Short-term Outdoor Temperature Change and Emergency Department Visits for Asthma Among Children: A Case-Crossover Study

Abstract

Although weather changes are known to cause asthma symptoms, their impact on asthma-related health-care utilization is poorly understood. The objective of the present study was to determine the association between short-term outdoor temperature change and asthma-related emergency department (ED) visits among children 3–18 years of age in Detroit, Michigan, in 2000–2001. Descriptive analyses of patient and ED visit characteristics were performed. A case-crossover study utilizing time-stratified controls was conducted to determine the impact of maximum temperature change and change rate measured during 4-, 8-, 12-, and 24-hour periods. Multivariable conditional logistic regression demonstrated the relation between ED visits and temperature change after controlling for other weather and pollutant measures. There were 4,804 asthma-related ED visits during the study period, and they occurred most frequently in the fall and during morning hours. The case-crossover study showed a statistically significant inverse relation between ED visits and maximum 24-hour temperature change after adjustment for climatic factors (for temperature change, odds ratio = 0.992, P = 0.04; for temperature change rate, odds ratio 0.972, P = 0.01). The association persisted after air pollutant measures were added to the model, although the association was not significant. Despite the finding that a greater 24-hour temperature change decreased the risk of asthma-related ED visits, the overall results suggested a negligible association with short-term temperature change.

asthma, child, cross-over studies, emergency, temperature, weather

Asthma is a chronic condition characterized by airway hyper-responsiveness to physiologic or environmental triggers. This hyper-responsiveness results in pronounced constriction of airway muscles, inflammation, swelling, mucus production, and subsequent respiratory distress. Research has documented a dramatic increase in asthma prevalence during the past 30 years in many parts of the world. Many demographic factors are associated with asthma, including age, sex, birth order, season of birth, ethnicity, region, and country (1). Minority groups and those in low socioeconomic groups have higher prevalences of asthma (1–3).

Numerous asthma triggers have been identified, including both indoor and outdoor environmental exposures. However, additional study is needed of factors that trigger the asthmatic response that leads to severe asthma outcomes. Of particular interest are the triggers that are the least amenable to moderation or elimination, such as weather conditions, because they have implications for patient education and maintenance of asthma control. Weather conditions, especially cold temperature, are often cited as triggers of asthma symptoms by patients and health-care providers. Anecdotally, emergency department (ED) physicians have commented that an influx of asthma patients occurs after a nighttime drop in temperature. There has been research documenting “epidemics” of asthma exacerbations related to thunderstorm events (4–6), cold temperature, low barometric pressure, high relative humidity, and fog presence (7–12). These studies, however, focused on climatic status, both on the day of the event or lagged, and long-term changes in weather conditions by measuring exposures over days and weeks. To our knowledge, no previous study has explored the critical window of exposure to temperature change using exposure periods as short as 4 to 24 hours. Also unique to our study are the simultaneous considerations of the temperature change value and the rate of temperature change.

The objective of this research was to determine whether there is an association between short-term outdoor temperature change and hospital ED visits for asthma among children. Further, this project assessed whether temperature change or the rate of temperature change and subsequently which exposure period is important in predicting asthma-related ED visits in children.

MATERIALS AND METHODS

Health data

ED data were collected for the years 2000 and 2001 using administrative billing records from the Children's Hospital of Michigan (CHM). Located in Detroit, Michigan, CHM serves a largely black, socioeconomically disadvantaged population residing in the inner city. An asthma-related ED visit was defined as a visit that resulted in a principal diagnosis of asthma at discharge from the ED (International Classification of Diseases, 9th Revision, Clinical Modification, codes 493.00–493.99) and did not result in an inpatient admission. Exposure was assigned based on ED registration time. In the administrative billing data set, when an ED registrant is subsequently admitted to the hospital, the registration time is removed from the patient record. Therefore, ED visits that resulted in an inpatient admission were omitted from the analyses. The study population was restricted to children 3 to 18 years of age; patients less than 3 years of age were excluded to minimize the potential for information bias due to diagnostic misclassification.

For each asthma-related ED visit, the following variables were collected from the administrative billing record: date of ED visit, time at registration, primary diagnosis, zip code of residence, date of birth, sex, and race. From the date and time of ED registration, the following categorical variables were created: time of day in 3-hour increments, day of the week, month, and meteorological season of the visit. Meteorological season was defined as winter (December, January, or February), spring (March, April, or May), summer (June, July, or August), or fall (September, October, and November).

This project was submitted to both the Tulane Health Sciences Center Institutional Review Board and the Wayne State University Human Investigation Committee for review. Both committees found it to be exempt research in accordance with 45 CFR 46.101 (b) (4).

Weather data

Hourly meteorological data for 2000 and 2001 were acquired from 3 National Oceanic and Atmospheric Administration weather stations located in the Detroit area at 3 airports: City Airport, Willow Run Airport, and Detroit Metropolitan Airport. The 2-year period of study was chosen to minimize the impact of an abnormal year, ensure variability in the temperature change exposure variables, and minimize the impact of unmeasured confounders. Exposure was assigned for each asthma-related ED visit based on the weather station nearest to the centroid of the patient's residential zip code reported at the time of the visit using ArcGIS, version 9.2 (ESRI, Redlands, California).

Two exposure variables were assessed: temperature change and temperature change rate. Temperature was measured hourly. We defined temperature change as the difference between 2 temperature values measured at varying intervals over the exposure periods. Temperature change rate was measured as the difference in temperature values divided by the duration of time over which the change took place. For both, the maximum value was designated as the study exposure. Because the appropriate period of time in which to measure an ecological exposure for asthma-related ED visits is generally unknown, the present investigation explored increasing exposure periods relative to the ED event time. ED visits were grouped in 3-hour blocks based on the registration time. The time periods for exposure were defined as 4, 8, 12, and 24 hours preceding the 3-hour block for the outcome. During these exposure periods, the maximum absolute temperature change and maximum absolute temperature change rate (per hour) were determined.

Relative humidity (%), barometric pressure (millibars), wind speed (mph), thunderstorm activity, and the presence of precipitation in the form of rain or drizzle were considered as covariates. Data for each were obtained from the 3 weather stations. For relative humidity, barometric pressure, and wind speed, the maximum values for each 4-, 8-, 12-, and 24-hour exposure period were used for analyses. Thunderstorm activity and the presence of precipitation in the form of rain or drizzle during the exposure periods were categorized as dichotomous variables.

Air pollutant data

A subanalysis was conducted to assess the association between temperature change and ED visits after controlling for environmental air pollutants. Data were obtained from the Michigan Department of Environmental Quality for measures of particulate matter with an aerodynamic diameter less than 2.5 microns (PM_2.5), ozone, sulfur dioxide, and nitrogen dioxide from 3 air monitors in the Detroit area in 2000 and 2001. The air monitors were chosen based on proximity to the CHM and the availability of data. For nitrogen dioxide, ozone, and sulfur dioxide, the daily maximum level of the air pollutant was determined. For PM_2.5, only 1 measurement was taken per day; therefore, this daily value was used. The measurement area considered valid for the environmental pollutants in residential areas is a 4-kilometer radius around the air-monitoring station (13). Therefore, this subanalysis was restricted to asthma-related ED visits for children residing in zip codes in which more than 50% of the total area falls inside the 4-kilometer radius of an air-monitoring station.

Analysis

For descriptive analysis, asthma-related ED visits were restricted to events not considered a treatment failure. Treatment failure was defined as an asthma-related ED visit within 7 days of a previous ED visit. Analyses were conducted to assess the distribution of age at visit, race, sex, closest weather station, and temporal characteristics of the visit (meteorological season, month, day of the week, and time of day).

To determine whether temperature change is related to asthma-related ED visits, a case-crossover study was conducted. This design is similar to a case-control study; however, cases serve as their own controls at a different point in time. The case-crossover design is appropriate for transient outcomes that are impacted by short-term events or exposures (14, 15). In the present study, each asthma-related ED visit, including those that were considered treatment failures, was counted as a case. Controls were defined as the absence of an ED visit at selected times for the case patient and were determined using a time-stratified approach (16). For each case event, up to 4 control times were designated as the times occurring on the same day of the week and during the 3-hour block when the case occurred during the same calendar month. Control times were matched to case times on day of the week, time of day, and calendar month to control for confounding by these variables and to reduce the likelihood that a patient made substantial changes in their asthma self-management. Exposure data for control times were collected from the weather- and air-monitoring stations providing data for the patient when they were a case.

Statistical analysis

Crude analysis of the distribution of temperature change and temperature change rate during case and control times, stratified by exposure period, was conducted using the nonparametric Friedman test for correlated repeated measures. Multivariable conditional logistic regression models stratified by exposure period were created to assess the relation between absolute temperature change and temperature change rate after controlling for other weather and air pollutant measures. We conducted analyses of the exposures as continuous variables, as well as categorized into tertiles. Odds ratios, 95% confidence intervals, and 2-sided P values associated with the Wald χ² statistics for the model parameters are presented; P < 0.05 was considered significant. These analyses were completed using SAS, version 9.2 (SAS Institute, Inc., Cary, North Carolina).

RESULTS

During 2000 and 2001, there were 6,659 ED visits due to asthma at the CHM. Excluding visits resulting in an inpatient admission (1,719) and treatment failures (136), 4,804 visits were available for descriptive analysis. The majority of the sample was black (87.6%), male (62.5%), and between the ages of 3 and 10 years (75.9%). ED visits were most likely to occur in the fall (36% of all visits), with September being the most frequent month. Asthma-related ED visits occurred most frequently Sunday through Tuesday and between 9:00 am and 12:00 pm. The vast majority of children resided nearest the City Airport weather station (92.5%), which is the weather station closest to the CHM (Table 1).

Table 1.

Description of Asthma Emergency Department Visits (n = 4,804), Children's Hospital of Michigan, Detroit, Michigan, 2000–2001

Characteristic	Proportion of Visits, %
Patient Characteristics at the Time of the Visit
Age, years
3–6	43.5
7–10	32.4
11–14	18.2
15–18	6.0
Race
White	2.3
Black	87.6
Other	10.1
Sex
Male	62.5
Female	37.5
Nearest weather station
Willow Run Airport	0.3
City Airport	92.5
Detroit Metropolitan Airport	7.2
Visit-Level Characteristics
Meteorological season
Winter	22.6
Spring	25.4
Summer	15.9
Fall	36.2
Day of the week
Sunday	15.6
Monday	16.5
Tuesday	15.3
Wednesday	14.8
Thursday	12.9
Friday	12.2
Saturday	12.6
Time of visit
12 am–2:59 am	10.6
3 am–5:59 am	5.7
6 am–8:59 am	9.4
9 am–11:59 am	18.3
12 pm–2:59 pm	13.4
3 pm–5:59 pm	12.6
6 pm–8:59 pm	14.3
9 pm–11:59 pm	15.5

There were 4,871 case events (including the 136 treatment failures) available for the analytic analysis and 16,546 controls included in the case-crossover study (69 case events were excluded because of left- and right-censored climatic data). Controls were matched 3:1 for 2,938 cases and 4:1 for 1,933 cases, depending on the number of days in the calendar month that matched the case event's day of the week. Comparisons between the average absolute maximum temperature change and absolute maximum temperature change rate during exposure periods for the case versus controls are presented in the Figure 1. The distribution of short-term temperature change was significantly different across exposures preceding cases and controls for all of the exposure periods considered (P < 0.01). Although the average temperature change and temperature change rates for case and control periods did not differ more than 2 degrees overall, exposures preceding controls were on average higher than those preceding cases, regardless of the exposure period. At maximum, temperature change during the 24 hours preceding controls was an average of 0.4 degrees higher than that preceding the cases.

Figure 1.

A) Average absolute maximum temperature change and B) temperature change rate per hour of exposure during case and control times, stratified by period of exposure measurement, Detroit, Michigan, 2000–2001. The results of analyses using the Friedman test for differences between correlated exposure values across repeated measures were statistically significant (P < 0.01) for both temperature change and temperature change rate and for all exposure periods. Sample sizes follow for control and case times from left to right (4 hours, 8 hours, 12 hours, and 24 hours): −28 days (379, 380, 384, and 384), −21 days (1382, 1393, 1398, and 1408), −14 days (2431, 2469, 2485, and 2517), −7 days (3578, 3648, 3666, and 3707), case (4686, 4766, 4792, and 4842), +7 days (3643, 3708, 3741, and 3788), +14 days (2634, 2669, 2686, and 2716), +21 days (1537, 1551, 1559, 1559), and +28 days (338, 341, 341, and 341).

Multivariable analysis of ED visits and each exposure, stratified by exposure period and adjusted for maximum relative humidity, maximum wind speed, maximum barometric pressure, thunderstorm occurrence, and rain/drizzle precipitation occurrence, are presented in Table 2. Absolute maximum temperature change during a 24-hour exposure period was significantly associated with reduced risk of an asthma-related ED visit after adjustment for climatic factors (odds ratio = 0.992, 95% confidence interval: 0.984, 0.999; P = 0.036). With the same model variables, absolute maximum temperature change rate during a 24-hour exposure period was significantly associated with a reduced risk of an asthma-related ED visit (odds ratio = 0.972, 95% confidence interval: 0.952, 0.993; P = 0.009). Despite this statistically significant result (P < 0.05), the odds ratio was essentially equal to 1.0, indicating no association between temperature change and risk of an asthma-related ED visit. For the remainder of the exposure periods, neither temperature change nor temperature change rate was associated with asthma-related ED visits in adjusted models.

Table 2.

Association of Absolute Maximum Temperature Change and Temperature Change Rate With Emergency Department Visits for Asthma, by Period of Exposure Measurement, Children's Hospital of Michigan, Detroit, Michigan, 2000–2001

Exposure Period and Parameter	Total No. of Events and Controls	Crude			Adjusted for Climactic Factors
		Parameter Estimatea	OR	95% CI	Parameter Estimatea,b	OR	95% CI
4 hours	12,782
Temperature change		0.00452	1.005	0.994, 1.015	0.00362	1.004	0.990, 1.018
Temperature change rate		0.00343	1.003	0.984, 1.023	0.00472	1.005	0.978, 1.033
8 hours	13,002
Temperature change		0.00360	1.004	0.996, 1.011	0.00016	1.000	0.990, 1.010
Temperature change rate		0.01085	1.011	0.993, 1.029	0.01230	1.012	0.987, 1.038
12 hours	13,053
Temperature change		0.00165	1.002	0.995, 1.008	−0.00298	0.997	0.988, 1.006
Temperature change rate		0.00469	1.005	0.988, 1.022	−0.00033	1.000	0.976, 1.024
24 hours	13,157
Temperature change		0.00057	1.001	0.995, 1.006	−0.00827*	0.992	0.984, 0.999
Temperature change rate		−0.00557	0.994	0.979, 1.010	−0.02808**	0.972	0.952, 0.993

Abbreviations: CI, confidence interval; OR, odds ratio.

*P < 0.05 (Wald χ² test); **P < 0.01 (Wald χ² test).

^a Units were 1 degree for temperature change and 1 degree per hour for temperature change rate.

^b Adjusted for maximum relative humidity, maximum wind speed, maximum barometric pressure, thunderstorm activity, and rain/drizzle precipitation during the exposure period.

Multivariable analysis results for each exposure, stratified by exposure period and adjusted for climatic factors (described above) and daily maximum sulfur dioxide, nitrogen dioxide, and daily PM_2.5 level, are presented in Table 3. Neither the absolute maximum temperature change nor absolute maximum temperature change rate was significantly associated with asthma-related ED visits in these analyses, regardless of the exposure period. Multivariable analyses of exposures categorized in tertiles or the addition of ozone in the full model did not alter the results (data not shown).

Table 3.

Association of Absolute Maximum Temperature Change and Temperature Change Rate With Emergency Department Visits for Asthma, Adjusted for Climatic Factors and Ambient Air Pollutant Exposure Excluding Ozone and Stratified by Period of Exposure Measurement, Children's Hospital of Michigan, Detroit, Michigan, 2000–2001

Exposure Periodand Parameter	Total No. ofEvents and Controls	Parameter Estimatea,b,c	OR	95% CI
4 hours	1,403
Temperature change		0.00394	1.004	0.953, 1.058
Temperature change rate		0.04302	1.044	0.943, 1.156
8 hours	1,423
Temperature change		−0.00003	1.000	0.964, 1.037
Temperature change rate		0.05396	1.055	0.957, 1.164
12 hours	1,427
Temperature change		−0.00356	0.996	0.964, 1.030
Temperature change rate		−0.00452	0.995	0.907, 1.092
24 hours	1,446
Temperature change		−0.00607	0.994	0.966, 1.022
Temperature change rate		−0.04473	0.956	0.884, 1.035

Abbreviations: CI, confidence interval; OR, odds ratio.

^a Units were 1 degree for temperature change and 1 degree per hour for temperature change rate.

^b Adjusted for maximum relative humidity, maximum wind speed, maximum barometric pressure, thunderstorm activity, rain/drizzle precipitation, daily maximum of sulfur dioxide and nitrogen dioxide, and daily measurement of particulate matter with an aerodynamic diameter less than 2.5 microns.

^c Restricted to cases who resided in zip codes in which 50% of the zip code area fell within a 4-km radius of the air-monitoring station at the time of the asthma emergency department visit.

DISCUSSION

The present study documented a pattern of asthma-related ED visits among low-income children in Detroit, Michigan. Whether the findings are applicable to children of higher socioeconomic status is unknown. As in other studies (17–21), we found that pediatric asthma-related ED visits peak during fall and occur most frequently early in the week, with 32.1% occurring on a Sunday or Monday (17, 19). Moreover, we found that ED registrations occur most often between 9:00 am and 3:00 pm. This observation contradicts previous research findings, although data on time of ED registration is limited. Buff et al. (22), studying the circadian patterns of asthma-related ED visits during the late fall and early winter, found that the majority of pediatric asthma-related ED visits occurred between 8:00 pm and 11:59 pm. In our study, we excluded visits that resulted in a hospitalization. It is possible that ED visits occurring at night are more severe and thus more likely to result in hospitalization. Further, it is unknown how symptom onset time and delays in seeking treatment are related to ED registration time and the severity of the event. Additional research of these patterns is warranted.

Analysis from the case-crossover study indicates that both the average maximum temperature change and temperature change rate were significantly different between case and control times, irrespective of the exposure window. Exposures during control times were on average higher than during case times. We found a small, statistically significant inverse relation between pediatric ED visits and maximum temperature change or temperature change rate during a 24-hour exposure period after adjustment for climatic factors. Even if the observed effect is real, the clinical significance is unknown. With further adjustment for air pollutants, the observed associations persisted, although it was not statistically significant. With increasing exposure windows, we observed a monotonic decrease in the odds ratio association between asthma-related ED visits and both temperature change and temperature change rate in both crude and adjusted models. However, our results taken in totality suggest a negligible relation between asthma-related ED visits and short-term temperature change because most of our analyses did not demonstrate a significant association.

Contrary to our finding of limited significance, results of previous research suggested that temperature change was related to an increased number of ED visits (7, 8, 11, 23–25). However, major methodological differences between the studies exist. Our study is unique because it examined relatively short exposure periods. This was done to capture the potentially rapid onset of asthma exacerbation and presentation to ED. Asthma symptoms have been shown to manifest both slowly, from 6 to 30 hours after exposure to a trigger, and quickly, in less than 6 hours and in some cases even less than 1 hour (26–28). Previous research measured temperature change anywhere from 1 to 15 days before the ED event (7, 8, 11, 23–25). Not all previous studies of temperature change, however, demonstrated a significant association. Using a case-crossover design similar to ours, Villeneuve et al. (29) found that 6-hour temperature change, change in relative humidity, and level of visibility had no association with asthma-related ED visits. Our use of the maximum value of the exposure during the exposure period was also unique. Previous work used the average temperature change as the exposure variable (8, 25). The maximum value was used to capture how the most extreme temperature changes might impact ED visits. Finally, our study used the absolute value of the maximum temperature change during exposure periods. Previous work studied the impact of decreases in temperature (8, 23, 24) or temperature change measures that accounted for directionality (7, 11, 25).

This study offers a unique analytic approach to the study of temperature change and asthma-related ED visits. It utilized the case-crossover design, rather than the commonly used time-series Poisson regression analyses, thereby offering a distinctly robust approach for studies of transient outcomes (14, 15). It is especially useful in studies of asthma events because, by design, it controls for seasonality, underlying disease severity, the level of disease control, self-management behavior, personal asthma symptom triggers, and access to primary care. The time-stratified approach was chosen based on its properties to produce unbiased conditional logistic regression estimates, to avoid bias due to time trend in the exposure series, and its flexibility to match on specific time-varying confounders (16).

There were several limitations to this study. First, we restricted our analysis to ED data from a single hospital site. CHM accounts for the majority of pediatric asthma-related ED visits in the Detroit area. Further, the choice of hospital is likely independent of exposure to short-term temperature change. Therefore, although it is possible that using a single site may have led to some selection bias, it is unlikely to have had a serious influence. Even within the selected source of outcome data, necessary restrictions resulted in data losses for analysis. Because asthma-related ED visits resulting in inpatient admission no longer retain the ED registration time in the billing record, they were dropped from the analyses. The assignment of exposure relied on the ED registration time. Patients may spend several hours in the ED before admission, and the amount of time spent is highly variable. Therefore, the inpatient admission time, which is retained in the billing record, would not serve as a good proxy for the ED registration time. Dropping these events from analysis conferred a loss of 26.4% of the total number of ED visits at the CHM during the study period. Additionally, aggregate exposures were used as a surrogate for individual exposure data. The absence of a stronger association in our study may be due to nondifferential misclassification of exposure resulting from assigning the exposure ecologically rather than individually. Although temperature change is an environmental exposure with little variability across small geographic areas, not all persons are exposed to the same degree. An additional limitation is that the length of time between onset of an asthma episode and registration in the ED is unknown. Brenner et al. (30) found that for pediatric ED visits, the onset of the asthma episode was within 24 hours for 70% of visits that occurred from midnight to 8:00 am and 59% for visits that occurred between 8:00 am and midnight. The timing of symptom onset in relation to the ED registration time would impact the exposure period of greatest importance in predicting asthma ED events, and these missing data could have led to misclassification of exposure.

An additional limitation is the lack of information on a few important confounders. Respiratory infections are triggers of asthma episodes and have been shown to be associated with asthma-related ED visits, as well as with season (31–35). Data on respiratory infection are not reliably captured in the administrative ED billing record and were unavailable for inclusion in the study. Other potentially important confounders not included in our study include pollen grain and fungal spore concentrations (5, 36–40).

Further study is suggested by these findings. Despite the methodological differences between our study and prior research, we were surprised to find the inverse relation between temperature change and pediatric ED visits for asthma. Future studies should examine this phenomenon to assess whether the observed relation is real. Future study should also evaluate temperature change exposures measured over sequential, not overlapping, time periods, which would indicate whether short-term distal temperature change is associated asthma-related ED visits. The length of exposure periods should also be reconsidered. The 12- and 24-hour exposure periods span the diurnal temperature changes that coincide with the rising and setting of the sun; the maximum exposure measured in these time periods is less likely to capture a weather event. Additional study of specific population subgroups, such as patients less than 10 years of age at the time of their asthma-related ED visit or those who had ED visits that resulted in a hospital admission, may reveal additional understanding of the impact of short-term temperature change. Finally, it would be advantageous to replicate this study using more current data to determine whether the observed relations persist.

The present study contributes to the growing body of knowledge regarding the impact of climate on health outcomes. It is unique in that it assessed the association between short-term temperature change and asthma-related ED visits, included multiple exposure periods, and measured both absolute temperature change and temperature change rate. From our study results, we found limited evidence that greater temperature change over 24 hours is associated with decreased risk of an asthma-related ED visit. Our overall results suggest that asthma-related ED visits are not likely to be associated with temperature change during the 24 hours before the event. Nevertheless, more research is needed to discern the role of temperature dynamics in asthma symptoms and health-care utilization.

ACKNOWLEDGMENTS

Author affiliations: Department of Epidemiology, School of Public Health and Tropical Medicine, Tulane University, New Orleans, Louisiana (Elizabeth A. Wasilevich, Felicia Rabito); Department of Biostatistics and Bioinformatics, School of Public Health and Tropical Medicine, Tulane University, New Orleans, Louisiana (John Lefante); and Department of Epidemiology, Fay W. Boozman College of Public Health, University of Arkansas for Medical Sciences, Little Rock, Arkansas (Eric Johnson).

This work was presented in part at the 2010 American Thoracic Society Conference, New Orleans, Louisiana, May 14–19, 2010.

Conflict of interest: none declared.

Author notes

Abbreviations: CHM, Children's Hospital of Michigan; ED, emergency department; PM_2.5, particulate matter with an aerodynamic diameter less than 2.5 microns.