TIME SERIES METHODS IN CLIMATE CHANGE RESEARCH

1
TIME SERIES METHODS IN CLIMATE CHANGE RESEARCH

• Paul Wilkinson
• Public Environmental Health Research Unit
• London School of Hygiene Tropical Medicine
• Wednesday 1st October 2003

2
OUTLINE

• principles
• methods
• an example

3
PRINCIPLES

• Short-term temporal association between health
  events and pollution
• Usually entails analysis of day to day
  fluctuations over several years
• Similar in principle to any regression analysis
  but with some specific features

4
150
125
100
75
Cardiovascular deaths/day
50
25
0
01jan1990
01jan1991
01jan1992
01jan1993
01jan1994
CVD deaths
Mean temperature
5
PRINCIPLES 2

• Same population is compared with itself focus
  is on cause of day to day variation
• Hence methodologically strong
• Confounders are time-varying risk factors
• Designed to quantify short-term associations
  (specifically not the long-term)
• Questions arise in relation to public health
  importance mortality displacement

6
STATISTICAL METHODS 1

• Confounders influenza day of the week, public
  holidays air pollution humidity trend season
• Modelling season/trend by smoothed functions of
  date moving averages indicator variables for
  month trigonometric terms smoothing splines

7
STATISTICAL METHODS 2

• Exposure-response functions smooth
  graphs thresholds linear functions (hockey
  stick)
• Lags simple lags distributed lags different
  lag structures for heat cold
• Temporal auto-correlation affects standard
  errors insignificant if effect of important risk
  factors captured

8
125
100
Cardiovascular deaths/day
75
50
25
0
5
10
15
20
25
Mean temperature /ºC
9
150
125
100
75
Cardiovascular deaths/day
50
25
0
01jan1990
01jan1991
01jan1992
01jan1993
01jan1994
CVD deaths
Mean temperature
10
50
25
Residual deaths/day
0
-25
-50
01jan1990
01jan1991
01jan1992
01jan1993
01jan1994
11
Source Anderson HR, et al. Air pollution and
daily mortality in London 1987-92. Br Med J
1996 312665-9
12
FUNCTIONS LAGS Heat 1 to 2 days Cold
approx. 2 weeks
13
RISK ESTIMATES
Delhi RR, heat 1.0394 (95 CI 1.029 to
1.0508) i.e. 3.94 (2.80 to 5.08) percent increase
in mortality for each degree Celsius increase in
temperature above heat threshold (28
Celsius) RR, cold 1.0278 (95 CI 1.0066 to
1.0494) i.e. 2.78 (0.66 to 4.94) percent
increase in mortality for each degree Celsius
decrease in temperature below cold threshold
(19 Celsius)
14
MORTALITY DISPLACEMENT
A
MORTALITY
B
POLLUTION
15
(No Transcript)
16
VULNERABILITY

• Intrinsic
• Age
• Pre-existing disease
• Other

17
AN EXAMPLE
18
Heat-related mortality, Delhi
Relative mortality ( of daily average)
Daily mean temperature /degrees Celsius
19
Future prediction
Temperature distribution
Relative mortality ( of daily average)
Daily mean temperature /degrees Celsius
20
RISK ASSESSMENT FOR CLIMATE CHANGE
GHG emissions scenarios Defined by IPCC
GCM model Generates series of maps of
predicted future distribution of climate variables
Health impact model Generates comparative
estimates of the regional impact of each climate
scenario on specific health outcomes
Conversion to GBD currency to allow summation
of the effects of different health impacts
21
Mortality ( of annual average)
Mean daily temperature in degrees Celsius
22
CONCLUSIONS

• Time-series studies are methodologically strong
• They provide robust quantification of relative
  risks which are often very small
• But, by design, they characterize only short-term
  associations
• They entail uncertainty regarding public health
  importance because of mortality displacement
• And, crucially, their evidence is only an
  indirect indication of health impacts through
  climate change

23
(No Transcript)