Study designs in air pollution epidemiology

1
Study designs in air pollution epidemiology

• A/Prof Bin Jalaludin
• MBBS MPH PhD MRCP (UK) FAFPHM
• South Western Sydney Area Health Service, and
• University of New South Wales
• Dr Geoff Morgan
• BSc DipED PhD
• Northern Rivers University Department of Rural
  Health -
• University of Sydney, and Northern Rivers Area
  Health Service

2
This talk

• Study designs in air pollution epidemiology
• Strengths and weaknesses of study designs
• Illustrative examples including brief
  discussion of statistical methods

3
http//biosun01.biostat.jhsph.edu/fdominic/
4
http//biosun01.biostat.jhsph.edu/fdominic/
5
Exposure Assessment

• Generally from fixed site ambient monitors
• Fixed site ambient monitors only practical
  solution when large numbers exposed
• More practical/low cost personal exposure
  monitoring data becoming available
• The weak link in epidemiological studies
• EPAs regulate ambient air pollution so exposure
  response estimates relating to ambient exposure
  are required

6
Determinants of exposure, dose and biologically
effective dose that underlie the development of
health effects (Modified from Jaakkola et al.,
1994)
7
Pollutant Exposure (External dose)
Susceptibility Factors Internal Dose Susceptibil
ity Factors Metabolites, Adducts,
or Biologically Effective Dose Susceptibility
Factors
Early Biological Effects Altered Structure and
Function Ultimately Clinical Disease
8
Types of studies

• Animal studies
• Human studies
• Experimental (chamber studies)
• Epidemiological (or observational)
• Cross-sectional
• Case-control
• Cohort
• Time series
• Quasi-experimental

9
Epidemiological studies of air pollution

• Aggregated data
• Geographical (spatial) observation unit
• Cross sectional
• Temporal observation unit
• Time Series Study
• Individual data
• Cross sectional (sometimes with spatial studies)
• Panel studies (diary studies)
• Cohort Studies

10
Holgate ST, Samet JM, Koren HS, editors. Air
Pollution and Health. UK Academic Press 1999.
11
Study Designs
(Sheppeard 2003)
12
Animal studies

• Easier and cheaper
• Can use higher exposures
• Can use precise mix of pollutants
• Can control most variables
• Fewer ethical issues
• BUT
• ??Extrapolation to humans

13
Experimental studies

• Also known as chamber studies
• NO2, SO2 and O3
• Often directed at understanding mechanisms of
  injury
• Endpoints studied
• pulmonary function
• airway hyperresponsiveness
• airway inflammation

14
Strengths of experimental studies

• Opportunity to create controlled and standardized
  environments
• Can be standardized to particular workloads,
  breathing patterns, types of exposure (single
  pollutant or more complex mixtures of gases),
  durations of exposure and number of exposures

15
Strengths of experimental studies (cont)

• Low concentrations with long duration exposures
  can be studied (to mimic ambient environments) as
  well as high concentrations with long durations
  (to mimic worse case scenarios)
• Dose-response information can be generated
• Interaction between two or more pollutants can be
  investigated by varying the concentrations of
  each of the pollutants

16
Strengths of experimental studies (cont)

• Minimization of bias
• Randomization, double blinding and controlled
  measurements of exposure and outcome factors
  ensure greater internal validity
• Results are more pertinent than results from
  animal studies as there often are intra- and
  inter-species differences

17
Weaknesses of experimental studies

• A small number of, usually healthy, adult
  volunteer subjects
• Hence, small, but important, changes may not be
  able to be detected
• Extrapolation and generalisability of results to
  the general population and in particular to high
  risk subgroups may not be possible

18
Weaknesses of experimental studies (cont)

• Most often only single pollutants are studied
• Concentrations studied are usually higher than
  that encountered in the ambient environment
• Chronic effects cannot be readily addressed

19
The effect of ozone on pulmonary function (after
Kleinman et al., 1989)
20
Aims of epidemiological studies

• to determine if air pollution poses a hazard to
  human health (Causality)
• to characterise the relationship between the
  level of exposure and the response (Dose-response
  relationship)
• to examine responses in potentially susceptible
  populations (Subgroup analyses)

21
Answers the complementary policy questions

• does the pollutant pose a hazard to human health?
• at what level of exposure are the risks
  acceptable?
• which groups need special consideration because
  of susceptibility?

22
Strengths of epidemiological studies

• Real people
• Usual environment
• Exposed to typical mixtures of air pollution
  (often complex mixtures)

23
Weaknesses of epidemiological studies

• Most studies report only acute effects (cheaper
  and easier to do than studies investigating
  chronic effects which may be of greater public
  health importance)

24
Weaknesses of epidemiological studies (cont)

• Provide little information about biological
  mechanisms
• Little information about the correlations between
  ambient and personal exposures
• Difficulty in modelling highly correlated risk
  factors
• Unable to explore contribution of PM size and
  composition

25
Issues in the interpretation of epidemiological
studies

• Biases - selection and measurement
• Confounders weather, pollens
• Low concentrations of pollutants
• Multi-pollutant exposures complex mixtures
• Different air pollution metrics
• Different outcome measures
• Different statistical models

26
Health outcome measures in studies of air
pollution
27
Some other health endpoints of interest

• Cardiovascular disease (AMI, heart failure)
• ECG changes
• Hospital admissions
• Mortality
• Perinatal outcomes
• Birth weight and gestational age
• Neonatal and infant mortality
• Cancer incidence lung cancer

28
Panel studies

• Also known as diary studies
• Panel of subjects keeping daily diaries
• An example would be asthma diaries
• Exposures
• measured at the ecological level
• Outcomes
• measured at the individual level

29
Acute effects of low levels of ambient ozone on
peak expiratory flow rate in a cohort of
Australian children Bin B Jalaludina,b, Tien
Cheya,b, Brian I O'Toolec,d, Wayne T Smithe,
Anthony G Caponf and Stephen R Leederd,g IJE
200029549-557
30
Aim

• To determine associations between ambient air
  pollution and lung function in children with
  asthma

31
Methods

• Children in the study were from six primary
  schools in west and southwest Sydney
• Each selected school was closest to an EPA air
  quality monitoring station
• Daily asthma diaries
• Daily air pollution concentrations

32
Location of EPA air monitoring stationsWestern
Sydney Childrens Asthma Study 1994

33
Mean daytime ozone and evening lung function,
Western Sydney Childrens Asthma Study
34
Mean PM10 and evening lung function, Western
Sydney Childrens Asthma Study 1994-1995
35
(No Transcript)
36
Conclusions

• Significant negative association between mean
  daytime ozone and lung function
• No association between mean PM10 and lung
  function
• No association between mean NO2 and lung function

37
Asthma camp associations between O3 levels and
both peak flow lung function change and asthma
exacerbations (as indicated by ?-agonist
medication use) (Thurston et al., 1997)
38
Time Series Studies

• Late 1980s
• Advances in statistical techniques
• Advances in computing technology
• Able to detect small effects
• Outcome and exposure aggregated over 1 day
• Investigate short term/acute effects of air
  pollution only (not chronic/long term effects)

39
Time Series Studies

• Strengths
• not confounded by personal risk factors (eg
  smoking, SES, diet)
• inexpensive (use routinely collected data)
• Weaknesses
• potential confounding by other environmental
  factors (eg season, weather, epidemics)
• potential bias due to aggregated exposure (ie
  exposure missclassification)

40
Time series studies

• Commonly used study design
• Uses routinely collected exposure and outcome
  data
• Outcomes studied include mortality,
  hospitalisations, emergency department
  presentations, general practitioner visits
• Both exposure and outcome measured at the
  ecological level

41
Sydney Time Series Studies

• American Journal of Public Health, 1998
  88759-764. Air Pollution and Daily Mortality in
  Sydney, Australia, 1989 through 1993. G Morgan et
  al
• American Journal of Public Health, 1998
  881761-1766. Air Pollution and Daily Hospital
  Admissions in Sydney, Australia, 1990 to 1994. G
  Morgan et al
• Epidemiology 14(5)S111, 2003. The effects of low
  level air pollution on daily mortality and
  hospital admissions in Sydney, Australia, 1994 to
  2000. Morgan G et al
• Epidemiology 14(5)S107, 2003. Associations
  between ambient air pollution and daily emergency
  department presentations for respiratory disease
  and cardiovascular disease in the elderly (65
  years), Sydney, Australia. Jalaludin B et al

42
Sydney Mortality Time Series Study

• Aim
• This study examined the effects of outdoor air
  pollutants in Sydney, Australia, on daily
  mortality.
• Methods
• Time-series analysis on counts of daily mortality
  and major outdoor air pollutants (PM10, PM2.5,
  ozone, NO2, CO, SO2) in Sydney (1994 to 2001)
• Adjusted for seasonal and cyclical factors
• Poisson regression
• Semi parametric smoothing

43
http//biosun01.biostat.jhsph.edu/fdominic/
44
Sydney Metropolitan Study Region
45
(No Transcript)
46
Time Series Modelling Strategy
Daily Deaths
long term trend
seasonal trends
weather (temp humidity)
day of the week
public / school holidays
influenza episodes
air pollution
Statistical Model (GAM/GLM)

- Poisson distribution (counts of daily deaths)
- Control for overdispersion and autocorrelation
47
(No Transcript)
48

• Results - Mortality
• both PM10 and PM2.5 are associated with all
  cause and cardiovascular mortality, primarily in
  the cool season
• both PM10 and PM2.5 are associated with
  respiratory mortality, primarily in the warm
  season at lags greater than 6 days
• ozone is associated with all cause,
  cardiovascular and respiratory mortality,
  primarily in the warm season when concentrations
  are highest.

49

• Results - Hospital admissions
• both PM10 and PM2.5 associated with all
  cardiovascular, ischemic heard disease, and all
  respiratory admissions 65 years, primarily in
  the cool season with PM2.5 showing the stronger
  associations
• CO and NO2 are strongly associated with all
  cardiovascular admissions and ischemic heart
  disease admissions in both the cool and warm
  season
• CO and NO2 are strongly associated with all
  respiratory admissions 65 years in both the cool
  and warm season
• PM2.5 and NO2 are strongly associated with
  childhood asthma admissions 1-14 years,
  particularly in the warm season with the effects
  decreasing at higher lags.

50
Conclusions

• Current levels of air pollution in Sydney are
  associated with daily mortality, hospital
  admissions, and hospital ED attendance

51
Time Series Multi city Studies
http//biosun01.biostat.jhsph.edu/fdominic/
52
http//biosun01.biostat.jhsph.edu/fdominic/
53
http//biosun01.biostat.jhsph.edu/fdominic/
54
NMMAPS 90 US Cities City and Regional
Estimates
http//biosun01.biostat.jhsph.edu/fdominic/
55
Time Series Studies - Multi City

• Improved precision of estimate
• Different studies can use different
• Health outcomes
• Exposures
• Statistical models

56
Time Series Mortality Effect Estimates

• Meta Analysis estimates (1990s)
• 10ug/m3 increase in PM10 produces a 1 increase
  in daily mortality
• Recent Meta/Pooled Analysis estimates (2000s)
• NMMAPS study of 90 US cities
• 10ug/m3 inc in PM10 produces a 0.4 inc in daily
  mortality
• APHEA study of 6 Western European cities
• 10ug/m3 inc in PM10 produces a 0.6 inc in daily
  mortality

57
Australian Multi-City Time Series Studies

• Epidemiology 14(5)S32, 2003. The short-term
  effects of air pollution on mortality A meta
  analysis for 4 Australian cities. Simpson R et al
  
• Epidemiology 14(5)S33, 2003. The short-term
  effects of air pollution on respiratory
  admissions A meta analysis for 4 Australian
  cities. Simpson R et al
• Epidemiology 14(5)S33, 2003. The short-term
  effects of air pollution on cardiovascular
  admissions A meta analysis for 4 Australian
  cities. Simpson R et al

58
Australian Multi City Study
59
Case crossover studies

• Case control analogue of the crossover study
• For each case, one or more time periods are
  chosen as the control periods for the case
• The exposure status at the time of disease onset
  is compared with the distribution of exposure
  status for that same individual
• Assumption that neither exposures or confounders
  are changing over time in a systematic way

60
Case crossover studies (cont)

• The exposure must vary over time within
  individuals rather than stay constant
• The exposure must have a short induction time and
  a transient effect
• Each case is his/her own control and all non-time
  varying confounders (whether measured or not) are
  controlled in the analysis

61
(No Transcript)
62
(No Transcript)
63
(No Transcript)
64
Reanalysis of the Effects of Air Pollution on
Daily Mortality in Seoul, Korea A Case-Crossover
DesignJong-Tae Lee1 and Joel Schwartz2 EHP
1999107(8)633-636
65
(No Transcript)
66
(No Transcript)
67
(No Transcript)
68
Cross sectional study

• Comparing serveral (more than two) geographical
  areas
• Difficult to control for confounders

69

• Outdoor air pollution and children's
  respiratory symptoms in the steel cities of New
  South Wales
• Peter R Lewis, Michael J Hensley, John
  Wlodarczyk, Ruth C Toneguzzi, Victoria J
  Westley-Wise, Trevor Dunn and Dennis Calvert MJA
  1998 169 459-463

70
Objective

• To investigate the relationship between outdoor
  air pollution and the respiratory health of
  children aged 8 to 10 years

71
Methods

• A cross-sectional survey (between October 1993
  and December 1993) of children's health and home
  environment. Summary measures of particulate
  pollution PM10 and SO2 were estimated for each
  area (using air quality monitoring station data
  from July 1993 to June 1994)

72
Survey participants

• Parents of 3023 primary school children (Years 3,
  4 and 5) from industrial and non-industrial areas
  with air quality monitoring stations in the
  Hunter and Illawarra regions of New South Wales

73
Main outcome measures

• Reported occurrence of four or more chest colds,
  four or more attacks of wheezing, and night-time
  cough without a cold for more than two weeks, all
  within the previous 12 months

74
Study areas
75
Results

• 77 response rate (range 66 to 88)
• The average annual outdoor air pollution for the
  nine areas was 18.6-43.7 µg/m for PM10 and
  0.16-0.90 pphm for SO2
• No significant association with SO2
• Significant association per 10 µg/m increase in
  PM10 for chest colds (OR1.43) and night-time
  cough (OR1.34), but not wheeze

76
Symptoms and PM10
77
Conclusions

• Evidence of health effects at lower than expected
  levels of outdoor air pollution PM10 in the
  Australian setting

78
Cohort studies

• Few such studies
• None from Australia
• Two study designs
• Initiation of new cohorts specifically to study
  air pollution effects
• Utilising existing cohorts

79
Southern Californian Childrens Health Study
(Gauderman 2002)

• To study chronic effects of air pollution
• 10-yr study that commenced in 1992
• 5,500 children in 12 communities
• Annual questionnaire and lung function measures
• Air quality measured outdoors, in schools and in
  homes

80
(No Transcript)
81
(No Transcript)
82
Initial findings

• A clear correlation between lower lung function
  and more intense air pollution. The effect was
  more pronounced in girls who reported spending
  the most time outdoors.
• High levels of NO2, PM10 and PM2.5, and acid
  vapour appeared to be associated with slower lung
  growth.
• Lung capacity was lower for girls living in the
  most polluted communities, especially those with
  the highest levels of NO2 and PM.
• Wheezing was more evident in boys who were
  exposed to higher levels of NO2 and acid vapour.
  
• Ozone did not appear to affect lung function
  growth.
• Girls with asthma who lived in communities with
  high ozone levels had lower lung capacity.
• Boys appear to have lower breathing capacity if
  they lived in communities with high ozone levels,
  provided they also spent above average time
  outdoors (where ozone concentrations are higher).

83

• Box 2 Summary of the major findings from the
  Southern California Childrens Health Study
• Associations were found between
• Wheeze and NO2 (Males only)
• Decreased MMEF, PEFR and O3 (Females only)
• Decreased FVC, FEV1, MMEF, and PM10 and NO2
  (Females only)
• Lung function growth and PM10, PM2.5, and NO2
  (fourth grade cohort)
• Lung function and acid vapour, and NO2
• Increased risk of developing asthma and O3 (in
  children who played three or more team sports)
• School absences (respiratory related) and O3.

84
Lung Cancer, Cardiopulmonary Mortality, and
Long-term Exposure to Fine Particulate Air
Pollution C. Arden Pope III, PhD Richard T.
Burnett, PhD Michael J. Thun, MD Eugenia E.
Calle, PhD Daniel Krewski, PhD Kazuhiko Ito,
PhD George D. Thurston, ScD JAMA. 20022871132
-1141.
85
Aims

• To assess the relationship between long-term
  exposure to fine particles and and all-cause,
  lung cancer and cardiopulmonary mortality

86
Methods

• Part of Cancer Prevention II Study
• Enrolled 1.2 million adults in 1982
• Baseline questionnaire collected individual
  information (age, sex, weight, height, smoking
  history, etc)
• 500,000 subjects risk factors linked to air
  pollution data and deaths through to 1998
• Study population drawn from from 157 cities
  throughout the USA(ie 157 data points)

87
(No Transcript)
88
(No Transcript)
89
Conclusions

• PM2.5 associated with all cause, lung cancer and
  cardiopulmonary mortality
• PM10 and TSP not consistently associated with
  mortality
• Long term exposure to combustion related PM2.5 is
  an important environmental risk factor for
  cardiopulmonary and lung cancer mortality

90
Epidemiological (observational) studies

• Study designs not commonly used
• Combination of designs in one study
• Exposure assessment (often weakest link)
• Coherence (Bates 1992)
• Causality (eg Hill 1965)
• Levels of evidence
• Drive criteria air pollutant standard setting

91
Concept of coherence

• ?symptoms
• ?restricted activity days
• ?GP visits
• ?ED presentations
• ?hospital admissions
• ?deaths

92
Criteria for assessing causality of associations
Data from Hill (1965) and Rothman (1986).
93
Thank you
94
http//biosun01.biostat.jhsph.edu/fdominic/ISRpap
er0920.pdf