Effect of particulate matter air pollution on hospital admissions ... - Nature

Journal of Exposure Science and Environmental Epidemiology (2007) 17, 478–487 r 2007 Nature Publishing Group All rights reserved 1559-0631/07/$30.00

www.nature.com/jes

Effect of particulate matter air pollution on hospital admissions and medical visits for lung and heart disease in two southeast Idaho cities GREGORY V. ULIRSCHa, LOUISE M. BALLb, WENDY KAYEc, CARL M. SHYd, CAROLYN V. LEEc, DOUGLAS CRAWFORD-BROWNb, MICHAEL SYMONSe AND TRACEY HOLLOWAYf a

Division of Health Assessment and Consultation, Agency for Toxic Substances and Disease Registry, Public Health Service, Atlanta, GA, USA Department of Environmental Sciences and Engineering, School of Public Health, University of North Carolina, Chapel Hill, NC, USA c Division of Health Studies, Agency for Toxic Substances and Disease Registry, Public Health Service, Atlanta, GA, USA d Department of Epidemiology, School of Public Health, University of North Carolina, Chapel Hill, NC, USA e Department of Biostatistics, School of Public Health, University of North Carolina, Chapel Hill, NC, USA f Nelson Institute for Environmental Studies, University of Wisconsin, Madison, WI, USA b

Few, if any, published time series studies have evaluated the effects of particulate matter air exposures by combining hospital admissions with medical visit data for smaller populations. We investigated the relationship between daily particulate matter (o10 mm in aerometric diameter or PM10) exposures with admissions and medical visits (emergency room, urgent care, and family practice) for respiratory and cardiovascular disease in Pocatello and Chubbuck, Idaho (population about 60,000), from November 1994 through March 2000. Within generalized linear models, time, weather, influenza, and day-ofweek effects were controlled. In single-pollutant models, respiratory disease admissions and visits increased (7.1–15.4% per 50 mg/m3 PM10) for each age group analyzed, with the highest increases in two groups, children and especially the elderly. Statistical analyses suggest that the results probably did not occur by chance. Sensitivity analyses did not provide strong evidence that the respiratory disease effect estimates were sensitive to reasonable changes in the final degrees of freedom choice for time and weather effects. No strong evidence of confounding by NO2 and SO2 was found from results of multipollutant models. Ozone and carbon monoxide data were not available to include multi-pollutant models, but evidence suggests that they were not a problem. Unexpectedly, evidence of an association between PM10 with cardiovascular disease was not found, possibly due to the lifestyles of the mostly Mormon study population. Successful time series analyses can be performed on smaller populations if diverse, centralized databases are available. Hospitals that offer urgent or other primary care services may be a rich source of data for researchers. Using data that potentially represented a widerange of disease severity, the findings provide evidence that evaluating only hospital admissions or emergency room visit effects may underestimate the overall morbidity due to acute particulate matter exposures. Further work is planned to test this conclusion. Journal of Exposure Science and Environmental Epidemiology (2007) 17, 478–487; doi:10.1038/sj.jes.7500542; published online 14 February 2007

Keywords: particulate matter, pulmonary disease, child exposure/health, population-based studies.

Introduction Much of the research into the health effects of daily exposures to particulate matter has used mortality, hospital admissions, or emergency room data in moderate to large cities worldwide (usually greater than 100,000 persons). Studies on these larger populations have produced valuable insights into the health effects of exposure to particulate matter. However, unique air pollution exposure scenarios or population demographic characteristics also need exploring in smaller populations. Moreover, only a few studies have

1. Address all correspondence to: Dr GV Ulirsch, Division of Health Assessment and Consultation, Agency for Toxic Substances and Disease Registry, 1825 Century Center Blvd., Atlanta, GA 30345 USA. Tel.: þ 1 404 498 0460. Fax: þ 1 404 498 0135. E-mail: [email protected]. Received 22 August 2006; accepted 14 November 2006; published online 14 February 2007

evaluated the effects of particulate matter exposures and visits to a primary care facility. Two studies in London evaluated exposure to particulate matter and daily general practitioner visits for asthma, lower respiratory diseases or allergic rhinitis (Hajat et al., 1999, 2001). Two other studies (Choudhury et al., 1997; Ostro et al., 1999) evaluated the effects of particulate matter exposures with daily asthma medical visits in Anchorage, Alaska or to primary care clinics in Santiago, Chile, for upper and lower respiratory symptoms. An assessment of two phosphate processing plants (FMC and Simplot Corporation, Figure 1), comprising the Eastern Michaud Flats Contamination Superfund Site, near Pocatello and Chubbuck, Idaho (population B60,000) determined that PM10 exposures from these plants and from other major sources (wood burning and wind-blown dust) from 1975 to 2000 constituted a public health hazard (ATSDR, 2001). As a follow-up to this assessment and to address community concerns regarding perceived increases in respiratory (RD) and cardiovascular disease (CVD), several

Effect of PM on lung and heart disease

Ulirsch et al.

Figure 1. Location of phosphate processing plants (FMC and Simplot), monitoring stations, and other features in the study area.

time series analyses were conducted. One of two hospitals that served the community also provided primary care services (two urgent care facilities and a family practice clinic). The authors conducted a time series analysis (TS1) of available data to evaluate the relationship between PM10 (particulate matter o10 mm in aerometric diameter) exposures with the combined counts of daily admissions and visits to emergency room and urgent care facilities. Family practice data (available after February 1997) was included in a second time series (TS2). Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)

To date, few time-series studies have evaluated air pollution health effects on mortality, hospital admissions, or emergency room visits in smaller communities, possibly due to concerns about limited statistical power. Here, we examine whether such limitations may be overcome by combining databases on hospital admissions with medical visit data, increasing daily counts. This unique approach allows us to examine health effects of PM10 exposures and RD and CVD on a small population. Moreover, since these data potentially represented a wide-range of disease severity, 479


Ulirsch et al.

the study may provide some insights into the particulate matter morbidity burden captured by outpatient data.

Materials and methods Hospital and Medical Visit Data for Study Population The Internal Review Board at the Centers for Disease Control and Prevention reviewed the protocol for this study and approved a waiver from the requirement to obtain informed consent, parental permission and HIPAA patient authorization. The study population was restricted to persons who resided (by street address) in or near the cities of Pocatello or Chubbuck and were admitted for inpatient care or visited the hospital run urgent care or family practice clinics during the study period November 1994 through March 2000. Family practice visit data for March 1997 through March 2000 did not specify whether a visit was scheduled or unscheduled. For the main study variables, we summarized the daily admissions and medical visits for RD (International Classification of Diseases, 9th revision (ICD-9) codes 460–519; 786.09 (reactive airway disease); excluding 500–508 for lung diseases caused by external causes) and for CVD (ICD-9 codes 390–429). For the control study variables, we summarized certain diseases of the eye (ICD-9 codes 365– 366), gastrointestinal tract (ICD-9 codes 531–536, 537, 540– 543, 558 and 574–575), kidney and urinary tract (ICD-9 codes 590 and 599) and injury and trauma cases (ICD-9 codes E800-E848, E880-E928, and E950–989). The ICD-9 codes for the control variable were chosen because they are not likely related to air pollution. Table 1 provides the summary statistics for the main and control variables. Only primary ICD-9 codes were counted as a case for the main or control variable analyses. No routinely collected data were available locally to characterize influenza outbreaks. Primary and secondary diagnosis codes for influenza were not counted as a case but were included as a separate influenza indicator variable. Environmental and Exposure Data for Study Population Adequate PM2.5 (o2.5 mm in aerometric diameter) data to conduct a time series analysis were not available for the study period. However, on any given day, sufficient PM10 data were available from none up to four monitoring stations (Figure 1). Most PM10 data from the monitors were moderately to strongly correlated (R ¼ 0.52–0.87). The least correlated monitors were the sewage treatment plant (STP) and the two Pocatello monitors (R ¼ 0.52–0.60). Nitrogen dioxide (NO2) data were available from one monitor (Garret & Gould (G&G)). Sulfur dioxide (SO2) data were available from two monitors (G&G and STP). Neither ozone (O3) nor carbon monoxide (CO) data were available from the four 480

Table 1. Hospital admission and medical visit summary statistics for main and control study variables.

All age RD* and CVD* Total counts of admissions/visits Daily mean 10th percentile 90th percentile Minimum Maximum % RD % CVD % Admissions % Emergency room visits % Urgent care visits % Family practice visits

Time series 1

Time series 2

39,347 20 11 31 4 63 90.1 9.9 11.4 41.5 47.1 0

29,513 26 14 41 4 64 81.2 18.8 9.3 31.7 32.8 27.2

Average RD and CVD counts per day (age group) All-age RD 18.0 0–17 years RD 5.5 65+ years RD 1.6 All-age CVD 2.0 Fa 30+CVD 65+RD/CVD 2.8 All-age control variables Total counts of admissions/visits Daily mean 10th percentile 90th percentile Minimum Maximum % Trauma and injury % Other diseases (eye, gastrointestinal, kidney, and urinary tract)

58,477 29.6 21 39 10 61 74.8 25.2

Average control variable counts per day (age group) 0–17 years 8.4 65+ 2.9

21.1 6.1 2.5 4.9 4.8 5.4

36,233 32.1 24 42 12 62 71.2 28.2

8.8 3.5

*RD, respiratory diseases; CVD, cardiovascular diseases. Not analyzed for in time series 1.

a

monitors. Air pollutant data were obtained from the US Environmental Protection Agency (US EPA) (US EPA, 2002). The Chubbuck School (CS) and STP monitors likely represented exposure for Chubbuck and northern Pocatello. The G&G and Idaho State University (ISU) monitors were more representative of exposures in downtown and southern Pocatello (Figure 1). A person who resided in one part of the study area would have likely worked, shopped, or recreated in other parts; thus, on average, the four monitors likely represent the exposure to the study population. As the STP monitor was not likely a good indicator of exposure to persons living in downtown or southern Pocatello, we excluded days from the analyses where PM10 or SO2 data Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)


Ulirsch et al.

where available from only the STP monitor. Table 2 shows other details of how we determined the final PM10 and SO2 exposure metric for any given day. Summary statistics for the environmental data are presented in Table 3. Daily weather data were available from the National Weather Service operating at the Pocatello Regional Airport (NCDC, 2003). Air pollutant data were mostly weakly correlated, with a moderate correlation between PM10 and NO2 (R ¼ 0.47). We did not have complete data sets for PM10 or for the co-pollutants, but it is not uncommon to have missing PM data in time series analyses. The National Morbidity, Mortality, and Air Pollution Study (NMMAPS) database, for example, had air monitoring data available for only 1 in 6 days (Roberts, 2005). In our study, PM10 data were available for 81% and 67% of the days for TS1 and TS2, respectively, compared with 17% available data in a 1 in 6 day monitoring schedule. We chose to run the co-pollutant analyses on days that we had concurrent PM10, NO2, and SO2 data.

Statistical Analysis of Daily Admissions and Visits with PM Exposures Although several time series studies in the literature have found statistically significant positive associations with daily counts as low as about 2.0 per day (Gwynn et al., 2000; Stieb et al., 2000; Atkinson et al., 2001), power was a potential concern in this study. To increase daily counts we chose to evaluate RD, CVD and combined RD/CVD in all ages in both time series. Representing potentially susceptible populations, we also evaluated the effects of PM10 exposures on children (0–17 years; RD only) and on the elderly (65 þ years; RD, CVD, and RD/CVD). For comparison to the RD results for children and the elderly, we also evaluated effects in non-elderly adults (18–64 years). To increase the daily counts for the CVD analyses, we evaluated two other age groups (30 þ and 50 þ years) in TS2. Separate analyses were performed for the TS1 all-age group during the cool months (October to March) and in the warm months (April to September) to examine if there were any

Table 2. Exposure metric development approach for PM10 and SO2. Available monitors

Description of approach

PM10 exposure metric 1 Use data from available monitor, except if from STP* since data from this monitor is not likely to represent the exposures to the entire study population. Days when only STP data were available were excluded from the analyses 2 Average two data points 3 Data from nearest monitor to one not available (i.e., STP to CS* or ISU* to G&G*) represented exposures to that portion of the study area and was counted as 50% of the total exposure contribution for that day. This value was then averaged with the mean value from the other two monitors 4 Averaged all four data points SO2 exposure metric 1 Data excluded because only available for STP monitor and not likely to represent SO2 exposures to entire community 2 Averaged two data points *STP, sewage treatment plant monitor; CS, Chubbuck school monitor; ISU, Idaho State University monitor; G&G, Garret and Gould monitor.

Table 3. Environmental data summary statistics. TS1* PM10 EMa TS2* PM10 EM TS1 NO2 TS2 NO2 TS1 SO2 EMb TS2 EM SO2 Minimum Maximum Average (mg/m3) (mg/m3) (ppb) (ppb) (ppb) (ppb) temperature temperature relative humidity (1F) (1F) (%) % Available datac Mean 10th percentile 90th percentile Minimum Maximum

80.7 24.2 10.5 40.7 3.0 183.0

66.5 23.2 10.0 37.4 3.0 183.0

60.2 15.2 6.6 24.6 0.04 63.0

34.7 15.8 6.9 25.9 0.04 63.0

63.6 3.0 0.2 7.9 oDLd 30.3

36.2 3.0 0.3 7.7 oDL 30.3

99.6 33.4 18.0 52.0 15.0 69.0

99.9 58.0 35.0 89.0 4.0 102.0

99.8 58.8 35.7 81.3 20.6 98.9

*TS1, time series 1; TS2, time series 2. PM10 exposure metric (EM)Fsee Table 2 for description of approach used to calculate the exposure metric. SO2 exposure metricFsee Table 2 for description of approach used to calculate the exposure metric. c Number of days data available divided by total number of days in study (1,978 for time series 1 (TS1) and 1,124 for time series 2 (TS2)). Temperature percentages are based on 1,978 days. d Less than detection limit. a

b

Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)

481


Ulirsch et al.

seasonal differences in the effect estimates due to PM10 exposures. We used log-linear generalized linear models (GLM) with natural splines to evaluate the relationship between daily admissions and visits with PM10 exposures. This approach has been recommended as an alternative to generalized additive models using other smoothing approaches (Dominici et al., 2002a), but no approach can be strongly preferred over another (HEI, 2003). The Poisson distribution has been widely used in time series analysis. One assumption of the model is that, at each level of the covariates, the number of cases has variance equal to its mean. If this is not true, then the Poisson model is either over- or under-dispersed (Insightful Corporation, 2001). Preliminary analyses indicated the models were over-dispersed; therefore, we choose to use quasi-likelihood to adjust the standard error to account for the over-dispersion. We varied the degrees of freedom (df), using natural splines, to account for time-varying effects and weather (minimum and maximum temperature and average relative humidity). To ensure that admission and visit peaks were not due to influenza outbreaks, we added a flu indicator variable. During the study period, influenza counts were Z10 on 4 days and 5–9 on 32 days. For the 4 days potentially indicating a severe outbreak, several days before and after these events were included. An additional indicator variable was added for day-of-week effects. The goodness-of-fit of these base models was evaluated by using the Akaike Information Criteria (AIC) (US EPA, 2004). Similar to other studies (Lipsett et al., 1997; Prescott et al., 1998), we also visually inspected autocorrelation function plots to determine the presence of autocorrelation. Once the best-fitting model was determined, single day lags from 0 to 4 were evaluated. To account for consecutive days of high PM10 levels that have occurred in this community during inversion events, a 0–4 day moving average lag was evaluated. The Health Effects Institute (HEI) recommends an exploration of the sensitivity of time series models to a wide range of alternative degrees of smoothing and to specifications of weather variables (Dominici et al., 2002b). Because the ‘‘right’’ extent of smoothing is not known, sensitivity analyses are critical (US EPA, 2004). We conducted sensitivity and co-pollutant analyses at one or more lags, based on which lags indicated the highest effect estimate and lowest two-sided P-values from the base modeling. The sensitivity analyses varied the final df choice for time and weather by dividing or multiplying by a factor of 72–3. The final step was to include the co-pollutants (NO2 and SO2) individually or together with PM10. The lag with the most robust results from the sensitivity and co-pollutant analyses was determined. In the re-analyses of selected particulate matter air pollution time-series studies, HEI (2003) arbitrarily determined that changes in effect estimates Z40% 482

were considered ‘‘substantial’’. Those between 20% and 40% to show ‘‘some’’ change and those r20% where considered to be ‘‘little to no’’ change. Although we did not use the same approach as the HEI, for both the sensitivity and co-pollutant analyses, we did not consider changes in the MPC of r33% to be important. All statistical analyses were performed with S-Plus version 6.1 software.

Results Respiratory Disease Particulate matter model results (single-pollutant). For both time series, a lag of 0 days (hereafter referred to as lag 0) for RD admissions or visits produced the most robust results (Table 4). Effect estimates were calculated as the mean percentage of change (MPC), which is the percent change in the mean number of daily admissions and visits per specified increase in PM10 levels. We found a positive association for all age groups analyzed. The MPC per 50 mg/m3 increase in PM10 ranged from 7.1% (TS2 all-age group) to 15.4% (TS2 65 þ age group). P-values below or near the 0.05 level were found for all results except for the TS1 65 þ age group analyses (Po0.12), which was reduced to 0.07 in the TS2 65 þ age group analyses. Several within and between age-group specific effect estimate comparisons are notable (see Table 4 for specific age-group MPC values). The MPC increased for the 0–17 versus the 18–64 age groups by 26% for TS1 and 55% for TS2. For the 65 þ versus the 18–64 age groups, the increases were 66% for TS1 and 117% for TS2. When individually comparing the 65 þ and 0–17 age groups across TS1 and TS2, the TS2 MPC values were 21–28% higher than for TS1. The MPC was positive in both the cool and warm season analyses, with larger effects seen in the warm months. For all 72 lags evaluated for the RD-alone analyses, 79% were positive. Given the number of analyses performed, a multiple-comparison analysis was conducted. The P-value counts in each significance category (r0.01, 40.01 to r0.05, and 40.05 to r0.10) exceeded by about four-fold what might be expected by chance alone. Co-pollutant Model Results. Multi-pollutant model results for SO2, NO2, or both with PM10 are shown in Table 5. None of the PM10 MPC values decreased by 433%, except for the TS1 0–17 age group. The inclusion of NO2 in the 0–17 age group model resulted in a 48% MPC reduction, whereas, including SO2 and NO2 into the 0–17 age group model resulted in a MPC reduction of 31%. In two-pollutant models, 55% of the MPCs for NO2 and SO2 were negative. Positive MPCs for NO2 and SO2 in two-pollutant models had P-values ranging from 0.14 to 0.82. In three-pollutant models, 64% of the MPCs for NO2 and SO2 were negative. Positive MPC values for NO2 and SO2 in three-pollutant Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)


Ulirsch et al.

Table 4. Summary of mean percentage of change results for particulate matter single-pollutant modelsa at lag 0 for respiratory diseases, combined respiratory/cardiovascular diseases, and control variable admissions and visits. Health outcome

MPCb

95% CI

P-value

MPC c

MPC d

RD* or RD/CVD* Time series 1 All-age RD (all year) All-age RD/CVD 18–64 years RD 0–17 years RD 65+ years RD 65+ years RD/CVD 0–17/65+ years RD All-age RD (cool season) All-age RD (warm season)

4.0 3.7 3.4 4.3 5.6 2.9 5.5 4.3 6.7

1.4, 1.3, 0.2, 0.1, 1.4, 2.9, 1.4, 1.3, 0.8,

6.7 6.3 6.7 8.9 13.1 8.7 9.6 7.5 14.8

0.03 o0.01 0.04 0.06 0.12 0.30 o0.01 o0.01 0.08

8.4 7.9 7.2 9.1 12.0 6.1 11.6 9.1 14.3

9.2 Fe 9.8 9.0 11.1 F 10.2 11.6 13.2

Time series 2 All-age RD All-age RD/CVD 18–64 years RD 0–17 years RD 65+ years RD 65+ years RD/CVD Combined 0-17/65+ years

3.3 2.7 3.3 5.0 6.9 2.4 3.8

0.3, 0.1, 0.4, 0.1, 0.4, 3.1, 0.3,

6.3 5.6 7.0 10.1 14.7 8.2 8.0

0.03 0.06 0.08 0.05 0.07 0.47 0.07

7.1 5.8 7.1 11.0 15.4 5.1 8.3

7.8 F 7.2 9.4 11.3 F 7.3

Control variables Time series 1 All-age (all year) 18–64 years 0–7 years 65+ years Combined 0-17/65+ years

1.1 0.3 3.5 3.3 1.6

3.0, 2.6, 7.1, 2.1, 4.7,

0.8 2.1 0.3 9.0 1.5

0.24 0.82 0.07 0.24 0.30

2.4 0.6 7.3 7.2 3.5

F F F F F

Time series 2 All-age (all year) 18–64 years 0–7 years 65+years Combined 0-17/65+years

1.6 0.5 4.6 1.5 3.9

4.1, 3.4, 9.3, 5.0, 7.9,

0.9 2.5 0.3 8.4 0.2

0.21 0.75 0.07 0.66 0.06

3.5 1.0 9.9 3.2 8.3

F F F F F

*RD, respiratory diseases; RD/CVC, combined respiratory and cardiovascular diseases. Models that did not include the co-pollutants sulfur dioxide and nitrogen dioxide. Mean percentage of change (MPC) in admissions/visit per mean PM10 increase during time series 1 (24.3 mg/m3) or time series 2 (23.2 mg/m3). c Mean percentage of change in admissions/visit per 50 mg/m3 increase in PM10. d Mean percentage of change in admissions/visit per 50 mg/m3 increase in PM10 for sensitivity analysis at df ¼ 8/year for time effects, per Slaughter et al. (2005). e Not analyzed. a

b

models had P-values ranging from 0.16 to 0.93. For TS1 multi-pollutant models, 72% of the PM10 MPCs were the same or increased. For TS2 multi-pollutant models, all MPCs increased, but were inflated as compared with TS1 results. This may be due to the number of missing days for either PM10 or co-pollutant data for TS2 versus TS1.

Sensitivity Analysis Results. The MPC values for most of the age groups analyzed were insensitive to a two- to threefold increase or decrease in the final df choice to model time effects (results not shown). For the TS2 0–17 age group and the combined 0–17/65 þ age group analyses, the MPC was sensitive to decreases in the df used to model time. Whereas, Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)

the MPCs for the same age groups in TS1 were insensitive. The MPCs were generally insensitive to changes to the final weather df choice.

Cardiovascular Disease and Combined Respiratory/ Cardiovascular Disease For the cardiovascular disease-alone analyses, 92% of the MPCs were negative. The positive lags had P-values ranging from 0.54 to 0.88. As no single lag emerged as the most robust, results for all lags are reported (Table 6). As with the RD-alone analyses, the most robust lag for the RD/CVD analyses was zero. Although all the MPCs for the RD/CVD analyses were positive (Table 4), the magnitude of 483


Ulirsch et al.

Table 5. Comparison of effect estimates for respiratory diseases from single- and multi-pollutant models at lag 0. Age group

Table 6. Summary of mean percentage of change results for particulate matter-alone models for cardiovascular disease admissions and visits at all lags.

MPCa for MPC for MPC for MPC for PM10 PM10 PM10 with PM10 alone with SO2 with NO2 SO2 and NO2

Time series 1 All-age (all year) 18–64 years 0–17 years 65+ years 0–17/65+ years All-age (cool season)b

8.4 7.2 9.1 12.0 11.6 9.1

10.8 8.0 10.8 8.7 14.2 11.9

10.5 9.3 4.6 12.4 9.5 11.1

11.3 9.0 6.2 12.0 10.3 11.0

Time series 2 All-age 18–64 years 0–17 years 65+ years 0–17/65+ years

7.1 7.1 11.0 15.4 8.3

17.5 9.1 32.7 31.3 25.3

16.3 17.3 18.7 32.7 16.8

21.7 23.1 23.6 45.2 23.3

a Mean percentage of change (MPC) reported for all values in table as the % increase in admissions/visits per 50 mg/m3 increase in PM10. b Multi-pollutant models were not analyzed for the warm season because of the lack of co-pollutant data.

MPCa Time series 1 All-age Lag 0 Lag 1 Lag 2 Lag 3 Lag 4 Lag 0–4 days Time series 2 All-age Lag 0 Lag 1 Lag 2 Lag 3 Lag 4 Lag 0–4 days

the MPC was reduced 6–66% as compared with the RDalone analyses. The largest MPC decreases were for the 65 þ age group analyses (50–66%).

Control Variable Results Excluding the 65 þ age group analyses, 69% of the MPCs analyzed were negative. The remaining positive MPCs were relatively weak (0.10–4.2% per 50 g/m3 PM10) and had P-values ranging from 0.22 to 0.77. For the 65 þ age group, all of the MPCs were positive (3.2–17.3%) with P-values ranging from 0.06 to 0.66. The 65 þ age group results indicate that the 0–4 day moving average lag was the most robust. The MPC values for the 0–4 day moving average lag ranged from 13.6% to 17.3%, with P-values of 0.06–0.08. For the lag 0 control variable results, 80% of the MPCs were negative. The two positive MPC values ranged from 3.2% to 7.2%, with P-values of 0.24 and 0.66. The overall pattern of a mostly negative association for the control variable analyses was quite distinct from the mostly positive association found for respiratory diseases, especially for lag 0, the most robust lag seen in the RD and RD/CVD analyses (Table 4). These results could be due to chance. No positive control variable MPC result achieved a 0.05 significance level. A multiple comparison analysis, at three different levels of significance (as before), indicated that the number of P-values in each of these levels was at or slightly above what may be expected by chance alone given the 60 multiple comparisons. Moreover, four of the significant results were negative. 484

0.02 1.9 3.1 0.5 1.7 0.5

5.0 0.7 -6.2 6.6 5.5 7.5

P-value

MPCb

6.3 8.4 3.4 6.9 0.9 7.6

0.99 0.54 0.35 0.88 0.58 0.90

0.05 4.0 6.2 1.0 3.6 1.1

0.6 5.2 0.2 0.7 0.4 0.5

0.08 0.82 0.04 0.03 0.07 0.04

10.6 1.5 12.9 13.7 11.4 15.5

CI

5.9, 4.1, 9.1, 5.6, 4.3, 8.0,

10.4, 6.3, 11.8, 12.1, 11.0, 14.1,

30+years Lag 0 Lag 1 Lag 2 Lag 3 Lag 4 Lag 0–4 days

5.1 0.04 5.1 5.7 4.7 6.4

10.6, 0.6 6.0,1.0 10.9, 0.1 11.4, 0.3 10.3, 1.3 13.1, 0.9

0.08 0.99 0.10 0.06 0.13 0.08

10.7 0.1 10.8 12.0 9.8 13.2

50+ Lag Lag Lag Lag Lag Lag

3.6 1.74 7.0 5.6 6.3 8.2

9.6, 7.9, 13.2, 11.7, 12.4, 15.3,

0.26 0.60 0.04 0.09 0.06 0.04

7.7 3.7 14.5 11.7 13.2 17.2

years 0 1 2 3 4 0–4 days

2.7 4.8 -0.3 0.9 0.2 -0.4

a

Mean percentage of change (MPC) in admissions/visits per mean PM10 increase during time series 1 (24.3 mg/m3) or time series 2 (23.2mg/m3). b Mean percentage of change (MPC) for both time series reported as the % increase in admissions/visits per 50 mg/m3 increase in PM10.

Discussion Effect estimates for short-term exposures with respiratoryrelated hospital admissions or emergency room visits for US and Canadian cities fall most consistently in the range of a 2–12% increase per 50 mg/m3 increase in PM10, with some effect estimates for respiratory medical visits ranging up to about 30% (US EPA, 2004). In this study, the RD alone results for both time series (7.1–15.4% increase per 50 mg/m3 PM10 increase) are at the upper end or above the range for studies of admissions and emergency room visits, but below the upper range for other medical visit studies. The RD effect seen in this study is not likely to be driven solely by the risk for a hospital admission or ER visit given that urgent care visits comprised about 50% of the total admissions and visits Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)


for TS1 and that the urgent care and family practice visits comprised about 60% of the total for TS2. A direct comparison of the effect estimate results from this study to other studies is not possible because no studies were identified that combined hospital admissions with emergency room, urgent care and family practice data. Moreover, many of the studies of larger populations were able to evaluate individual RD such as asthma, COPD, etc. Individual RD analyses were not performed for this study because of the potential reduction in power due to small counts per day. The results provide evidence of a larger risk of a hospitalization or medical visit in children, and especially the elderly, who are potentially more susceptible to the effects of PM10 exposures. Comparing TS2 to TS1 results, the only MPC value that increased with the addition of family practice visits were for the 0–17 and 65 þ age groups. This may be due to the increased likelihood for a child or a senior citizen to obtain a same-day doctor’s appointment since we have no information on which family practice visits were scheduled or unscheduled. We chose to use two diagnostic tools (AIC and autocorrelation function plots inspection) to optimize the model while testing the sensitivity of this approach to a wide-range of df choices for time and weather. Slaughter et al. (2005) reported that using AIC as a rigid optimization criterion tended to over-fit the time trends, which induced autocorrelation in the residuals. The modeling approach in this study appears to have minimized the residual deviance and autocorrelation since the base models did not show any strong evidence of autocorrelation or pattern in the time series data. Except for the TS2 0–17 age group and combined 0–17/ 65 þ age group analyses, none of the age group analyses were sensitive to df changes to model time and weather effects. As the pure modeling approach used produced relatively high df choices for time effects (10–28 df/year), an additional sensitivity analysis was performed using only 8 df/year for all the RD GLM models, as was used in Spokane, Washington, by Slaughter et al. (2005). Using this approach, all of the original MPCs, including the TS2 0–17 and combined 0–17/65 þ age groups, were insensitive to df changes for time effects (Table 4). These results do not provide strong evidence that the effect estimates, using the modeling approach in this study, were sensitive to reasonable changes to the final time and weather df choice. Similar to other studies (US EPA, 2004), the multipollutant analyses of PM10 with SO2 and NO2 produced a ‘‘mixed-bag’’ of results regarding the likelihood of confounding by these co-pollutants. Overall, the co-pollutant analyses show that the PM10 effect is not likely confounded by SO2 and NO2 in the model and that SO2 and NO2 have no effect. However, the results for the TS1 0–17 age group indicate the potential for confounding by NO2. Given the potential for variance inflation in these models from multi-colinearity Journal of Exposure Science and Environmental Epidemiology (2007) 17(5)

Ulirsch et al.

between PM10 and NO2 (HEI, 2003), we cannot assume that the respiratory effect associated with PM in the 0–17 age group was due to NO2. Although we had no direct measurements of O3 or CO, we do have some limited, indirect evidence that these co-pollutants were not confounders in our study. Most significantly, the State of Idaho does not consider O3 or CO to be a problem in the area (Tom Edwards, Idaho Department of Environmental Quality, ‘‘personal communication,’’ 2005). Not having O3 or CO data to include in the models was a major limitation of this study. However, one of the major strengths of the study was the high percentage of days where we were able to calculate the PM10 exposure metric (about 81% and 67% for TS1 and TS2, respectively). Moreover, for many other studies in the literature, only one or two monitoring locations provided exposure information for much larger populations and metropolitan areas. For a relatively small community, both in terms of population size and area, the availability of daily PM10 data from up to four monitoring stations (two in Pocatello and two in the Chubbuck area) was a strength in terms of enhancing the exposure information for this study. The cardiovascular disease-alone analyses provided little evidence of a positive association with PM10. For RD/CVD, in all age groups analyzed, the magnitude of the MPC was reduced as compared with the respective RD-alone analysis. The risk reduction was particularly pronounced in the potentially susceptible 65 þ age group. The positive results for the combined RD/CVD analyses were likely driven by the RD risk. The findings of no association between PM10 and cardiovascular disease admissions and visits was not expected given that excess risk estimates from other studies of hospital admissions or emergency room visits in US cities most consistently fall in the range of about 3–9% per 50 mg/m3 increase in PM10, especially in the elderly (US EPA, 2004). For studies that evaluated total CVD admissions or emergency room visits in all ages, as we did in our study, several researchers (Burnett et al., 1997, 1999; Atkinson et al., 1999; Stieb et al., 2000) found statistically significant excess risk, in single-pollutant models, ranging from 3.2% to 32.5% for a 50 mg/m3 increase in PM10. In another age-group evaluated in our study, Linn et al. (2000) found that persons 30 years and older living in Los Angeles, California, had a 3.3% increased risk of a CVD admission per 50 mg/m3 increase in PM10. One reason for our CVD findings may relate to the lifestyles of the study population, which is estimated to be about 52% Mormon (Robert Chambers, City of Pocatello, ‘‘personal communication,’’ 2006). Members of the Mormon Church advocate abstinence from the use of tobacco, alcohol, coffee, and tea. One study found fewer excess deaths from ischemic heart disease (IHD) among Mormons and a lower death rate from hypertensive disease among Mormon men (Lyon et al., 1978). 485


Ulirsch et al.

The data sets used for this study contained mostly nonIHD cases (73–83%). Therefore, as discussed by Pope et al. (2004), there may have been little to no interactive effect between particulate matter exposure and smoking that would increase the risk of non-IHD outcomes. The finding of a possible interactive effect by Pope et al. (2004) was for longterm exposures to particulate matter, not short-term exposures. However, evidence from some studies suggests that short-term exposures to particulate matter may increase the risk in persons and animals with pre-existing cardiovascular disease (Liao et al., 1999; Gold et al., 2000; Peters et al., 2000; HEI, 2000). Pope et al. (1992) evaluated mortality in Utah Valley in relation to PM10 exposures from a steel mill. Respiratory diseases accounted for 10% of the deaths while cardiovascular disease accounted for 46% of the deaths during the study period. That study apparently had a larger pool of CVD-related health outcomes (as compared with RD), in a population of 90% Mormons, than did this study. Followup studies in humans and animals indicated that exposure to the filter extracts (rich in transition metals) when the steel mill was operating produced significantly higher markers of an inflammatory response than did exposure to extracts when the steel mill was closed (Dye et al., 2001; Ghio and Devlin, 2001). Other studies suggest that the sulfate and transition metal components may play a role in the onset of particulate matter-related CVD outcomes (Burnett et al., 1995; Levy et al., 2001; Sullivan et al., 2005). The degree to which the sulfate and transition metal composition in the study area affected the CVD results is uncertain. No long-term average sulfate concentrations are available. Limited data indicate that sulfate may be relatively low (about 6%) on ‘‘typical’’ days, but on ‘‘high’’ days, sulfate increased to about 30% (IDEQ, 2003). Higher sulfate contributions to PM10 have been reported during severe inversions in the community (IDEQ, 2000). The levels and types of transition metals in the study area PM10 are not well understood. The primary reason to conduct a control variable analysis was as an internal validity check of the methods used. The results of the control variable analyses provided little evidence of an association between the control variables and PM10, especially for lag 0, the most robust for the RD analyses. In general, the methods used for the control variable analyses produced the expected result. Therefore, the findings from the control variable analyses provide support that the methods employed can produce potentially unbiased results. The findings from this study show that successful time series analyses can be performed on smaller populations if diverse, centralized databases can be identified. The more severe illnesses represented by RD admissions and emergency room visits were unlikely to be the sole driver of the effect observed. The results of this study provide some support to the suggestion by the US EPA (2004) that looking at only 486

hospital admissions or emergency hospital visit effects may underestimate the overall morbidity due to acute ambient particulate matter exposures. Additional analyses are planned to further test this conclusion. Many more hospitals throughout the US are offering urgent and other primary care to the communities they serve. If researchers can identify hospitals or other major medical centers that offer a range of services, investigations of other smaller communities with unique exposure or population characteristics could be more robust. Moreover, these centralized databases may offer researchers an opportunity to better evaluate the overall morbidity burden of exposure to particulate matter by comparing the risk of a severe illness, represented by hospital admissions and emergency rooms visits, to a less severe illness, potentially represented by urgent care and doctor’s visits.

Acknowledgements The authors appreciate the assistance provided by the Pocatello Regional Medical Center and Intermountain Health Care in providing the hospital admission and medical visit data vital to conducting this study. In addition, we appreciate all of the assistance provided by the Idaho Department of Environmental Quality and the Idaho Department of Public Health.

References Atkinson R.W., Anderson H.R., Sunyer J., Ayres J., Baccini M., Vonk J., Boumghar A., Forastiere F., Forsberg B., Touloumi G., Schwartz J., and Katsouyanni K. Acute effects of particulate air pollution on respiratory admissions. Am J Respir Crit Care Med 2001: 164: 1860–1866. Atkinson R.W., Bremner S.A., Anderson H.R., Strachan D.P., Bland J.M., and Ponce de Leon A. Short-term associations between emergency hospital admissions for respiratory and cardiovascular disease and outdoor air pollution in London. Arch Env Health 1999: 54: 398–411. ATSDR. Air contamination health consultation for Eastern Michaud Flats Contamination Superfund site. Agency for Toxic Substances and Disease Registry, US Department of Health and Human Services, Atlanta, GA, 2001. Burnett R.T., Cakmak S., Brook J.R., and Krewski D. The role of particulate size and chemistry in the association between summertime ambient air pollution and hospitalization for cardiorespiratory diseases. Env Health Perspect 1997: 105(6): 614–619. Burnett R.T., Dales R., Krewski D., Vincent R., Dann T., and Brook J.R. Associations between ambient particulate sulfate and admissions to Ontario hospitals for cardiac and respiratory diseases. Am J Epidemiol 1995: 142: 15–22. Burnett R.T., Smith-Doiron M., Stieb D., Cakmak S., and Brook J.R. Effects of particulate and gaseous air pollution on cardiorespiratory hospitalizations. Arch Env Health 1999: 54: 130–139. Choudhury A.H., Gordian M.E., and Morris S.S. Associations between respiratory illness and PM10 air pollution. Arch Env Health 1997: 52(2): 113–117. Dominici F., McDermott A., Daniels M., Zeger S.L., and Samet J.M. A report to the Health Effects Institute: reanalyses of the NMMAPS database 2002b, Baltimore, MD: Johns Hopkins University, Available: (http://www.biostat. jhsph.edu/Bfdominic/NMMAPS/HEIReanalysis.pdf). Dominici F., McDermott A., Zeger S.L., and Samet J.M. On the use of generalized additive models in time-series of air pollution and health. Am J Epidemiol 2002a: 156(3): 193–203.



Dye J.A., Lehmann J.R., McGee J.K., Winsett D.W., Ledbetter A.D, Everitt J.L., Ghio A.J., and Costa D.L. Acute pulmonary toxicity of particulate matter filter extracts in rats: coherence with epidemiological studies in Utah residents. Env Health Perspect 2001: 109(Suppl 3): 395–403. Ghio A.J., and Devlin R.B. Inflammatory lung injury after bronchial instillation of air pollution particles. Am J Crit Care Med 2001: 64(4): 704–708. Gold D.R., Litonjua A., and Schwartz J. Ambient air pollution and heart rate variability. Circulation 2000: 10(11): 1267–1273. Gwynn R.C., Burnett R.T., and Thurston G.D. A time series analysis of acidic particulate matter and daily mortality and morbidity in the Buffalo, New York, region. Env Health Perspect 2000: 108(2): 125–133. Hajat S., Haines A., Bremer S.A., Anderson H.R., and Emberlin J. Association of air pollution with daily consultations with general practitioners for allergic rhinitis in London, United Kingdom. Am J Epidemiol 2001: 153(7): 704–714. Hajat S., Haines A., Goubet S.A., Atkinson R.W., and Anderson H.R. Association of air pollution with daily GP consultations for asthma and other lower respiratory conditions in London. Thorax 1999: 54: 597–605. HEI. Mechanisms of morbidity and mortality from exposure to ambient air particles. Health Effects Institute, Boston, MA, 2000. HEI. Special report: Revised analyses of time series studies of air pollution and health. Health Effects Institute, Boston, MA, 2003. IDEQ. CMB analysis for PM10 scenario for 1999, Pocatello, Idaho. Idaho Division of Environmental Quality, Boise, ID, 2000. IDEQ. Pocatello chemical mass balance analysis. Idaho Department of Environmental Quality, Boise, ID, 2003. Insightful Corporation. S-Plus 6.0 for Windows user guides. Seattle, WA, 2001. Levy D., Sheppard L., Checkoway H., Kaufman J., Lumley T., Koenig J., and Siscovick D. A case-crossover analysis of particulate matter air pollution an out-of-hospital primary cardiac arrest. Epidemiology 2001: 12(2): 193–199. Liao D., Creason J., Shy C., Williams R., Watts R., and Zweidinger R. Daily variation of particulate air pollution and poor cardiac autonomic control in the elderly. Env Health Perspect 1999: 107: 521–525. Linn W.S., Szlachcic Y., Gong H., Kinney P., and Berhane K.T. Air pollution and daily hospital admissions in metropolitan Los Angeles. Env Health Perspect 2000: 108: 427–434. Lipsett M., Hurley S., and Ostro B. Air pollution and emergency room visits for asthma in Santa Clara County California. Env Health Perspect 1997: 105(2): 216–222.


Ulirsch et al.

Lyon L.J., Wetzler H.P., Gardner J.W., Klauber M.R., and Williams R.R. Cardiovascular mortality in Mormons and non-Mormons in Utah, 1969– 1971. Am J Epidemiol 1978: 108(5): 357–365. NCDC (National Climatic Data Center). Climatic data from National Weather Service office in Pocatello, Idaho for 1994–2000, Available: http:// www.ncdc.noaa.gov/oa/ncdc.html[accessed 20 June 20 2003]. Ostro B.D., Eskeland G.S., Sanchez J.M., and Feyzioglu T. Air pollution and health effects: a study of medical visits among children in Santiago, Chile. Env Health Perspect 1999: 107: 69–73. Peters A., Liu E., Verrier R.L., Schwartz J., Gold D.R., Mittleman M., Baliff J., Oh J.A., Allen Monahan K., and Dockery D.W. Air pollution and incidence of cardiac arrhythmia. Epidemiology 2000: 11: 11–17. Pope C.A., Schwartz J., and Ransom M.R. Daily mortality and PM10 pollution in Utah Valley. Arch Env Health 1992: 7(3): 211–217. Pope C.A., Burnett R.T., Thurston G.D., Thun M.J., Calle E.E., Krewski D., and Godleski J.J. Cardiovascular mortality and long-term exposure to particulate air pollution: epidemiological evidence of general pathophysiological pathways of disease. Circulation 2004: 109(1): 71–77. Prescott G.J., Cohen G.R., Elton R.A., Fowkes F.G., and Agius R.M. Urban air pollution and cardiopulmonary ill health: a 14.5 year time-series study. Occup Env Med 1998: 55: 697–704. Roberts S. Using moving total mortality counts to obtain improved estimates for the effect of air pollution on mortality. Env Health Perspect 2005: 113(9): 1148–1152. Slaughter J.C., Kim E., Sheppard L., Sullivan J.H., Larson T.V., and Claiborn C. Association between particulate matter and emergency room visits, hospital admissions and mortality in Spokane, Washington. J Exposure Anal Env Epidemiol 2005: 15: 153–159. Stieb D.M., Beveridge R.C., Brook J.R., Smith-Doiron M., Burnett R.T., Dales R.E., Beaulieu S., and Mamedov A. Air pollution, aeroallergens and cardiorespiratory emergency department visits in Saint John, Canada. J Exposure Anal Env Epidemiol 2000: 10: 461–477. Sullivan J., Sheppard L., Schreuder A., Ishikawa N., Siscovick D., and Kaufman J. Relation between short-term fine-particulate matter exposure and onset of myocardial infarction. Epidemiology 2005: 16(1): 41–48. US EPA (US Environmental Protection Agency) Aerometric Information Retrieval System (AIRS). Available: http://www.epa.gov/airs/airs.html (accessed 10 June 2002). US EPA. Air quality criteria for particulate matter. US Environmental Protection Agency, National Center for Environmental Assessment, Research Triangle Park, NC, 2004.

487