Multi-site Time Series Analysis

Size: px
Start display at page:

Download "Multi-site Time Series Analysis"

Transcription

1 Multi-site Time Series Analysis Epidemiological Findings from the National Morbidity, Mortality, Air Pollution Study (NMMAPS) + the Medicare and Air Pollution Study (MCAPS) SAMSI Spatial Epidemiology Fall 2009 Howard Chang hhchang@jhsph.edu 1

2 Scientific Questions (1) Is there an association between daily mortality (and hospital admissions) and short-term exposure to air pollution? (2) Are the health effects spatially heterogeneous? Can we identify effect modifiers that explain the spatial variation in risk estimates? (3) Is there temporal variation in the health effects? (4) Is exposure to co-pollutant an confounder? (5) What is the exposure-response relation between air pollution and mortality? 2

3 Exploring Health Effects Spatial Heterogeneity 3

4 Regional Health Effects of PM 2.5 on Hospital Admissions Percent increase in hospital admissions per 10 mg/m3 increase in lag 0 PM 2.5. The size of the circle is proportional to the inverse of the estimate s variance. More certain estimates are shown with larger circles. Cardiovascular Admissions Respiratory Admissions Bell ML, Ebisu K, Peng RD, Samet JM, Dominici F (2009). "Hospital admissions and chemical composition of fine particle air pollution," American Journal of Respiratory and Critical Care Medicine, 179 (12),

5 Regional Health Effects of PM 10 on Daily Mortality A Three-level Hierarchical Model for Regional Differences ˆ β cr β cr ~ Normal ( β, Vˆ βcr ~ Normal β r ~ cr cr 2 ( β, τ 2 Normal ( β, σ r ) ) ) c = city; r = region F. Dominici, A. McDermott, M. Daniels, S. L. Zeger, and J. M. Samet. Mortality among residents of 90 cities. In Revised Analyses of Time-Series Studies of Air Pollution and Health, pages The Health Effects Institute, Cambridge, MA,

6 Spatial Correlation in PM 10 on Mortality Health effects from cities close to each other are more similar. ˆ β β c c ~ Normal ( β, Vˆ β ~ Normal c c c 2 ( β, τ d ( c, c') = Euclidean distance between city c and city c based on their longitude and latitude coordinates. ) ) cor( βc, βc' ) = exp{ φ [ d( c, c')] 0.5 } 6

7 Sensitivity to Heterogeneity Model Daily Mortality and PM 10 A. Marginal posterior distributions of national average estimates of PM10 effects on total mortality from nonexternal causes at lag 1. B Marginal posterior distribution of the standard deviation among cities of the true relative rate.the histogram represents the profile likelihood. F. Dominici, A. McDermott, M. Daniels, S. L. Zeger, and J. M. Samet. Mortality among residents of 90 cities. In Revised Analyses of Time-Series Studies of Air Pollution and Health, pages The Health Effects Institute, Cambridge, MA,

8 Sensitivity to Heterogeneity Model Daily Hospital Admission for Heart Failure and PM 2.5 cor ( ' βc, βc ) = exp{ φ [ d( c, c')]} φ = 1 (dotted), 0.1 (dashed), 0.01 (solid) Peng RD, Dominici F (2008). Statistical Methods for Environmental Epidemiology in R: A Case Study in Air Pollution and Health, Springer. 8

9 Particulate Matter as a Complex Mixture Example distribution of PM by Size Wilson, W. E. and Suh, H. H. (1997), \Fine particles and coarse particles: concentration relationships relevant to epidemiologic studies." J Air Waste Manag Assoc, 47,

10 Particulate Matter as a Complex Mixture Sources Constituents Coarse PM (PM ) Agricultural activities Windblown dust Sea spray Grinding and crushing Construction/demolition Combust burnout (fly ash) Crustal compounds Tire, brake, and pavement abrasion residue Pollen, endotoxin Fine PM (PM 2.5 ) Atmospheric reactions of gases Combustion processes Vehicle emission Industrial operations Power generation Organic carbon compounds Carbon black Inorganic ions Metals PM composition varies both spatially and temporally. Perhaps the PM composition can explain the observed risk heterogeneity. 10

11 Effect Modification by PM Chemical Constituents? Differences in mortality risk coefficients shown as the 5th 95th percentile difference in concentrations of fine PM components (long-term average) for the 60 NMMAP cities. ˆ β β β c c ~ c ~ Normal ( β, Vˆ Normal ( β 0 + β1zc, τ c c ) 2 ) Z c = longer-term average concentration of a constituent in county c. Abbreviations: EC, elemental carbon; OC, organic carbon. Lippmann M, Ito K, Hwang JS, Maciejczyk P, Chen LC Cardiovascular effects of Ni in ambient air. Environ Health Perspect 114:

12 Effect Modification by Nickel and Vanadium With longer study period: Ni Concentration (µg/m 3 ) V Concentration (µg/m 3 ) Dominici F, Peng RD, Ebisu K, Zeger SL, Samet JM, Bell M. Does the Effect of PM10 on Mortality Depend on PM Nickel and Vanadium Content? A Reanalysis of the NMMAPS Data, Environmental Health Perspectives 12

13 Effect Modification by Nickel and Vanadium Weaker effect and non-statistically significant when three counties from NY are excluded. NY cities are particularly high in Ni and V. Setting aside the three counties that belong to the New York community, the between-community variance of Ni is reduced by 68%. The Ni and V concentrations in the three New York counties were 8.9 and 3.4 times higher than the other counties, respectively. Ni and V PM2.5 chemical components in New York are likely attributed to oil-fired power plants and emissions from ships using oil Dominici F, Peng RD, Ebisu K, Zeger SL, Samet JM, Bell M. Does the Effect of PM10 on Mortality Depend on PM Nickel and Vanadium Content? A Reanalysis of the NMMAPS Data, Environmental Health Perspectives 13

14 Constituents on PM 2.5 and Hospital Admissions Percent increase in health effects estimates for PM 2.5 lag 0 Bell ML, Ebisu K, Peng RD, Samet JM, Dominici F (2009). "Hospital admissions and chemical composition of fine particle air pollution," American Journal of Respiratory and Critical Care Medicine, 179 (12),

15 Sulfate in PM 2.5 on Hospital Admissions % increase in admissions per 10 µg m 3 increase in PM cerebrovascular disease Northeast South Northwest Southeast Midwest West Central slope = 0.54 (0.25) heart failure Central South Northeast Northwest West Southeast Midwest slope = 0.03 (0.27) peripheral vascular South Midwest Northeast West Southeast Central Northwest slope = 0.56 (0.42) COPD Southeast Central West South Northeast Northwest Midwest slope = (0.57) ischemic heart disease West Northwest Central South Midwest Northeast Southeast slope = 0.21 (0.16) respiratory infection Northwest West Central South Southeast Midwest Northeast slope = (0.18) heart rhythm South Northeast Midwest Southeast West Northwest Central slope = 0.83 (0.29) Average sulfate µg m3 15

16 Rural versus Urban Coarse PM on Hospital Admissions For each 10 µg/m 3 increment of PM , a county with 1% higher urbanicity with respect to another county was estimated to have an additional 0.065% (95%PI, 0.002% %) increase in risk for hospital admission for cardiovascular diseases. Peng RD, Chang HH, Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F (2008). "Coarse particulate matter air pollution and hospital admissions for cardiovascular and respiratory diseases among Medicare patients," Journal of the American Medical Association, 299 (18),

17 PM 2.5 Speciation Network! A national network that measures more than 50 PM 2.5 constituents. Approximately 250 monitoring locations from year Can directly assess the health effects of each constituent. There are seven major constituents by % mass. Median (across 112 counties) correlation between daily concentrations of time series for pairs of pollutants. Peng RD, Bell ML, Geyh AS, McDermott A, Zeger SL, Samet JM, Dominici F (2009). "Emergency admissions for cardiovascular and respiratory diseases and the chemical composition of fine particle air pollution," Environmental Health Perspectives, 117 (6),

18 PM 2.5 Constituents and CVD Admissions National average estimates and 95% PIs for the percent increase in hospital admissions for CVD per IQR increase in each of the seven PM 2.5 components in 119 U.S. counties during : singlepollutant model (S; top row) and multipollutant model (M; bottom row). The multipollutant model (M*) for ammonium excludes sulfate and nitrate. Peng RD, Bell ML, Geyh AS, McDermott A, Zeger SL, Samet JM, Dominici F (2009). "Emergency admissions for cardiovascular and respiratory diseases and the chemical composition of fine particle air pollution," Environmental Health Perspectives, 117 (6),

19 Effect Modification by Air Conditioning The Hypothesis: Higher AC use lower ambient pollution penetration reduce personal exposure Bell ML, Ebisu K, Peng RD, Dominici F. "Adverse health effects of particulate air pollution: Modification by air conditioning (with commentary)," Epidemiology, 20 (5),

20 Effect Modification by Air Conditioning Samet J.M. Dominici F. Curriero F. Coursac I. and Zeger S.L. Particulate Air Pollution and Mortality: Findings from 20 U.S. Cities. New England Journal of Medicine, 343,24,

21 Effect Modifications in Ozone versus Daily Mortality L06 = ozone during the previous week Bell M. Dominici F. Effects Modification by Community Characteristics on the Short-Term Effects of Ozone on Mortality in 98 U.S. Communities, American Journal of Epidemiology 21

22 Exploring Time-varying Health Effects Short-term effects of air pollution may exhibit seasonality due to changes in Human activities and exposure Air pollution composition One example to include seasonality: log µ ct = lognct + αc + βc ( t) xct + confounders β c( t) = βc + βc 1 sin(2πt / 365) + βc2 cos(2π / 365) 0 t 22

23 Health Effects of PM 10 on Mortality National and regional smooth seasonal effects of PM 10 at a lag of 1 day for 100 US cities. Peng RD, Dominici F, Pastor-Barriuso R, Zeger SL, Samet JM (2005). "Seasonal analyses of air pollution and mortality in 100 U.S. cities," American Journal of Epidemiology, 161 (6) 23

24 Health Effects of PM 10 on Mortality Samples from the national and regional joint posterior distributions of the pooled coefficients from the sine/cosine seasonal model. The solid and dashed lines indicate the 75% and 95% regions for the joint posterior distribution. 24

25 Time-varying Health Effects of PM 10 on Mortality Seasonal patterns varied by geographic region, with a strong pattern for lag 1 appearing in the Northeast. The log relative rate is greater in the spring and summer in the northern regions. Some Hypotheses: Higher level of more toxic PM constituents? People spending more time outdoor and less exposure error? In the winter, more powerful risks factors (influenza) swamp the air pollution effect. An unknown source of bias that varies seasonally Peng RD, Dominici F, Pastor-Barriuso R, Zeger SL, Samet JM (2005). "Seasonal analyses of air pollution and mortality in 100 U.S. cities," American Journal of Epidemiology, 161 (6) 25

26 Health Effects of PM 2.5 on CVD Hospital Admissions % increase in total CVD hospital admissions ( ) per 10 µg/m 3 increase in same-day (lag 0) PM 2.5, by US region, for results from the harmonic model (curved lines) and seasonal interaction model using indicators for each season (straight lines). Bell ML, Ebisu K, Peng RD, Walker J, Samet JM, Zeger SL, Dominici F (2008). "Seasonal and regional short-term effects of fine particles on hospital admissions in 202 U.S. counties, ," American Journal of Epidemiology, 168 (11),

27 Perhaps PM 2.5 Constituents Again? Bell ML, Ebisu K, Peng RD, Walker J, Samet JM, Zeger SL, Dominici F (2008). "Seasonal and regional short-term effects of fine particles on hospital admissions in 202 U.S. counties, ," American Journal of Epidemiology, 168 (11),

28 A More Ambitious Search for Effect Modifiers To increase information for Stage II analysis, we can fractionate city-specific data into month-year-city data. Assess how city information, and monthly co-pollutant can modify health effects. Eckel SP and Louis TA (2007). Identifying eect modiers in air pollution time-series studies using a two-stage analysis. Johns Hopkins University, Dept. of Biostatistics Working Papers. 28