1. 01/22/2004 Assessing the Health Effects of Atlanta’s Air Pollution Jennifer L. Peel, PhD, MPH Emory University Rollins School of Public Health 01/22/2004

2. Emory University Funded by EPA, NIEHS, and EPRI

3. 01/22/2004 SOPHIA/APNEA Study Team  Paige Tolbert, PI  Mitch Klein  Jennifer Peel  Kristi Metzger  Dana Flanders  Sarah Tinker  Matt Strickland  Jim Mulholland  Ted Russell  Knox Todd  Lance Waller  Bob Lyles …And others!

4. 01/22/2004 Background: London Fog of 1952 Temperature inversion concentrated smoke and other pollutants Mortality increased 200% in week following episode Cardiorespiratory conditions accounted for 81% of excess deaths (compared to usual 30%)

5. 01/22/2004 Unresolved Questions  NRC Research Priorities, 1998: Need to understand physicochemical characteristics of PM underlying observed associations of PM with cardiorespiratory outcomes »Responsible components »Control for copollutants »Mechanism of injury

6. 01/22/2004 Aerial View of ARIES Monitoring Station, Atlanta, GA

7. 01/22/2004 ARIES Air Quality Data PM 10 PM 2.5 PM 10-2.5 PM 2.5 components Ultrafine PM O 3 NO 2 CO SO 2 Polar VOCs Sulfates Water-soluble metals Acidity Organic carbon Elemental carbon

8. 01/22/2004 Ongoing ARIES-Related Studies at Emory u ED study (NIH, EPA, EPRI) u ICD study (EPRI) u Medicare hospital admission study (NIH) u APNEA study (NIH)

9. 01/22/2004 ED Study: Methods  Obtain air quality data –ARIES monitor back to 8/98 –state monitoring network back to 1/93  Collect billing records from all Atlanta EDs back to 1/93  Define multiple cardiorespiratory case groups based on ICD-9 codes  Examine relation of case counts with pollutant measures

10. 01/22/2004 Hospital Participation through 2000

11. 01/22/2004

12. Data Analysis u A priori model: Poisson model –GEE –GLM u 3-day moving average (lag 0,1,2) for pollutant u Control for time trends, day of week, hospital entry/exit, holidays, mean temperature, mean dewpoint u Secondary models assessed (e.g., GAMs, alternative lag structures, season-specific)

13. 01/22/2004  Model specification based on literature –Air quality measures –Lag structure –Control for confounding  Reduce “data mining” for primary results  Reduce reporting bias based on “best” result Use of A Priori Model

14. 01/22/2004 Explanation of Estimates 2.5% increase in all CVD ED visits per 20 ppb increase in NO 2 OutcomePollutantUnitRR95% CI All CVDNO 2 20 ppb1.0251.012, 1.039

15. 01/22/2004 Results for All Respiratory Disease Visits through 2000

16. 01/22/2004 Results for RD Subgroups through 2000 RDURIAsthmaPneumCOPD Finger Wounds PM 10 ++ O3O3 ++ NO 2 +++ CO+++ SO 2 PM 2.5 Coarse PM Ultrafines- Metals Sulfates Acidity OC+ EC OHC 8/98-8/00 1/93-8/00

17. 01/22/2004 Results for All CVD Visits through 2000

18. 01/22/2004 Results for CVD Subgroups through 2000 CVDDYSCACHFIHDAMIPERI Finger Wounds PM 10 O3O3 NO 2 +++ CO++ SO 2 PM 2.5 +++ Coarse PM Ultrafines- Metals Sulfates Acidity OC++ EC++ OHC++ 8/98-8/00 1/93-8/00

19. 01/22/2004 Findings (through 2000) u CVD outcomes –Observed associations with NO 2, CO, PM 2.5, EC, OC, OHC –Pollution from mobile sources may play important role –Strongest effects with same-day levels –Epidemiology 2004;15(1):46-56

20. 01/22/2004 Findings (through 2000) u Respiratory outcomes –Associations from a priori models »All respiratory illness and URI: ozone, NO 2, CO, PM 10 »COPD: NO 2, CO »Pneumonia: OC –Strongest associations for all RD with pollution levels from recent days –Asthma: strongest associations with pollution levels lagged 5-7 days –In review, Epidemiology

21. 01/22/2004 Interpretation Issues u Ambient vs. personal exposure –Ambient exposure is of interest »Population-level health effects »Regulatory purposes –Personal behavior (e.g., air conditioning use, time spent outdoors) »May affect personal exposure levels »May affect magnitude of associations when compared to other locations with different behavior profiles

22. 01/22/2004  Measurement error in AQ data –Single, centrally-located monitor »Instrument error »Error from local sources »Error due to spatial heterogeneity –May attenuate estimates »Unlikely to lead to spurious results »Possible explanation where no effect observed Interpretation Issues (cont’d)

23. 01/22/2004 Interpretation Issues (cont’d) u Correlated co-pollutants acting as surrogates for less well-measured or unmeasured pollutant? –Relative amount of measurement error in air quality measures important –Multi-pollutant models can be as misleading as single- pollutant models

24. 01/22/2004 Current Work  Spatial analysis  Measurement error assessment  Source apportionment  Alternative analyses (e.g., distributed lag models, case- crossover)  Air quality data through 2002  Updating outcome databases for all studies (ED, ICD, APNEA) through 2002