Health Impacts of Aerosols

1
Health Impacts of Aerosols

• Dr. Patrick L. Kinney
• Associate Professor
• Columbia University

2
Overview

• What do we know about health effects of PM?
• Experimental vs. epidemiologic methods
• Time series and cross sectional study designs and
  methods
• Case studies
• New health research directions
• Monitoring and modeling data needs

3
The Human Respiratory System

• Three key regions
• Extrathoracic
• Tracheobronchial
• alveolar
• Particle penetration and deposition vary by
  region
• Lung defenses vary by region
• Vulnerability varies by region

4
(No Transcript)
5
Estimated Deposition for Adult Male based on
LUDEP model
6
How Can Air Pollution Cause Problems?

• Irritation of the airways in the extrathoracic
  region, resulting in symptoms such as runny nose
  and sneeze
• Irritation and inflammation of the conducting
  airways in the tracheobroncial region, resulting
  in symptoms of cough, wheeze, and shortness of
  breath (asthma-like symptoms), and possibly
  long-term effects such as bronchitis or lung
  cancer
• Damage to the alveolar cells, resulting in
  scarring, remodeling, and decreased lung
  capacity, which may lead eventually to
  clinically-significant fibrosis or emphysema
• Penetration through the epithelial lining to the
  circulatory system and thence to other organs,
  such as the heart

7
Health Effects of Airborne Particulate Matter

• Historical experience during severe episodes
  provides strong evidence for a cause-effect
  relationship between air pollution and premature
  death
• For example, London 1952
• It has been argued, though it cannot be proven,
  that PM was the responsible pollutant in the
  London episode

8
London Killer Fog, December, 1952
9
London Mid-day in December 1952
10
(No Transcript)
11
Health Effects, continued

• Modern epidemiology studies have repeatedly found
  statistically significant associations between PM
  and the risk of death and disease
• Two primary epidemiologic study designs
• Studies examining changes in exposure over time,
  which assess mainly acute effects
• Studies examining changes in exposure over space,
  which assess mainly chronic effects
• Experimental studies also provide supporting
  information

12
What are acute studies telling us?

• Time-series studies show that
• Risk of death is acutely elevated on days when PM
  levels are high
• Risk of hospital visits or admissions are also
  acutely related to PM levels
• Respiratory and cardiovascular causes of death
  and disease are most associated with PM

13
Important Disease Categories in the Study of PM
Health Effects

• Respiratory
• Asthma
• Bronchitis
• Pneumonia
• Emphysema
• COPD

• Cardiovascular
• Heart attack
• Arrhythmia
• Congestive heart failure

• Control conditions
• Digestive diseases

14
What are chronic studies saying?

• Spatial studies show that
• Risk of death is higher in cities with higher
  long-term PM concentrations
• This relationship remains after controlling for
  differences in smoking rates, economic status,
  diet, occupation, and other factors
• The chronic PM effect is substantially larger
  than the acute effect and is probably more
  significant in its overall population impacts

15
Schematic view of the relationship between
long-term and short-term effects Adapted from
Kunzli et al. 2001
16
Description of Categories
17
What else does the health literature tell us?

• More recent epidemiologic studies have
  investigated possible mechanisms of acute
  effects, by observing physiologic or biochemical
  changes in groups of subjects followed over time.
  
• For example, recent studies have demonstrated
  associations of PM with
• decreased heart rate variability
• increased arrhythmias
• levels of substances in the blood that promote
  inflammation and blood clotting.

18
Mechanistic Hypotheses
19
What havent we learned?

• Epidemiologic studies tell us very little about
• Which particle components are responsible for the
  observed health effects?
• Coarse vs. fine vs. ultrafine modes
• Sulfate vs. nitrate vs. elemental carbon vs.
  organic carbon vs. trace elements and metals
• What sources are most responsible?
• Motor vehicles vs. coal vs. fuel oil vs.
  windblown dust
• Why Not?

20
Epidemiology is opportunistic

• In most cases, exposure assessment relies upon
  air monitoring data collected for regulatory
  purposes
• Regulatory air monitoring for PM has seldom
  provided detailed size or chemical speciation
• If theres no speciation data, then epidemiologic
  studies cannot look at differential effects of
  different PM components
• Even in cases where speciated PM data exist, it
  is hard to statistically separate the various
  components when they correlate with each other

21
Experimental Studies Can Help Answer the PM
Component Questions

• Controlled exposures of animals to component
  particles
• Controlled exposures of human volunteers to
  component particles
• In-vitro exposures to cells

22
What are some advantages of Experimental Studies?

• Well-characterized air pollution exposures
• Control of other environmental conditions that
  might affect the outcome
• With appropriate study design, can prove a
  cause-effect relationship
• Can reliably evaluate shape of exposure-response
  relationship

23
What are some limitations of Experimental Studies?

• Complex pollution mixtures are technically
  challenging
• To address this challenge, ambient pollution is
  sometimes used to generate exposures
• In the case of PM, concentrators have become
  popular in experimental studies
• Animal and in-vitro results may not be directly
  applicable to human risk assessment
• Human studies involve healthy volunteers, so may
  not tell us whats going on in susceptible
  populations
• Exposures much higher than ambient are often
  used, necessitating extrapolation to lower levels
• Human studies limited to acute effects

24
Air Pollution Epidemiology

• Pros
• Provides results which are directly relevant for
  policy makers
• Assesses effects of the real-world mix of
  pollutants on relevant human populations
• No need to extrapolate across species
• Little need for extrapolation across exposure
  levels
• Study population may include susceptible subgroups

25
Air Pollution Epidemiology

• Cons
• Pollutants tend to co-vary, making it hard to
  identify effects of a specific pollutant
• Can demonstrate associations between outcome and
  exposure, but cannot prove cause and effect.
  Were stuck with causal inference
• Various schemes exist for building an argument
  for causality
• Shape of exposure-response relationship difficult
  to discern. Difficult to identify thresholds
• Must control for confounding factors
• Exposures are always ecologic. Outcomes and
  covariates often are too

26
Confounding a serious issue

• In an analysis of the effects of exposure on an
  outcome, confounding occurs when there is a third
  variable, omitted from the analysis, that
  independently affects the outcome and which is
  also correlated with the exposure variable of
  interest
• In a multi-city study of smoking and heart
  disease, high fat diet might be a confounder
• In a time series study of ozone and daily deaths,
  temperature might be a confounder

27
Confounding, continued

• When confounding is present, the estimate of the
  exposure-response relationship of interest may be
  biased up or down
• This is because some of the effect of the omitted
  confounding variable is attributed to, or is
  picked up by, the exposure variable of interest
  
• If not addressed, confounding may invalidate the
  findings of a study
• Solution control for the confounding variable,
  either by exclusion, stratification, or by
  including it as a covariate in the analysis

28
Effect Modification

• Occurs when the level of a third variable
  influences the exposure-response relationship for
  the variable of interest
• For example, people who smoke cigarettes are more
  likely to get lung cancer following asbestos
  exposure than are non-smokers
• Same idea as an interaction
• Does not invalidate a study, but may affect the
  generalizability of results

29
What is meant by Ecologic Variable?

• An ecologic variable is one for which information
  is not available uniquely for each individual in
  the study, but rather is available only for
  groups of individuals
• Air pollution variables are always ecologic, in
  that data from one or a few monitoring sites is
  used to represent exposure for a large group of
  people
• Outcome and covariate data sometimes are ecologic
• For example, daily death counts in a city

30
Why does this matter?

• Especially when outcomes and potential
  confounding variables are ecologic, there is
  greater concern about potential confounding by
  group-level factors that are hard to measure and
  control
• Old cross sectional epidemiology studies were
  criticized for this reason (see Lave and Seskin
  studies from 1970s)

31
Two Main Study Designs

• Time Series or temporal, acute, short-term
• Cross Sectional or cohort, chronic, long-term

32
Time Series Study DesignSee Pope and Schwartz
1996 handout

• Examines acute exposure-response relationships
• Data must be available at equally spaced
  intervals (usually days) over extended time
  period
• Outcome data may be continuous (e.g., lung
  function), dichotomous (e.g., symptoms), or count
  (e.g., deaths)
• Multiple regression is commonly used to examine
  exposure-response relationships

33
The Multiple Linear Regression Model

• Yt a b1 X1t b2 X2t et

Slopes
Intercept
Residual
Outcome at time t
Level of variable 1 at time t
Level of variable 2 at time t
34
Time Series Analysis

• Statistical Challenges
• Residuals (deviations between observed and
  modeled outcome data) may not be normally
  distributed
• For counts, use Poisson regression
• Residuals may not be independent of each other
  especially over time
• Autoregressive terms address this issue
• Seasonal cycles and weather variables are often
  important confounders
• Co-pollutants may also be confounders

35
Control for Seasonal Confounding

• General approach is to include a new variable or
  function which fits the seasonal pattern of the
  outcome variable, thereby eliminating the
  opportunity for pollution to explain the seasonal
  variations
• The goal is to eliminate any cyclic patterns from
  the data with periodicities of greater than
  several weeks, leaving only high frequency data
  variations
• Options include filtering using a weighted moving
  average, fitting of sinusoidal functions of time,
  fitting generalized additive models (GAM)
• In general, all seem to perform similarly

36
Control for Weather Confounding

• For many health outcomes, such as death, we know
  that weather is a risk factor. Both heat spells
  and cold snaps are associated with higher death
  rates
• Since weather is closely tied to air pollution
  concentrations, confounding is possible
• This problem is easily addressed by including
  weather variables, such as temperature, in the
  regression analysis

37
A Note About GAM

• In the 1990s, the Splus GAM function became a
  popular method for fitting the non-linear
  exposure-response function for seasonal and
  temperature influences on daily deaths
• GAM with LOESS involves fitting a local, weighted
  moving regression between two variables
• In 2002, it was realized that the Splus GAM/loess
  algorithm has some problems

38
GAM continued

• When GAM/loess is used to fit season and/or
  temperature effects in a multiple regression
  model with pollution,
• The slope estimate for pollution may not converge
  to its optimal solution, resulting in positive
  bias
• Also, the standard error estimate for the slope
  on pollution is negatively biased
• But, the effects are small and not qualitatively
  important

39
Time Series Results

• A large number of studies have reported
  significant associations between daily deaths
  and/or hospital visit counts and daily average
  air pollution
• Particles often appear most important, but CO,
  SO2, NO2, and/or ozone may also play roles
• For example, NMMAPS Study

40
National Mortality and Morbidity Air Pollution
Study (NMMAPS)
Objectives

• To investigate acute effects of PM10 on daily
  deaths and hospital admissions, controlling for
  other pollutants
• To carry out a comprehensive analysis for
  multiple US cities using a consistent statistical
  approach

41
(No Transcript)
42
Cohort Epidemiology

• Address long-term exposure-response window
• Large populations in multiple cities enrolled and
  then followed for many years to determine
  mortality experience
• Cox proportional hazards modeling to determine
  associations with pollution exposure
• Must control for spatial confounders, e.g.,
  smoking, income, race, diet, occupation
• Assessment of confounders at individual level is
  a major advantage over former cross-sectional,
  ecologic studies

43
Pope, C.A. et al., Journal of the American
Medical Association 287, 1132-1141, 2002 (see
handout)
44
Pope et al., 2002

• Context although acute impacts of PM on
  mortality have been well-documented, studies of
  health effects from long-term PM have been less
  conclusive
• Objective to assess the relationship between
  long-term exposure to fine PM and all-cause, lung
  cancer, and cardiopulmonary mortality

45
Methods

• Added air pollution exposure assessment to an
  existing long-term study of cancer among 500,000
  adults enrolled in 1982 from 50 states in the US
• Vital status and cause of death recorded for each
  subject through the end of 1998
• Analyzed relationship of mortality risk to fine
  PM and other pollutant exposures
• Controlled for important confounders
• Tested for effect modification

46
Study Participants

• Were enrolled by American Cancer Society
  volunteers, and consisted of friends, neighbors,
  or acquaintances
• Aged 30 and over
• At enrollment, each completed a questionnaire
  addressing age, sex, weight, height, smoking
  history, alcohol use, occupational exposures,
  diet, education, and marital status

47
Exposure Assessment

• Each participant was assigned a metropolitan area
  of residence based on their location at
  enrollment
• The annual mean concentrations of all monitors
  throughout the metropolitan area were averaged
• Fine PM and other air pollutants were available
  for various time periods between 1980 and 1998

48
(No Transcript)
49
(No Transcript)
50
Overall Results for PM2.5
51
Is it Linear?
52
How Sensitive are the PM2.5 Results to Various
Confounders and Modeling Choices?
53
Is there Effect Modification?
54
Do Other Pollutants Play a Role?
55
Conclusion

• Long-term exposure to combustion-related fine
  particle air pollution is an important
  environmental risk factor for cardiopulmonary and
  lung cancer mortality.

56
Summary of PM Epidemiology

• Daily time-series studies have demonstrated small
  but consistent associations of PM with mortality
  and hospital admissions, reflecting acute effects
  
• Multi-city prospective cohort studies have shown
  increased mortality risk for cities with higher
  long-term PM concentrations, reflecting chronic
  effects

57
Implications

• Acute effects are well documented but of
  uncertain significance due to questions about how
  much life is lost
• Chronic effects imply very large impacts on
  public health.
• A new US national ambient air quality standard
  for PM2.5 was established in 1997, largely based
  on the cohort epidemiology evidence
• Mechanistic explanation for effects remains
  unclear but is the subject of current research
• Weaknesses in exposure assessment limits
  interpretation

58
It is also unclear

• Whether a threshold exists
• Who is at risk due to
• Higher exposures
• Greater susceptibility
• What particle components are most toxic
• Which sources should be controlled

59
Effects of Long-term Air Pollution Exposures on
Lung Function in Young Adults the Yale Study

• There is increasing concern about the human
  health impacts of long-term particulate matter
  (PM) exposures.
• Small airways function represents a potentially
  sensitive measure of early PM-related effects on
  the lung. We know e.g., that first signs of
  adverse effects due to smoking occur in the small
  airways.
• Few environmental epidemiology studies have
  examined lung function outcomes in the age range
  (late teens to mid 20s) when maximal function is
  achieved, or have included long-term PM exposure
  estimates.
• The Yale cohort study combined physiological
  measurements of small airways function, with
  life-time PM (and ozone) exposure estimates among
  young adults, to test the hypothesis that
• Life-time exposures to PM10 and/or Ozone are
  associated with diminished lung function in a
  nationwide cohort of young adults.

60
Yale Study Methods

• 1723 subjects were recruited over a three year
  period from freshman (first year) classes at Yale
  University in New Haven, CT.
• The present report focuses on the subset of 1578
  subjects who lived in the United States prior to
  attending Yale University.
• Each subject completed a questionnaire addressing
  respiratory disease and symptom history,
  residential history, home characteristics,
  childhood activity patterns, personal and family
  smoking history, parental education (SES), and
  other factors.

61
Lung Function Assessment

• Standing height measured in stocking feet.
• At least 3 maximal forced expirations performed
  by each subject.
• Blow selection based on standard ATS
  recommendations. Adjustment to BTPS.
• Lung function variables
• FVC Forced vital capacity
• FEV1 Forced expiratory volume in 1 second,
• FEF25-75 Mid-maximal flow rate
• FEF75 Flow rate at 75 of FVC

62
Exposure Assessment

• Annual mean PM10 concentrations were computed for
  all US sites in operation from 1983-1997.
• Annual means were extrapolated to the years
  1972-1982 using site-specific linear regressions
  (annual mean regressed on year) from 1983-1997.
• Annual mean PM10 concentrations were interpolated
  from 3 nearest monitoring sites to subject
  residential locations for all years from
  1972-1997.
• Life-time average PM10 concentrations were
  computed for each subject.
• A similar procedure was used to estimate
  long-term ozone exposures, based on June-August
  mean daily maximal.

63
Data Analysis

• Multiple linear regression was used to relate
  lung function to long-term average PM10 and ozone
  exposures, controlling for covariates.
• Covariates in the lung function models included
  height, height squared, sex, race, personal and
  maternal smoking, and parental education level.

64
Subject Characteristics
65
Distributions of long-term average air pollution
exposures
66
Percent changes in lung function per 1 SD change
in life-time average PM10 exposure results from
two-pollutant models, adjusted for covariates.
plt.05
67
Yale Study Conclusions

• We found associations between long-term average
  exposures to ambient PM10 and diminished
  small-airways function in a college student
  cohort with varying life-time residential and
  exposure histories. No associations were
  observed for ozone in two-pollutant models.
• To the extent that the study population was of
  high socioeconomic status, these results may
  underestimate effects (recent evidence from the
  ACS cohort showed higher chronic mortality risk
  at lower SES Pope et al., JAMA, 2002).
• Results of this study provide new insights into
  potential pathophysiologic linkages between
  long-term PM exposures and ill-health.

68
Remaining Questions on Yale Study Results

• Do lung function decrements persist into
  adulthood?
• Do they increase the risk of chronic lung
  disease?
• What component of the particle mix is responsible?

69
Monitoring Data Needs

• Daily speciated PM data
• Daily size-selected PM data for ultrafine and
  accumulation mode particles
• New portable, lightweight monitors for personal
  sampling of PM with speciation
• Correlation between personal and central-site
  concentrations for speciated PM
• Better data on fine-scale spatial patterns of PM
  species and size classes resulting from source
  influences in urban areas

70
Modeling Data Needs

• Models for near-source impact assessment over
  fine spatial, but not necessarily temporal,
  scales
• Integration of modeling into exposure
  assessments, to fill the gaps in available
  monitoring data