Measuring respiratory impacts of particulate matter reductions by environmental public health indicators

Measuring Respiratory Health Impacts of Particulate Matter Reductions by Environmental Public Health Indicators,
Minneapolis-St. Paul, 2003-2009
NAOMI SHINODA, Paula Lindgren, Jean Johnson, Allan Williams, Minnesota Department of Health, St. Paul, MN
Gregory C. Pratt, Kari Palmer, Lisa Herschberger, Minnesota Pollution Control Agency, St. Paul, MN; Barbara Yawn, Peter Wollan, Olmsted Medical Center, Rochester, MN
Results
Background
Air Pollution Reduction Initiatives in Minneapolis-St. Paul metro, 2003-2009
• Concurrent nationwide rules and local Minneapolis-St. Paul (MSP) metro initiatives
targeted air pollution reduction during 2003-2009
• Large SO2 and NOx power plant emission reductions occurred in MN and regionally,
including 3 large power plants under MN Metro Emissions Reduction Project (MERP)
• National reductions in fuel sulfur content also occurred
24-hr PM2.5 NAAQS revision
20
2007
Early Implementation
2008
% Change per 10 µg/m3 PM2.5
Baseline Period
2006
Time Period (Years)
12
2009
Implementation
MERP Plant 3
converted coal to natural gas
MERP Plant 2
converted coal to natural gas
MERP Plant 1
CLRD
Hosp.*
33,429
Asthma
Hosp.*
16,703
Mean PM2.5
(µg/m3)
10.8
8
4.7
4
2.8
0
Population Exposure:
• Daily 24-hr PM2.5 averages from six MSP metro Beta Attenuation
Mass (BAM) continuous monitors
• Short-term exposures: lag0, lag1, lag2, and averages of lag days
Health Outcomes:
• Total respiratory, chronic lower respiratory disease (CLRD), and asthma
hospitalizations from MN Hospital Discharge Data
Analyses:
• Case-crossover analysis with time-stratified referent selection (28-day strata)
• Poisson model time series analysis for daily counts of health outcomes
• Covariates: average temperature, relative humidity, influenza ED visits, holidays
Indicators of Exposure-Health Association:
• Odds ratios (from case-crossover and time series) -- % change in hospitalizations
per unit increase in PM2.5
• Population attributable fractions (from case-crossover) -- excess hospitalizations
that occurred in study population that were triggered by observed levels of PM2.5
0.4
-4
-8
2003-2005 2006-2007 2008-2009 2003-2005 2006-2007 2008-2009
Case-crossover
12
8
4
5.1
5.5
3.7
5.2
0.3
0
1.1
-4
-8
2003-2005 2006-2007 2008-2009 2003-2005 2006-2007 2008-2009
Time Series
Case-crossover
Time Series
8
6.8
6.2
6.1
4
1.3
0
2.9
4.6
-4
-8
2003-2005 2006-2007 2008-2009 2003-2005 2006-2007 2008-2009
Case-crossover
Time Series
** Intervals shown are 95% CIs. All time series models adjusted for avg. temperature lags 0 and 1, relative humidity lags 0 and 1, day after national holidays, day of week, time period, daily flu count, and trend variables for linear trend as well as 2yr 1yr, 6mo, 4mo, and 3mo cycles.
Case crossover models for total respiratory hospitalizations adjusted for cubic splines of avg. temperature lag 0, relative humidity lag 0, and day after national holidays.
Case-crossover models for CLRD hospitalizations adjusted for avg. temperature lag 0, relative humidity lag 0, day after national holidays, and number of influenza ED visits per week.
Case-crossover models for asthma hospitalizations adjusted for avg. temperature lag 0, relative humidity lag 0, day after national holidays, and number of influenza ED visits per week.
What are population attributable fractions (PAFs) ?
• PAFs quantify excess hospitalizations triggered by PM2.5 exposure, assuming a causal
association
• They take into account the proportion of people exposed to above-background
ambient PM2.5 levels
• PAFs can put ORs into policy-relevant context; can interpret in terms of economic
burden
Population attributable fractions (PAFs) from case-crossover results
• During baseline (2003-2005), the % of hospitalizations triggered by a 3-day mean of
ambient PM2.5 exposure greater than policy-relevant background (5µg/m3) were:
3% (1.5% - 4.5%) of total respiratory hosp. (about 1,050 hospitalizations)
3% (0.7% - 5.5%) of CLRD hosp. (about 440 hospitalizations)
4% (0.7% - 7.2%) of asthma hosp. (about 290 hospitalizations)
• During implementation period (2008-2009), these PAFs approached zero
Discussion and Conclusions
Why did ORs change over time?
• Change in susceptibility of the underlying population, health care utilization, or other
external factors that affected respiratory hospitalization trends?
• Change in PM2.5 composition or surrogate chemicals?
• Observed downward trend (p < 0.05) during 2003-2009 of ambient NO2, SO2, and
PM2.5 sulfate concentrations in MSP
• Regional decreases in NOx and SO2 emissions, including at local MERP power
plants:
Tons (in thousands)
Study Area: Minneapolis-St. Paul (MSP) 7-county metropolitan area
Study Time Periods: 2003-2005, 2006-2007, 2008-2009
2.3
-1.0
2005-2009 continuous
Methods
4.6
Asthma Hospitalizations**
16
12
Diesel vehicle retrofits added emissions controls
Objectives:
• Develop and evaluate environmental public health indicators (EPHI) to measure and
track impacts of air pollution reduction on PM2.5 exposure and related respiratory
health outcomes
• Utilize data readily available to state/local agencies
20
CLRD Hospitalizations**
16
30
20
NOx
SO2
10
0
2003 2004 2005 2006 2007 2008 2009
10
8
6
4
2
0
NOx
SO2
2003 2004 2005 2006 2007 2008 2009
Tons (in thousands)
2005
20
Total Respiratory Hospitalizations**
16
Tons (in thousands)
2004
Total Resp.
Hosp.*
82,213
All Years
Case-Crossover and Time Series Analyses
• Odds ratios (ORs) approached null in implementation period (2008-2009) compared with
Baseline (‘03-’05)
34,926
14,218
7,209
11.2
baseline (2003-2005) for all analyses except time series analysis for asthma hospitalizations
Early Implement. (‘06-'07)
23,838
9,697
5,069
10.4
• Time series results generally show smaller PM2.5-health associations than case-crossover results,
Implementation (‘08-’09)
23,449
9,514
4,425
10.4
but association time trends are similar (with exception of asthma hospitalizations)
* Total respiratory hosp.: ICD9-CM = 464, 466, 480-487, 490-496; CLRD hosp: ICD9-CM = 490-496; Asthma hosp.: ICD9-CM = 493
% Change per 10 µg/m3 PM2.5
Heavy Duty Diesel Rule
(adopted 2005; remanded 2008)
2003
Number of hospitalizations and mean PM2.5 concentrations, MSP, 2003-2009
Associations between respiratory hospitalizations and 3-day mean PM2.5 concentrations (lags 0-2) by analysis method, MSP, 2003-2009
Ultra Low Sulfur Fuel
Clean Air Interstate Rule
PM2.5 time trends
• mean PM2.5 concentrations lower in later time periods compared with baseline time period
(p < 0.05)
% Change per 10 µg/m3 PM2.5
R833627010
20
15
10
NOx
SO2
5
0
2003 2004 2005 2006 2007 2008 2009
• Pittsburgh case-crossover study (Xu 2008) for PM10 and cardiorespiratory
hospitalizations also saw change in ORs before and after closure of a steel coke oven
Using EPHI to measure local area impacts
• Some impacts difficult to measure because centralized ambient air monitoring data may
not accurately reflect very localized emission sources (e.g. diesel vehicle retrofits)
• PAFs are potentially useful measures , but are dependent on reliable OR estimates as well
as the chosen policy-relevant background level of exposure
Conclusions
• Analysis method (case-crossover vs. time series) can depend on available resources:
• Case-crossover requires individual-level health data but less statistical expertise
• Time series does not require individual-level health data, but requires more
statistical expertise
• Collaborative relationships (epidemiologists, air quality scientists) helpful to fully
understand complexities of both health and air quality data
• Recommend further evaluation/exploration with other health outcomes, more years of
data