Derek A.T. Cummings, University of Pittsburgh Graduate School of Public Health and Johns Hopkins Bloomberg School of Public Health

1
Spatial coherence and association of temperature,
rainfall and the incidence of dengue hemorrhagic
fever in Thailand
Derek A.T. Cummings, University of Pittsburgh
Graduate School of Public Health and Johns
Hopkins Bloomberg School of Public Health Timothy
P. Endy, Walter Reed Army Institute of
ResearchAnanda Nisalak, Armed Forces Research
Institute of Medical SciencesDonald Burke,
University of Pittsburgh Graduate School of
Public Health
2
Research Question

• How is climate variation associated with the
  incidence of dengue hemorrhagic fever?
• Does the dengue season occur at the same time as
  the rainy season or hot season across Thailand?

3
Effects of Temperature and Rainfall on Dengue
Incidence
Temperature Affects -Extrinsic incubation period
of virus -Survival rates of mosquito
vector -Biting rates of mosquito -Fecundity of
mosquito vector Rainfall Affects -Survival
rates of mosquito vector -Fecundity of mosquito
vector
4
Watts et al, 1987
DEG C
34
O
X
X
X
X
O
O
X
X
X
32
O
O
X
X
X
30
O
O
O
X
X
28
26
O
O
O
O
O
24
14
3
7
21
28
X TRANSMISSION, O NONE
DAYS INCUBATION
5
Source of Incidence Data
Monthly DHF Incidence Passive Surveillance 72
Provinces 1983-1996 Gathered by Thai Ministry of
Public Health, serologically confirmed to
the extent logistically possible. (20)
Roughly 850,000 cases
6
Normalized Log of Raw Incidence of DHF in Thailand
N S
Log(Incidence/1000)
Rank of Province
Year
7
Source of Weather Data
Monthly Mean Temperatures and Monthly Rainfall
for 52 weather stations in 41 provinces 1983-1996
NOAA Climate Anomaly Monitoring System
8
First Hypothesis
The timing of the DHF season is associated with
the timing of the rainy or warm seasons
9
Spatial Synchrony of DHF, Temperature and Rainfall
DHF Incidence
Rainfall
Temperature
10
Normalized Log of Raw Incidence of DHF in Thailand
N S
Log(Incidence/1000)
Rank of Province
Year
11
Decomposition of Time Series Using the Empiric
Mode Decomposition
12
Seasonal Mode of DHF Incidence
13
Phase Differences
14
(No Transcript)
15
(No Transcript)
16
Mean Phase Difference of Seasonal Mode of DHF
Incidence vs. Mean Phase Differences of Rainfall
plt0.003 R20.2
17
(No Transcript)
18
Temperature

• The timing of changes in ambient temperature are
  not associated with the timing of the dengue
  season in the same analysis

19
Second Hypothesis
Annual DHF incidence is related to annual means
of temperature or annual totals of
rainfall Used a two-stage hierarchical model to
generate pooled estimates of the association
between Annual mean temperature, annual total
rainfall and total annual DHF incidence across
the entire country
20
Used a two-stage hierarchical model to generate
pooled estimates of the association between
annual mean temperature, annual total rainfall
and total annual DHF incidence across the entire
country Higher Mean Annual Temperatures are
statistically significantly associated with
higher Annual DHF Incidences (plt10-4) Modified
by Mean Temperatures (plt0.04) and Distance from
Bangkok (plt10-5)
21
Betas from Generalized Linear Model of Annual
Incidence on Temperature vs. Distance from
Bangkok
Beta Coefficients from Linear Regression of
Annual DHF Inc on Temp.
Annual Mean Temperatures
22
Difficulties with this Analysis
Only monthly data is available. May not be able
to detect association between small changes in
the timing of seasons Association may be
non-linear, with thresholds above or below which,
variation is less important Seasonal process may
not be separable from underlying intrinsic
dynamics
23
Conclusions
Higher mean annual temperatures are associated
with higher incidences of DHF across Thailand
and are modified by mean temperature and
distance from Bangkok The phase of the dengue
seasonal variance in DHF incidence is associated
with the phase of rainfall across Thailand
24
Average proportion of countrywide DHF cases per
month
25
Relative timing of rainy season and the dengue
season regional
26
Next Steps
Examine weekly data available from later
years Examine correlation structure of each of
these variables
27
Thank you