The Impact of Heat Waves on Emergency Department Visits in Roanoke, Virginia

Abstract

Extreme heat, often coupled with high humidity, is a source of environmental stress to the human body. The resulting strain can lead to morbidity or mortality. Prior heat-related research has emphasized the impact of heat events on mortality, but there is a growing interest in understanding the influence of heat on emergency department (ED) visits, hospitalizations, and other disease metrics.1, 2 With increasing global temperatures, the number and/or intensity of heat events is likely to increase. Because heat is underreported as a potential cause of morbidity, traditional data set filtering by heat stress codes results in a significant underestimate of the actual heat impact. Furthermore, short-term surges in ED visits associated with heat can overtax the delivery of health services and place patients at undue risk. Given the ability of meteorologists to accurately forecast heat waves and the potential to use that knowledge in ED scheduling, understanding the relationship between heat wave events and ED visits is a potentially important, yet often unconsidered, factor in ED staffing. The purpose of this study was to determine the extent to which ED visits were associated with heat waves. In addition to the total daily ED visit count, we examined age, demographic (race/sex) groups, and discharge diagnoses and their relationship to heat waves. This retrospective study utilized daily ED visit encounters from 2010 to 2017 for residents of the Roanoke, Virginia, region who visited hospitals operated by Carilion Clinic, a six-hospital system in southwest Virginia. The Carilion Clinic Institutional Review Board deemed this study as having minimal risk to study subjects and did not require individual subject consent. Patient-level data acquired from an electronic data patient archive included basic demographic data (age, sex, race/ethnicity), zip code, date of service, and discharge diagnosis. To be included in the study data set, patients must have lived in Roanoke or the surrounding area during the study period. Inclusion was based on the patient zip code tabulation area (ZCTA) of residence. The full data set was subdivided by age class, sex, and race/ethnicity. The latter variable was coded as white/nonwhite given that small sample sizes for specific nonwhite groups impacted statistical robustness. We examined CCS level one discharge diagnoses (17 categories) based upon the appropriate ICD-9 or ICD-10 code. ICD-9 codes were converted to the appropriate ICD-10 code using publicly available tools (http://www.icd10codesearch.com). When code comparisons were unclear, they were mapped to the closest clinically appropriate term. Weather data were downloaded from archived measurements taken at the Roanoke airport (station code ROA) by the National Weather Service. We identified 54 ZCTAs in the vicinity of the ROA weather station that covered an area of approximately 2400 km2. Weather data were retrieved at the observation time closest to 1 p.m. local standard time and included air temperature, wet bulb temperature, and dew point temperature. Apparent temperature (AT) was calculated from the raw data. AT is a combination of air temperature and humidity used to estimate the added stress placed on the body during hot conditions based on the efficiency of evaporative cooling. AT is commonly used in heat-health studies and is statistically equivalent to the popular “Heat Index.”2, 3 We used a time-stratified case-crossover approach in which ED visit counts during heat waves (with lags of up to 2 days) were compared to a prior control period that was not part of any heat wave. We defined heat waves as 3 or more days exceeding a 1 p.m. AT threshold of 33°C with no more than 1 intervening day below the threshold.2 The 33°C threshold was chosen based on the overall warm-season AT distribution and prior research in the region.2 The 33°C AT was exceeded 9.3% of the time from April to September at ROA. Total ED visits during heat waves were compared to control periods prior to each heat wave. With the time-stratified case-crossover approach, each person serves as their own control.3 Odds ratios (OR) and their 95% confidence intervals (CIs) were calculated using the “casecross” function in the “season” package in R, which utilizes conditional logistic regression to estimate the OR.4 To provide more complete information about the significance of the relationships, we used two different control periods. In one method (defined as “Control 1”), the control was the period of identical duration immediately prior to the onset of the heat wave (e.g., a 4-day heat wave that began on a Tuesday was paired with a 4-day control ending on Monday). Since this method does not account for possible day-of-week variations in ED visit patterns, a second control (“Control 2”) was defined as the identical period 1 week prior to the heat wave. Controls were only used if they were not also part of a heat wave. Cases for which no appropriate control period could be identified were removed from the analysis. The identical procedure was repeated for 1- and 2-day lags. Results demonstrated a mean (±SD) of 167 (±19) ED encounters per day over the 8-year period of record. Higher ED visits were evident in September and on Mondays, whereas lower frequencies occurred in November and February and on weekends. Based on a 1 p.m. AT threshold of 33°C, there were 16 heat waves (an average of two per year) over the study period. The odds of an ED visit during a heat wave were 6% to 7% higher than during the prior, non–heat wave period (Table 1). For this facility, this translates to an additional 10 to 12 patients per day. This effect size for total visits is consistent with other published studies on the impact of heat on morbidity.2, 5 Heat wave–related ED visit rates were higher for males than females (OR = 1.10 vs. 1.05), but both groups showed statistically significant effects (Table 1). ED visits were elevated during high heat periods for whites and nonwhites, for patients in the 1- to 4-, 10- to 19-, and 50- to 64-year age groups and for patients diagnosed with infectious, mental, respiratory, circulatory, and genitourinary diseases (Table 1). In almost all cases, stronger effects were found at lag zero than at longer lags. Our data demonstrated ED visit rates to be higher for the male and nonwhite subgroups during heat events. Prior research has demonstrated a prevalence of heat impacts on the male and preteen/teen population. This finding has been related to outdoor work and/or exertional factors.6 Prior studies have had inconsistent findings regarding differences in ED visits during heat events with respect to race/ethnicity.7, 8 Heat has been associated with higher ED visits for hypotension,7 ischemic heart disease,7 and stroke. 7, 9 Likewise, respiratory disease increases, while most common in winter, have also been associated with heat events in some studies.9 Finally, higher genitourinary ED visits during heat events have been commonly linked to renal failure.5, 7-9 We were unable to find any consistent relationships between heat events and either infectious disease or mental disorders in the literature. A fundamental limitation of the retrospective study structure is the lack of information on the heat exposure of individual patients. This exposure varies greatly depending on individual work and home environments and may be affected by factors such as building quality and access to air conditioning. Although the exact exposure of each person is unknown, the short time window used in the case-crossover approach assures that, on average, individual exposure before and during heat events would be similar for most subjects. For the groups with smaller sample sizes, the type of control selected impacted statistical significance; however, for the larger categories, the results were generally consistent. Although these results are specific to this location and population and may not be generalizable to other groups, the method used here could be applied to other EDs and regions. Additionally, using administrative data in research has unique limitations given its primary purpose in billing patients. This approach can lead to lack of specificity in coding, misclassification, and the omission of comorbidities and other information. Heat waves can be accurately forecasted at least several days in advance and are rarely surprise occurrences. Given that heat has a demonstrable impact on ED visits, which in some cases can exceed the baseline rate by 10%, heat index and heat wave forecasts could be employed to affect advanced plans for ED staffing. As the climate warms and heat waves become more common, our findings suggest that heat-related ED encounters will likely increase, resulting in higher ED patient volumes. ED surges and crowding are associated with poorer patient outcomes and reductions in the quality of health care delivery. The cost of care is also relatively high in the ED. Consequently, ED surges could implicitly shift health care delivery into a higher cost of care model. Developing weather forecast–based predictive models could improve the ability of clinics to staff EDs during heat wave surge periods and to implement strategies to reduce the ultimate health care impacts of high heat events. By exploiting established relationships between heat and the course of disease, EDs can be strategically positioned to handle increased patient volumes that will occur during future heat events. The authors thank David M. Hondula (Arizona State University) and Philip J. Stenger (University of Virginia Climate Office) for their assistance with various aspects of this research.

Topics & Concepts