Cardiovascular

802 views

Published on

Published in: Health & Medicine, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
802
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
14
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Cardiovascular

  1. 1. Fine Particulate Air Pollution and Hospital Admission for Cardiovascular and Respiratory Diseases Francesca Dominici; Roger D. Peng; Michelle L. Bell; et al. Online article and related content current as of February 16, 2009. JAMA. 2006;295(10):1127-1134 (doi:10.1001/jama.295.10.1127) http://jama.ama-assn.org/cgi/content/full/295/10/1127 Contact me if this article is corrected. Correction This article has been cited 126 times. Citations Contact me when this article is cited. Topic collections Occupational and Environmental Medicine; Pulmonary Diseases; Pulmonary Diseases, Other Contact me when new articles are published in these topic areas. Related Letters Hospital Admissions and Fine Particulate Air Pollution George D. Thurston. JAMA. 2006;296(16):1966. Paolo F. Ricci et al. JAMA. 2006;296(16):1966. In Reply: Francesca Dominici et al. JAMA. 2006;296(16):1966. Subscribe Email Alerts http://jama.com/subscribe http://jamaarchives.com/alerts Permissions Reprints/E-prints permissions@ama-assn.org reprints@ama-assn.org http://pubs.ama-assn.org/misc/permissions.dtl Downloaded from www.jama.com by guest on February 16, 2009
  2. 2. ORIGINAL CONTRIBUTION Fine Particulate Air Pollution and Hospital Admission for Cardiovascular and Respiratory Diseases Francesca Dominici, PhD Context Evidence on the health risks associated with short-term exposure to fine Roger D. Peng, PhD particles (particulate matter 2.5 µm in aerodynamic diameter [PM2.5]) is limited. Re- sults from the new national monitoring network for PM2.5 make possible systematic Michelle L. Bell, PhD research on health risks at national and regional scales. Luu Pham, MS Objectives To estimate risks of cardiovascular and respiratory hospital admissions Aidan McDermott, PhD associated with short-term exposure to PM2.5 for Medicare enrollees and to explore heterogeneity of the variation of risks across regions. Scott L. Zeger, PhD Design, Setting, and Participants A national database comprising daily time- Jonathan M. Samet, MD series data daily for 1999 through 2002 on hospital admission rates (constructed from the Medicare National Claims History Files) for cardiovascular and respiratory out- N UMEROUS EPIDEMIOLOGICAL comes and injuries, ambient PM2.5 levels, and temperature and dew-point tempera- studies have shown associa- ture for 204 US urban counties (population 200 000) with 11.5 million Medicare en- tions of acute and chronic ex- rollees (aged 65 years) living an average of 5.9 miles from a PM2.5 monitor. posures to airborne particles Main Outcome Measures Daily counts of county-wide hospital admissions for pri- with risk for adverse effects on morbid- mary diagnosis of cerebrovascular, peripheral, and ischemic heart diseases, heart rhythm, ity and mortality.1,2 The recent evi- heart failure, chronic obstructive pulmonary disease, and respiratory infection, and in- dence on adverse effects of particulate air juries as a control outcome. pollution on public health has led to Results There was a short-term increase in hospital admission rates associated with more stringent standards for levels of par- PM2.5 for all of the health outcomes except injuries. The largest association was for ticulate matter in outdoor air in the heart failure, which had a 1.28% (95% confidence interval, 0.78%-1.78%) increase United States and in other countries. In in risk per 10-µg/m3 increase in same-day PM2.5. Cardiovascular risks tended to be 1997, the US National Ambient Air Qual- higher in counties located in the Eastern region of the United States, which included the Northeast, the Southeast, the Midwest, and the South. ity Standard for airborne particulate mat- ter was revised, maintaining the previ- Conclusion Short-term exposure to PM2.5 increases the risk for hospital admission ous indicator of particulate matter of less for cardiovascular and respiratory diseases. than or equal to 10 µm in aerodynamic wwww.jama.com JAMA. 2006;295:1127-1134 diameter (PM10) and creating a new in- logical studies on this size range of par- The national data on PM2.5 concen- dicator for fine particulate matter of less ticulate matter had been reported at that trations were used to assess associa- than or equal to 2.5 µm in aerodynamic diameter (PM2.5).3 Following the imple- time. The EPA heavily weighted the few tions of short-term exposure to PM2.5 studies with available PM2.5 data when with risk for hospitalization regionally mentation of the PM2.5 National Ambi- it considered the level that should be set and by city among Medicare partici- ent Air Quality Standard, a nationwide for the standard.4 The EPA also consid- pants. We followed the model of the Na- monitoring system of this pollutant was ered the dosimetry of particles in the tional Morbidity, Mortality and Air Pol- implemented. Data on PM2.5 are now lung. Particles in the size range of PM2.5 available for many parts of the United Author Affiliations: Departments of Biostatistics (Drs have a much greater probability of reach- States starting from 1999 through the Dominici, Peng, McDermott, and Zeger and Mr Pham) ing the small airways and the alveoli of present. and Epidemiology (Dr Samet), Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Md; the lung than do larger particles. The Although the US Environmental Pro- and School of Forestry and Environmental Studies, Yale availability of the new monitoring net- tection Agency (EPA) added a PM2.5 stan- University, New Haven, Conn (Dr Bell). work for PM2.5 allows epidemiological dard in 1997 based on available evi- Corresponding Author: Francesca Dominici, PhD, Bloomberg School of Public Health, Johns Hopkins analyses at the national level on the dence that these small particles were University, 615 N Wolfe St, Baltimore, MD 21205 health effects of fine particles. particularly damaging, few epidemio- (fdominic@jhsph.edu). ©2006 American Medical Association. All rights reserved. (Reprinted) JAMA, March 8, 2006—Vol 295, No. 10 1127 Downloaded from www.jama.com by guest on February 16, 2009
  3. 3. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION Table 1. Percentage Change in Hospitalization Rate per 10-µg/m3 Increase in PM2.5 on Average Across 204 Counties National Average Relative Rate, PE (95% PI)† Daily Hospitalization Rates for 1999-2002, All Medicare Reason for Lag Day Median (IQR) per Enrollees All Counties, Hospital Admission No.* 100 000 Individuals (Aged 65 y) Aged 65-74 y Aged 75 y PE (95% PI)‡ Injury 0 4.1 (3.7 to 4.5) −0.41 (−1.00 to 0.18) 0.22 (−1.01 to 1.45) −0.46 (−1.16 to 0.24) 1.47 (0.33 to 2.48) Cerebrovascular disease 0 5.4 (4.8 to 6.0) 0.81 (0.30 to 1.32) 0.91 (0.01 to 1.82) 0.80 (0.21 to 1.38) 1.24 (0.35 to 2.05) Peripheral vascular disease 0 1.7 (1.5 to 1.9) 0.86 (−0.06 to 1.79) 1.21 (−0.26 to 2.67) 0.86 (−0.39 to 2.11) 2.11 (0.79 to 3.40) Ischemic heart disease 2 8.1 (7.1 to 9.4) 0.44 (0.02 to 0.86) 0.37 (−0.22 to 0.96) 0.52 (−0.01 to 1.04) 1.15 (0.40 to 1.77) Heart rhythm 0 3.8 (3.3 to 4.2) 0.57 (−0.01 to 1.15) 0.46 (−0.63 to 1.54) 0.72 (0.02 to 1.42) 1.05 (0.26 to 1.91) Heart failure 0 5.5 (4.7 to 6.6) 1.28 (0.78 to 1.78) 1.21 (0.35 to 2.07) 1.36 (0.78 to 1.94) 1.09 (0.42 to 1.78) COPD 0 2.6 (2.1 to 3.2) 0.91 (0.18 to 1.64) 0.42 (−0.64 to 1.48) 1.47 (0.54 to 2.40) 1.61 (0.56 to 2.66) Respiratory tract infection 2 5.4 (4.7 to 6.2) 0.92 (0.41 to 1.43) 0.93 (0.04 to 1.82) 0.92 (0.32 to 1.53) 1.39 (0.60 to 2.09) Abbreviations: COPD, chronic obstructive pulmonary disease; IQR, interquartile range; PE, point estimate; PI, posterior interval; PM2.5, particulate matter of less than or equal to 2.5 µm in aerodynamic diameter. *Results are reported for the lag at which the greatest effect of PM2.5 was estimated. †Percentage change in hospital admission rates per 10-µg/m3 increase in PM2.5 concentration. ‡The SD of the true relative rates among counties (heterogeneity). Earth-Info CD database.12 To protect lution Study, which used PM10 data for Eight outcomes were considered time-series analyses.5-8 The Medicare co- based on the ICD-9 codes for 5 cardio- against consequences of outliers, we hort covers nearly all members of an el- vascular outcomes (heart failure [428], used a 10% trimmed mean to average derly population considered to be vul- heart rhythm disturbances [426-427], across monitors after correcting for nerable to air pollution; the size of this cerebrovascular events [430-438], is- yearly averages for each monitor. population allows for assessments of spe- chemic heart disease [410-414, 429], pe- County names and location, air pol- cific cardiac and respiratory diagnostic ripheral vascular disease [440-448]), 2 lution data, weather data, county- categories that have been associated with respiratory outcomes (chronic obstruc- specific estimates of health risk, and particulate air pollution. tive pulmonary disease [COPD; 490- software developed to construct county- 492], respiratory tract infections [464- specific time-series data are available METHODS 466, 480-487]), and hospitalizations online (http://www.biostat.jhsph.edu caused by injuries and other external /MCAPS). Billing claims of Medicare This analysis is based on daily counts causes (800-849). The county-wide daily enrollees are not publicly available. of hospital admissions for 1999-2002 hospitalization rates for each outcome Calculations were implemented using obtained from billing claims of Medi- R statistical software version 2.2.0.13 for 1999-2002 appear in TABLE 1. care enrollees. Because the Medicare The study population includes 11.5 We applied Bayesian 2-stage hierar- data analyzed for this study did not in- chical models14-16 to estimate county- million Medicare enrollees residing an volve individual identifiers, consent was not specifically obtained. This study was average of 5.9 miles from a PM2.5 moni- specific, region-specific, and national reviewed and exempted by the institu- tor. The analysis was restricted to the average associations between day-to- tional review board at Johns Hopkins 204 US counties with populations larger day variation of PM2.5 (at lags 0, 1, and Bloomberg School of Public Health. than 200 000. Of these 204 counties, 90 2 days) and day-to-day variation in the Each billing claim contains the date of had daily PM2.5 data across the study pe- county-level hospital admission rates, service, treatment, disease (Interna- riod and the remaining counties had accounting for weather, seasonality, and tional Classification of Diseases, Ninth PM2.5 data collected once every 3 days long-term trends. A lag of 0 days cor- Revision9 [ICD-9] codes), age, sex, self- for at least 1 full year. The locations of responds to the association between reported race, and place of residence the 204 counties appear in FIGURE 1. PM2.5 concentration on a given day and (ZIP code and county). The daily counts The counties were clustered into 7 geo- the risk of hospitalization on the same of each health event within each county graphic regions by applying the K- day. We also applied distributed lag models17-20 to the 90 counties with daily were obtained by summing the num- means clustering algorithm to longi- tude and latitude for the counties.10,11 ber of hospital admissions for each of PM2.5 data available to estimate the rela- the diseases considered a primary di- The PM2.5 and ozone data were ob- tive rate (RR) of hospitalization asso- agnosis. To calculate hospitalization tained from the EPA’s Aerometric In- ciated with cumulative exposure over rates, we constructed a time series of formation Retrieval Service (now re- the current day and the 2 previous days. the numbers of individuals at risk in ferred to as the Air Quality System Significance is assessed by the poste- each county for each day (defined as the database). Temperature and dew-point rior probability that the RR is larger than number of individuals enrolled in Medi- temperature data were gathered from the zero. Values greater than .95 are con- care on a given day). National Climatic Data Center on the sidered significant. ©2006 American Medical Association. All rights reserved. 1128 JAMA, March 8, 2006—Vol 295, No. 10 (Reprinted) Downloaded from www.jama.com by guest on February 16, 2009
  4. 4. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION In the first stage, single lag and dis- Figure 1. US Counties With Populations Larger Than 200 000 Included in Analysis tributed lag overdispersed Poisson re- gression models21,22 were used for esti- mating county-specific RRs of hospital Northwest Midwest admissions associated with ambient lev- Northeast els of PM2.5. These county-specific mod- els include as explanatory variables: (1) the logarithm of the daily number of in- dividuals at risk; (2) indicator variables for the day of the week to allow for dif- ferent baseline hospital admission rates for each day; (3) smooth functions of cal- endar time (natural cubic splines) with 8 degrees of freedom per year to adjust for seasonality and for other time- West Southeast varying influences on admissions (eg, in- fluenza epidemics and longer-term trends Central due to changes in medical practice pat- South terns); and (4) smooth functions of tem- perature (6 degrees of freedom) and dew- Population, In Millions point temperature (3 degrees of freedom) on the same day and of the 3 previous 0.20 0.25 0.30 0.50 0.66 0.84 1.30 9.50 0.40 days’ temperature and dew-point tem- perature to control for the potential con- founding effect of weather. For the smooth functions of calendar 10-µg/m3, and N is the number of hos- same 2-stage hierarchical model de- time, we chose 8 degrees of freedom per scribed above but separately within each pital admissions across the 204 coun- year so that little information at the time of the 7 regions. ties for 2002. scales of longer than 2 months would be To explore effect modification of air The sensitivity of key findings was ex- retained in estimating the risks. For tem- pollution risks by location-specific char- amined with respect to the lag of expo- perature, we chose 6 degrees of free- acteristics, we fitted a weighted linear sure; degrees of freedom in the smooth dom so that the model has sufficient flex- regression model with the dependent functions of time; and degrees of free- ibility to take account of potential variable as the location-specific RR es- dom in the smooth functions of tem- nonlinearity in the relationship of tem- timate and the independent variable as perature and dew-point temperature. perature with hospitalization.23 the location-specific characteristic. The This modeling approach was devel- observations were weighted inversely RESULTS oped for the National Morbidity, Mor- to the statistical variance of the location- More than 2 years of PM2.5 data were tality and Air Pollution Study analy- specific estimate. ses22,24 and applied to national databases available for most of the 204 counties. The county and regional averages of The average of the county mean an- for estimating short-term effects of PM10 PM2.5 concentration, ozone concentra- and ozone on mortality.5,12 Statistical nual values for 1999-2002 was 13.4 tion, and temperature for 2000 through µg/m3 (interquartile range [IQR], 11.3- properties of this modeling approach 2002 were calculated as potential modi- 15.2 µg/m3). There was substantial ho- and alternative modeling specifica- fiers. A regional average was calculated mogeneity of fine particulate matter con- tions for confounding adjustment are by using all of the county-specific con- reported elsewhere.7,25 centrations across geographic areas. The centrations within the region. median of pairwise correlations among In the second stage, to produce a na- Finally, the annual reduction in hos- PM2.5 monitors within the same county tional average estimate of the short- pital admissions (H) attributable to a for 2000 was 0.91 (IQR, 0.81-0.95). term association between PM2.5 and hos- 10-µg/m3 reduction in the daily PM2.5 The point estimates and 95% poste- pital admissions, we used Bayesian level for the 204 counties by cause- hierarchical models14-16,26 to combine rior intervals (PIs) for the percentage specific admissions were calculated. increase in daily admission rates per RRs across counties accounting for H is defined as 10-µg/m3 increase in PM2.5 concentra- within-county statistical error and for H=(exp( x)– 1) N tion (national average RRs) for single lags between-county variability of the “true” of 0, 1, and 2 days and the distributed where is the national RR estimate RRs (also called heterogeneity). To pro- for a 1-µg/m3 increase in PM2.5, x is lag models for lags 0 through 2 for all dis- duce regional estimates, we used the ©2006 American Medical Association. All rights reserved. (Reprinted) JAMA, March 8, 2006—Vol 295, No. 10 1129 Downloaded from www.jama.com by guest on February 16, 2009
  5. 5. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION ease outcomes (total) appear in FIGURE 2. the single lag estimates, the wider 95% the exposure indicator, positive asso- The single lag model estimates the effect PIs for the distributed lag estimates ciations also were not found. The main of exposure on 1 day only, lagged by 0, reflect the restriction of the analysis to results were robust to the number of de- 1, or 2 days, while the total estimate 90 of the 204 counties with daily data. grees of freedom used to adjust for tem- from the distributed lag model sum- The results for the single lag models poral confounding, to the adjustment marizes the effect of 3 days of expo- were also stratified by age group at the for weather, and to adjust for the prior sure (lag 0, 1, and 2 days). We found lag with the greatest effect (Table 1). distributions used for the analysis. evidence of positive associations The national average RR estimates were The point estimates and 95% PIs of the between day-to-day variation in PM2.5 larger for the oldest group for some out- heterogeneity parameter, defined as the concentration and hospital admis- comes including ischemic heart dis- between-county SD of the “true” county- sions for all outcomes, except injuries, ease, heart rhythm disturbances, heart specific rates in relation to their mean, for at least 1 exposure lag. The largest failure, and COPD. appear in Table 1. For example, the es- effect was found at lag 0 for all of the Several analyses were conducted as timate of the heterogeneity parameter for cardiovascular outcomes except ische- internal checks. Analyses for lag −1 COPD is 1.61. This value indicates that mic heart disease, for which the larg- were run to predict today’s outcome by with a national average RR of 0.91% per 10-µg/m3 increase in PM2.5, 95% of the est effect was at lag 2. For respiratory using the next day’s pollution and for outcomes, the largest effects occurred hospitalizations caused by injuries and “true” county-specific RRs are within the at lags 0 and 1 for COPD and at lag 2 other external causes. Positive associa- interval of 0.91 to 1.96 1.61=−2.24% for respiratory tract infections. Distrib- tions were not found for injuries or for and 0.91 1.96 1.61=4.06%. To de- uted lag estimates were statistically sig- other external causes, which was ex- termine the strength of evidence sup- nificant for heart failure. Compared with pected. When lag −1 PM2.5 was used as porting the null hypothesis of no het- erogeneity, we calculated the posterior probability that the heterogeneity pa- Figure 2. Percentage Change in Hospitalization Rate by Cause per 10-µg/m3 Increase in rameter is smaller than .05 (the Bayesian PM2.5 on Average Across 204 US Counties analogue of a P value) and this was found Injury to be close to 0 for all outcomes. Lag 0 Lag 1 To determine whether there was sig- Lag 2 Total nificant variation of risks across the Cardiovascular Outcomes 7 geographic regions, the RR for each Cerebrovascular Disease Lag 0 outcome was estimated separately Lag 1 Lag 2 within the regions, which excluded Ho- Total nolulu, Hawaii, and Anchorage, Alaska. Peripheral Vascular Disease Lag 0 The point estimates and 95% PIs of the Lag 1 Lag 2 regional RRs for each outcome at the Total lag with the greatest estimated RR Ischemic Heart Disease Lag 0 appear in FIGURE 3 and Table 1. For the Lag 1 Lag 2 2 groups of outcomes (cardiovascular Total and respiratory), the estimated RRs have Heart Rhythm Lag 0 distinct regional patterns. For cardio- Lag 1 Lag 2 vascular diseases, all estimates in the Total Midwestern, Northeastern, and South- Heart Failure Lag 0 ern regions were positive, while esti- Lag 1 Lag 2 mates in the other regions were close Total to 0. Compared with cardiovascular dis- Respiratory Outcomes Chronic Obstructive Pulmonary Disease eases, there was greater consistency Lag 0 Lag 1 between the regions for respiratory Lag 2 diseases. However, there were larger Total Respiratory Tract Infection effects in the Central, Southeastern, Lag 0 Lag 1 Southern, and Western regions than in Lag 2 the other regions. Total Regional differences were investi- –2 –1 0 1 2 % Change in Hospital Admissions per 10-µg/m3 Increase in PM2.5 gated by dividing the United States into an Eastern region (Northeast, South- Point estimates and 95% posterior intervals of the percentage change in admission rates per 10 µg/m3 (na- tional average relative rates) for single lag (0,1, and 2 days) and distributed lag models for 0 to 2 days (total) east, Midwest, and South) and a for all outcomes. PM2.5 indicates particulate matter of less than or equal to 2.5 µm in aerodynamic diameter. Western region (West, Central, and ©2006 American Medical Association. All rights reserved. 1130 JAMA, March 8, 2006—Vol 295, No. 10 (Reprinted) Downloaded from www.jama.com by guest on February 16, 2009
  6. 6. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION Figure 3. Percentage Change in Hospitalization Rate by Region and Cause per 10-µg/m3 Increase in PM2.5 Within Each Region Northeast (60 Counties) US East Region Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection Southeast (35 Counties) Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection Midwest (48 Counties) Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection South (25 Counties) Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection West (16 Counties) Injury US West Region Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection Central (9 Counties) Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection Injury Northwest (9 Counties) Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection All Regions (202 Counties)∗ Injury Cardiovascular Outcomes Cerebrovascular Disease Peripheral Vascular Disease Ischemic Heart Disease Heart Rhythm Heart Failure Respiratory Outcomes COPD Respiratory Tract Infection –6 –4 –2 0 2 4 6 % Change in Hospital Admissions per 10-µg/m3 Increase in PM2.5 Point estimates and 95% posterior intervals of the percentage change in admission rates per 10 µg/m3 (regional relative rates). PM2.5 indicates particulate matter of less than or equal to 2.5 µm in aerodynamic diameter; COPD, chronic obstructive pulmonary disease. *Honolulu, Hawaii, and Anchorage, Alaska, were excluded. ©2006 American Medical Association. All rights reserved. (Reprinted) JAMA, March 8, 2006—Vol 295, No. 10 1131 Downloaded from www.jama.com by guest on February 16, 2009
  7. 7. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION Northwest). The average effect esti- concentrations, temperature, and izations for heart failure by 3156 for the mates and 95% PIs of the RRs for each ozone. Both county and regional aver- 204 urban counties in 2002. outcome and for the lags with the great- age temperature positively modified the COMMENT est estimated national average effects association between PM2.5 and hospi- The Medicare National Claims History appear in FIGURE 4. There were 168 tal admission rates for the 2 respira- Files were used in this study to esti- counties included in the Eastern re- tory outcomes. For example, compar- mate the short-term effects of PM2.5 on gion and 34 counties included in the ing 2 regions that differ by 1°C, there cause-specific hospitalization rates. Data Western region. Using analysis of vari- would be an estimated 18 additional obtained from national databases on ance, the differences in risk of hospi- hospital admissions per 10 000 indi- health were combined with data on air talization between the 2 regions were viduals for COPD and 9 additional hos- pollution and weather.5,27 This is a rep- statistically significant for outcomes ex- pital admissions per 10 000 individu- licable approach that can be applied pe- cept for heart failure and COPD. All RR als for respiratory tract infections per 10-µg/m 3 increase in PM 2.5 in the riodically for air pollutants or other en- estimates for cardiovascular outcomes vironmental factors as a component of were positive in the US Eastern region warmer region. We did not find evi- a national health surveillance system to but not in the US Western region. The dence of the effect modification by av- track adverse health effects. This ap- RR estimates for respiratory tract in- erage concentrations of either PM2.5 or proach also has the strength of analyz- fections were larger in the Western re- ozone. ing the national data uniformly, avoid- gion than in the Eastern region. The yearly hospital admissions attrib- utable to a 10-µg/m3 reduction in the daily ing the potential for publication bias that Effect modification of short-term ef- occurs when data from only 1 or sev- fects of PM2.5 on hospital admission PM2.5 also were calculated (TABLE 2). For example, a 10-µg/m3 reduction in PM2.5 eral counties are analyzed and positive rates was investigated by using both findings are selectively reported.27 county and regional averages of PM2.5 would reduce the number of hospital- In interpreting the findings, consid- eration needs to be given to the inher- Figure 4. Percentage Change in Hospitalization Rate by Cause per 10-µg/m3 Increase in ent limitations of the data analyzed and PM2.5 for the US Eastern and Western Regions for all Outcomes to the possibility that even the complex Outcome statistical models used are not adequate West Injury East to eliminate all bias. Medicare data are Cardiovascular Outcomes collected for administrative purposes and Cerebrovascular Disease diagnoses are known to be subject to Peripheral Vascular Disease some degree of misclassification28-30 and Ischemic Heart Disease to vary geographically.31,32 The result- Heart Rhythm ing misclassification and geographic vari- Heart Failure ability would introduce a bias in daily Respiratory Outcomes time-series analyses only if patterns of di- COPD agnosis and coding were associated with Respiratory Tract Infection level of PM2.5. We used only primary di- –3 –2 –1 0 1 2 3 agnosis, an approach that should re- % Change in Hospital Admissions per 10-µg/m3 Increase in PM2.5 duce misclassification of outcomes. To investigate whether geographic differ- Point estimates and 95% posterior intervals of the percentage change in admission rates per 10 µg/m3. PM2.5 indicates particulate matter of less than or equal to 2.5 µm in aerodynamic diameter; COPD, chronic obstruc- ences in diagnosis rates could modify the tive pulmonary disease. risks, a second-stage analysis was per- formed using county-specific hospital ad- Table 2. Annual Reduction in Admissions Attributable to a 10-µg/m3 Reduction in the Daily mission rates (number of admissions per PM2.5 Level for the 204 Counties in 2002 100 000 individuals) as an independent Cause-Specific Annual No. of Annual Reduction in variable and county-specific RR esti- Hospital Admissions Admissions Admissions (95% PI)* Cerebrovascular disease 226 641 1836 (680 to 2992) mates as a dependent variable. This Peripheral vascular disease 70 061 602 (−42 to 1254) analysis did not find such evidence of Ischemic heart disease 346 082 1523 (69 to 2976) effect modification by underlying diag- Heart rhythm 169 627 967 (−17 to 1951) nosis rates. While we relied on moni- Heart failure 246 598 3156 (1923 to 4389) tors cited for regulatory purposes, the av- COPD 108 812 990 (196 to 1785) erage distance from the centroid of a ZIP Respiratory tract infection 226 620 2085 (929 to 3241) code to the monitor was only 5.9 miles Abbreviations: COPD, chronic obstructive pulmonary disease; PI, posterior interval; PM2.5, particulate matter of less and PM2.5 levels tend to be uniform across than or equal to 2.5 µm in aerodynamic diameter. *Per 10-µg/m3 reduction in PM2.5. such distances. ©2006 American Medical Association. All rights reserved. 1132 JAMA, March 8, 2006—Vol 295, No. 10 (Reprinted) Downloaded from www.jama.com by guest on February 16, 2009
  8. 8. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION ment of atherosclerosis41; parallel hu- The modeling approach used in this fied so that control strategies can be tar- man findings also were found.42 study enabled extensive exploration of geted efficiently. Because the source mix the sensitivity of the findings. At the first Although many time-series studies for PM2.5 varies across locations, we ex- stage of the hierarchical model, we speci- have used PM10 as an exposure indica- plored spatial variation of the effect of fied the same number of degrees of free- tor, only a few studies have specifically PM2.5 on risk for hospitalization. Strong dom in the smooth functions of time and assessed associations of PM2.5 with hos- evidence for spatial heterogeneity in the temperature used to control for con- pitalization or other morbidity mea- effect of PM2.5 on risk for hospitaliza- sures.43 Lippmann et al44 and Ito et al45 founding for all the locations. This ap- tion was found. The pattern and degree proach does not necessarily lead to a used Medicare admission data for De- of heterogeneity tended to vary by out- similar degree of control for confound- troit, Mich, for 1992-1994, along with come measure. Because the magnitude ing across counties, but it does give simi- size-fractionated particle concentra- of the effects contrasted greatly in the lar flexibility to the smooth functions, tion data from a nearby monitoring sta- comparisons between the 7 regions, allowing their shape to vary across coun- tion in Windsor, Ontario. As reported counties were grouped into an Eastern by Ito et al,45 updated analyses of these ties. An alternative is to allow a differ- region and a Western region. There are ent number of degrees of freedom across data showed positive associations of known differences in the composition of counties, an approach used in multisite PM2.5 for hospitalization for pneumo- particles at this geographic scale, includ- time-series studies in Europe.33-36 Re- nia, COPD, ischemic heart disease, and ing a greater sulfate component in the cently we have compared these 2 mod- heart failure. In comparison with the East and a greater nitrate component in the West. 2 There are also well- eling strategies and found that national present study, the reported risk esti- estimates of PM10 risks were robust to the mates were several-fold higher. Mool- characterized differences in the mix of choice of method.19 We also have ex- gavkar46,47 used data for Los Angeles sources across these broad areas that may plored the sensitivity of the estimated RRs County, California, for 1987-1995 and be relevant, including power genera- to different degrees of adjustment for found that PM2.5 was significantly asso- tion and the smokestack industry in the weather and seasonality and found the ciated with risk for hospital admission East and a larger contribution from trans- results to be robust. Statistical chal- for cardiovascular disease in persons portation sources in parts of the West. aged 65 years or older. Sheppard et al48,49 lenges inherent to the adjustment for With clear and continuing indica- temporal confounding have been ex- reported a positive association of PM2.5 tion that inhaled particles affect pub- plored elsewhere.19,25,37 with risk for hospital admission for lic health adversely, the emphasis of re- Overall, we found evidence of an as- asthma in Seattle, Wash, for 1987- search should shift toward the difficult sociation between recently measured 2004, but elderly persons were ex- issue of identifying those characteris- cluded. Finally, Burnett et al50 assessed PM2.5 concentrations and daily hospi- tics of particles that determine their tox- icity.1 The EPA’s Speciation Trends Net- talizations on a national scale. Our find- risk for hospitalization for cardiores- ings complement substantial evidence piratory diseases in relation to particu- work, which is now providing extensive on particulate matter and hospitaliza- late air pollution over 3 summers in data on characteristics of PM2.5 at se- tion for respiratory or cardiovascular Toronto, Ontario. Positive associations lected sites, offers a needed resource for this research.52 causes using exposure measures other were found in univariate models that Under the Clean Air Act,53 the EPA than PM2.5 and the more limited evi- were attenuated with consideration of dence using PM2.5 specifically. While gaseous pollutants in bivariate models. is required to set a particulate matter Na- mechanisms underlying the adverse ef- There is much more literature on PM10 tional Ambient Air Quality Standard that fects of particulate matter on the respi- and risk for hospitalization, which gen- protects public health with an “ad- erally shows positive associations.2,51 In ratory and cardiac systems remain a fo- equate margin of safety.” Our findings cus of research, the leading hypotheses most urban locations across the United indicate an ongoing threat to the health emphasize inflammatory responses in States, PM2.5 accounts for at least half of of the elderly population from air- the lung and release of cytokines with the PM10 mass, and a scaling factor of borne particles and provide a rationale local and systemic consequences.1,38-40 In 0.55 has been used to convert PM10 con- for setting a PM2.5 National Ambient Air the lung, particulate matter may pro- centrations to PM2.5. With this assump- Quality Standard that is as protective of mote inflammation and thereby exac- tion, our quantitative findings for PM2.5 their health as possible. Our national ap- erbate underlying lung disease and re- are quite similar to those for both PM10 proach offers a method for continuing duce the efficacy of lung-defense and for PM2.5 as recently summarized by to search for the characteristics of par- the EPA.43 The comparability of the PM10 ticles that determine their toxicity.53 mechanisms. Cardiovascular effects may reflect neurogenic and inflammatory and PM2.5 estimates suggests that the Author Contributions: Drs Dominici, Peng, and processes.40 Experimental studies of ath- effect of PM10 on hospital admissions McDermott and Mr Pham had full access to all of the erosclerosis using genetically suscep- largely reflects its PM2.5 component. data in the study and take responsibility for the integ- rity of the data and the accuracy of the data analysis. tible mice also suggest that particulate The sources of particles contributing Study concept and design: Dominici, Peng, Zeger, Samet. matter may accelerate the develop- to the observed risks need to be identi- Acquisition of data: Dominici, McDermott. ©2006 American Medical Association. All rights reserved. (Reprinted) JAMA, March 8, 2006—Vol 295, No. 10 1133 Downloaded from www.jama.com by guest on February 16, 2009
  9. 9. HOSPITAL ADMISSION RATES AND EXPOSURE TO FINE PARTICULATE POLLUTION 33. Samoli E, Analitis A, Touloumi G, et al. Estimating mortality in 100 US cities. Am J Epidemiol. 2005;161: Analysis and interpretation of data: Dominici, Peng, the exposure-response relationships between particu- 585-594. Bell, Pham, McDermott, Zeger. late matter and mortality within the APHEA multicity 9. International Classification of Diseases, Ninth Re- Drafting of the manuscript: Dominici, Peng, Pham, project. Environ Health Perspect. 2005;113:88-95. vision (ICD-9). Geneva, Switzerland: World Health Or- Samet. 34. Touloumi G, Samoli E, Quenel P, et al. Short-term ganization; 1977. Critical revision of the manuscript for important in- effects of air pollution on total and cardiovascular 10. MacQueen JB. Some Methods for Classification and tellectual content: Dominici, Peng, Bell, McDermott, mortality. Epidemiology. 2005;16:49-57. Analysis of Multivariate Observations. Berkeley: Uni- Zeger, Samet. 35. Le Tertre A, Medina S, Samoli E, et al. Short-term versity of California Press; 1967. Statistical analysis: Dominici, Peng, Bell, Pham, effects of particulate air pollution on cardiovascular dis- 11. Hartigan JA. Clustering Algorithms. New York, NY: McDermott, Zeger. eases in eight European cities. J Epidemiol Community Wiley; 1975. Obtained funding: Dominici, Bell, Samet. Health. 2002;56:773-779. 12. Bell ML, McDermott A, Zeger SL, Samet JM, Do- Administrative, technical, or material support: Dominici. 36. Touloumi G, Katsouyanni K, Zmirou D, et al. Short- minici F. Ozone and short-term mortality in 95 US ur- Study supervision: Dominici. term effects of ambient oxidant exposure on mortality: ban communities, 1987-2000. JAMA. 2004;292:2372- Financial Disclosures: None reported. a combined analysis within the APHEA project. Am J 2378. Funding/Support: Funding for Drs Dominici, Epidemiol. 1997;146:177-185. 13. R: A Language and Environment for Statistical McDermott, Zeger, and Samet was provided by the US 37. Dominici F, McDermott A, Hastie TJ. Improved semi- Computing. Vienna, Austria: R Foundation for Statisti- Environmental Protection Agency (RD-83241701). parametric time series models of air pollution and cal Computing; 2005. Funding for Drs Dominici, Peng, and Zeger and Mr Pham mortality. J Am Stat Assoc. 2004;99:938-948. 14. Lindley DV, Smith AFM. Bayes estimates for the lin- was also provided by the National Institute for Envi- 38. Pope CA III, Hansen ML, Long RW, et al. Ambi- ear model. J R Stat Soc [Ser B]. 1972;34:1-41. ronmental Health Sciences (ES012054-03) and by the ent particulate air pollution, heart rate variability, and 15. Du Mouchel WH, Harris JE. Bayes methods for com- National Institute of Environmental Health Sciences’ blood markers of inflammation in a panel of elderly bining the results of cancer studies in humans and other Center in Urban Environmental Health (P30 ES 03819). subjects. Environ Health Perspect. 2004;112:339-345. species. J Am Stat Assoc. 1983;78:293-315. Funding for Dr Bell was provided by the Health Effects 39. Pope CA III, Burnett RT, Thurston GD, et al. Car- 16. Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian Institute, an organization jointly funded by the Envi- diovascular mortality and long-term exposure to par- Data Analysis. 2nd ed. New York, NY: Chapman & Hall; ronmental Protection Agency and automotive manu- ticulate air pollution. Circulation. 2004;109:71-77. 2003. facturers through the Walter A. Rosenblith New Inves- 40. Brook RD, Franklin B, Cascio W, et al. Air pollution 17. Almon S. The distributed lag between capital ap- tigator Award (4720-RFA04-2/04-16). and cardiovascular disease. Circulation. 2004;109:2655- propriations and expenditures. Econometrica. 1965;33: Role of the Sponsor: The funding agencies/sponsors 2671. 178-196. listed above had no involvement in the design and con- 41. Sun Q, Wang A, Jin X, et al. Long-term air pollu- 18. Schwartz J. The distributed lag between air pollu- duct of the study; in the collection, management, analy- tion exposure and acceleration of atherosclerosis and tion and daily deaths. Epidemiology. 2000;11:320-326. sis, or interpretation of the data; or in the prepara- vascular inflammation in an animal model. JAMA. 2005; 19. Welty LJ, Zeger SL. Are the acute effects of par- tion, review, or approval of the manuscript. 294:3003-3010. ticulate matter on mortality in the National Morbidity, Disclaimer: The research described in this article was 42. Kunzli N, Jerrett M, Mack WJ, et al. Ambient air pol- Mortality, and Air Pollution Study the result of inad- funded wholly or in part by the US Environmental Pro- lution and atherosclerosis in Los Angeles. Environ Health equate control for weather and season? Am J Epidemiol. tection Agency through a grant to Johns Hopkins Uni- Perspect. 2005;113:201-206. 2005;162:80-88. versity but it has not been subjected to the Environ- 43. Clean Air Scientific Advisory Committee. Review of 20. Zanobetti A, Wand M, Schwartz J. Generalized ad- mental Protection Agency’s required peer and policy ditive distributed lag models: quantifying mortality the National Ambient Air Quality Standards for Par- review and therefore does not necessarily reflect the displacement. Biostatistics. 2000;1:279-292. ticulate Matter: Policy Assessment of Scientific and Tech- views of the Environmental Protection Agency and no nical Information. Research Triangle Park, NC: US En- 21. McCullagh P, Nelder JA. Generalized Linear Models. official endorsement should be inferred. The con- vironmental Protection Agency; 2005. 2nd ed. New York, NY: Chapman & Hall; 1989. 44. Lippmann M, Ito K, Nadas A, Burnett RT. Associa- 22. Kelsall JE, Samet JM, Zeger SL, Xu J. Air pollution tents of this article do not necessarily reflect the views tion of particulate matter components with daily mor- and mortality in Philadelphia, 1974-1988. Am J and policies of the Health Effects Institute. tality and morbidity in urban populations. Res Rep Health Epidemiol. 1997;146:750-762. Acknowledgment: We thank Charlotte Gerczak, MLA, Eff Inst. 2000;95:5-72. 23. Curriero FC, Heiner KS, Samet JM, et al. Tempera- for her editorial input and Keita Ebisu, MS, for his aid 45. Ito K, De Leon SF, Lippmann M. Associations be- ture and mortality in eleven cities of the eastern United in collecting the particulate matter data. Neither Ms tween ozone and daily mortality: analysis and States. Am J Epidemiol. 2002;155:80-87. Gerczak nor Mr Ebisu received any financial compen- meta-analysis. Epidemiology. 2005;16:446-457. 24. Dominici F, Samet J, Zeger SL. Combining evi- sation for their work. 46. Moolgavkar SH. Air pollution and hospital admis- dence on air pollution and daily mortality from the larg- sions for chronic obstructive pulmonary disease in three est 20 U.S. cities: a hierarchical modeling strategy. J R metropolitan areas in the United States. Inhal Toxicol. Stat Soc [Ser A]. 2000;163:263-302. REFERENCES 2000;12(suppl 4):75-90. 25. Peng RD, Dominici F, Louis TA. Model choice in time 47. Moolgavkar SH. Air pollution and daily mortality 1. National Research Council; Committee on Re- series studies of air pollution and mortality. J R Stat Soc in three U.S. counties. Environ Health Perspect. 2000; search Priorities for Airborne Particulate Matter. Re- [Ser A]. 2006;169(part 2):179-203. 108:777-784. 26. Everson P, Morris C. Inference for multivariate nor- search Priorities for Airborne Particulate Matter, IV: Con- 48. Sheppard L, Levy D, Norris G, Larson TV, Koenig tinuing Research Progress. Washington, DC: National mal hierarchical models. J R Stat Soc [Ser B]. 2000;62: JQ. Effects of ambient air pollution on nonelderly asthma Academies Press; 2004. 399-412. hospital admissions in Seattle, Washington, 1987-1994. 2. National Center for Environmental Assessment. Air 27. Bell ML, Dominici F, Samet JM. A meta-analysis of Epidemiology. 1999;10:23-30. Quality Criteria for Particulate Matter. Research Tri- time-series studies of ozone and mortality with com- 49. Sheppard L. Reanalysis of effects of ambient air pol- angle Park, NC: US Environmental Protection Agency; parison to the national morbidity, mortality, and air pol- lution on nonelderly asthma hospital admissions in Se- 2004. lution study. Epidemiology. 2005;16:436-445. attle, WA, 1987-1994 . In: Revised Analyses of the Na- 3. US Environmental Protection Agency. National am- 28. Losina E, Barrett J, Baron JA, Katz JN. Accuracy of bient air quality standards for particulate matter. Fed Medicare claims data for rheumatologic diagnoses in total tional Morbidity, Mortality, and Air Pollution Study, Part II. Boston, Mass: Health Effects Institute; 2003: Regist. 1997;62:138. Available at: http://www.epa.gov hip replacement recipients. J Clin Epidemiol. 2003;56: 227-230. /ttn/caaa/t1/fr_notices/pmnaaqs.pdf. Accessed Feb- 515-519. 50. Burnett RT, Cakmak S, Brook JR, Krewski D. The ruary 9, 2006. 29. Fisher ES, Whaley FS, Krushat WM, et al. The ac- role of particulate size and chemistry in the association 4. US Environmental Protection Agency; Office of Re- curacy of Medicare’s hospital claims data: progress has between summertime ambient air pollution and hospi- search and Development. Air Quality Criteria for Par- been made, but problems remain. Am J Public Health. talization for cardiorespiratory diseases. Environ Health ticulate Matter. Washington, DC: US Government Print- 1992;82:243-248. Perspect. 1997;105:614-620. ing Office; 1996. 30. Kiyota Y, Schneeweiss S, Glynn RJ, et al. Accuracy 51. Stieb DM, Judek S, Burnett RT. Meta-analysis of 5. Samet JM, Dominici F, Curriero FC, Coursac I, Zeger of Medicare claims-based diagnosis of acute myocar- time-series studies of air pollution and mortality. J Air SL. Fine particulate air pollution and mortality in 20 U.S. dial infarction. Am Heart J. 2004;148:99-104. Waste Manag Assoc. 2003;53:258-261. cities, 1987-1994. N Engl J Med. 2000;343:1742-1749. 31. Baicker K, Chandra A, Skinner JS, Wennberg JE. Who 52. Office of Air Quality Planning and Standards. Qual- 6. Dominici F, Daniels M, Zeger SL, Samet JM. Air pol- you are and where you live. Health Aff (Millwood). 2004; lution and mortality. J Am Stat Assoc. 2002;97:100-111. 23:VAR33-VAR44 doi:10.1377/hlthaff.var.33. Ac- ity Assurance Project Plan: PM2.5 Speciation Trends Net- 7. Dominici F, McDermott A, Daniels M, Zeger SL, Samet cessed February 3, 2006. work Field Sampling. Research Triangle Park, NC: En- JM. Revised Analyses of Time-Series Studies of Air Pol- 32. Havranek EP, Wolfe P, Masoudi FA, Rathore SS, vironmental Protection Agency; 2000. Krumholz HM, Ordin DL. Provider and hospital char- 53. US Environmental Protection Agency Clean Air Act lution and Health: Mortality Among Residents of 90 Cities. Boston, Mass: Health Effects Institute; 2003. acteristics associated with geographic variation in the (amended in 1990). Available at: http://www.epa.gov 8. Peng RD, Dominici F, Pastor-Barriuso R, Zeger SL, evaluation and management of elderly patients with /oar/caa/contents.html. Accessibility verified February Samet JM. Seasonal analyses of air pollution and heart failure. Arch Intern Med. 2004;164:1186-1191. 15, 2006. ©2006 American Medical Association. All rights reserved. 1134 JAMA, March 8, 2006—Vol 295, No. 10 (Reprinted) Downloaded from www.jama.com by guest on February 16, 2009

×