freedom to control long-term time and seasonal trend, second degree polynomial of anomaly temperature, and indicator variables to control calendar year effect. The drought exposure was used as three-level categorial variables (no drought, moderate, and severe). The values of logarithm for population size in each county were used an offset variable in the model.

In the second stage, we applied Bayesian meta regression model to estimate the overall risk ratio. Specifically, we combined county-specific risk ratio by a random effect Bayesian linear regression model to evaluate the overall risk ratio. We performed the entire analysis using two-stage model and stratified by age group, sex, and urbanicity. For all analysis, Statistical Analysis Software (SAS version 14.2) with PROC GLIMMIX and PROC MCMC procedures was used.

Results Descriptive Statistics

Table 1 summarizes all-cause death counts in the Northern Rockies and Plains area from 2000 to 2018. The total number of all-cause deaths in this region was 813,121, with male having a higher mortality risk than female. The total number of deaths during the course of the study period was used to calculate the monthly mean death counts. Table 1 also shows that the observed and mean death counts increased substantially among the elderly population over 65, with females having higher mean death counts than males. Other age groups, on the other hand, followed an opposite pattern.

Table 1: Summary of all-cause death counts for study region during study period (2000-2018). N is the number of all-cause death county by each subgroup and SD indicates standard deviation.

Figure 1 presents the distribution and the trajectory of percentage experiencing binary, moderate, and severe drought events in the Northern Rockies and Plains region. The percentage was calculated using the total number of drought events for binary, moderate, or severe in the counties over the study period and summarized by states (left panel) and year (right panel). We remind the readers that the binary drought events are further divided into either moderate or severe drought events. Various spatial and temporal patterns over the entire study period were identified within the same climate region (See Figure 1).

Figure 1: The percentage of months exposed to binary, moderate, and severe USDM drought in the Northern Rockies and Plains region by year 2000-2018. Abbreviations indicate states (MT: Montana, ND: North Dakota, NE: Nebraska, SD: South Dakota, WY: Wyoming). Abbreviations for the USDM drought (B: binary, M: moderate, S: severe).

Estimate Overall Risk Ratio

We conducted a Bayesian analysis and the posterior distribution is the main inference tool. The key quantity for interpretation is referred somewhat differently. Posterior mean and the 95% credible interval (Cr) are similar interpretation to an estimate and 95% confidence interval in classical statistics. Statistical significance is determined if the 95% Cr does not include the value of one.

Figure 2 shows the posterior means and their corresponding 95% Crs of the estimated overall risk ratios by USDM moderate and severe droughts. We see in Figure 2 that the overall risk ratio of severe drought over no drought is 1.050 (95% Cr: 1.034 to 1.068), indicating that severe drought increased the risk ratio 5.0% with experiencing the severe drought. Moderate drought had the risk ratio 1.009 (95% Cr: 0.999 to 1.018), however, this association is not statistically significant. These results indicate drought severity has negatively impact on all-cause mortality risk ratio in the Northern Rockies and Plains region.

Figure 2: Overall risk ratio of the all-cause mortality risk ratios and their corresponding 95% Crs by general population and subgroups and drought types, moderate (left) and severe (right).