Stephen Jun Villejo Paolo Redondo Angela Nalica Erniel Barrios School of Statistics UP Diliman Climate change and dengue According to the World Health Organization climate change affects occurrence of infectious diseases apart from rapid demographic environment ID: 754630
Download Presentation The PPT/PDF document "Weather and incidence of dengue in the P..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Weather and incidence of dengue in the Philippines: evidence of climate change
Stephen Jun Villejo | Paolo Redondo
Angela
Nalica
|
Erniel
Barrios
School of Statistics, UP
DilimanSlide2
Climate change and dengue
According to the World Health Organization, climate change affects occurrence of infectious diseases, apart from rapid demographic, environmental, social, technological and other changes.Slide3
Approaches in predictive modelling
Statistical Models
Derives an empirical relationship between climatic conditions and the actual distribution of the disease
Process-based (Mathematical Models)
use equations that express the scientifically documented relationship between climatic variables and biological parameters
Landscape-based Models
combining the climate-based models described above with the rapidly-developing use of spatial analytical methods, to study the effects of both climatic and other environmental factorsSlide4
There is a need to understand complex causal relationships and apply this information to the prediction of future impacts, using more complex, better validated, integrated models. Slide5
Dengue
a fast emerging pandemic viral disease in many parts of the world.
a mosquito-borne viral infection causing a severe flu-like illness and, sometimes causing a potentially lethal complication called severe dengue.
described by Murray et al (2013) as an “acute mosquito-borne viral infection that places a significant socioeconomic and disease burden on many tropical and subtropical regions of the world.” Slide6
incidence of dengue has increased 30-fold over the last 50 years. Up to 50-100 million infections are now estimated to occur annually in over 100 endemic countries, putting almost half of the world’s population at risk.
The World Health Organization says…Slide7
Case for the Philippines:
the National Epidemiology Center of the Philippines’ Department of Health reported a total of 59,943 dengue cases from January 1 to September 6, 2014.
Based on the Disease Surveillance Report by the Department of Health, a total of 101, 401 suspect dengue cases were reported nationwide from January 1 to August 20, 2016, which is 16% higher compared to the same time period of the previous year.
Of all the cases, 11.1% are from Region VI and 10.7% are from Region IV-A. Slide8
Incidence of dengue in Selected provinces
There is high incidence of dengue in the third quarter of years 2010 to 2013.
Incidence of dengue seems to be high during the third quarter of every year. Slide9
Incidence of dengue in Selected provinces
There is high incidence of dengue in the third quarter of years 2010 to 2013.
Incidence of dengue seems to be high during the third quarter of every year. Slide10
Empirical evidence
There is much evidence of associations between climatic conditions and infectious diseases.
Naish
etal
(2014) have found that the transmission of dengue is highly sensitive to climatic conditions, particularly temperature, rainfall, and relative humidity.
Earnest et al (2012) used the approaches of
poisson
regression and sinusoidal function in modelling dengue incidence. They found out that temperature, relative humidity and SOI are associated with dengue cases.
Colon-Gonzales
etal
(2011) found that incidence of dengue was higher during El Nino. Temperature was also an important factor in the incidence of dengue.
Chen et al (2012) also arrived at the same conclusion, that a change in climate influences dengue outbreaks. Lag effects were also observed.Slide11
Empirical evidence
Hales
etal
(2002) used logistic regression and IPCC scenarios to conclude a potential increase in the dengue risk areas under climate change scenarios, holding all risk factors constant.
Bulto
etal
(2006) using multivariate methods found a strong association between climate anomalies and dengue.
Su (2008) investigated the effect of temperature and rainfall as weather parameters on dengue incidence in Metro Manila from 1996 to 2005. Rainfall was found to be a significant predictor of dengue incidence, but there’s not significant relationship between dengue incidence and temperature. Slide12
Superimposed Time plots of incidence of dengue and weather variables for
albaySlide13
Superimposed Time plots of incidence of dengue and weather variables for BoholSlide14
Superimposed Time plots of incidence of dengue and weather variables for
quezon
citySlide15
Superimposed Time plots of incidence of dengue and weather variables for
quezon
citySlide16
Research objectivesSlide17
Research objectives
A dynamic spatiotemporal model is proposed and estimated through a hybrid of maximum likelihood, forward search, and
boostrap
in the context of the
backfitting
algorithm. The model is then used in understanding space-time dynamics of prevalence rate of some diseases in various provinces of the Philippines.
Climate change, viewed as a structural change, is tested through a nonparametric bootstrap-based approach.Slide18
Proposed modelSlide19
Proposed model
The model is a modification of the
Landagan
and Barrios (2007) and Villejo
etal
(2016) model.
is the count of dengue in the
th
station for the
th
time point
are the set of covariates whose effects are assumed to be different across locations
are the set of covariates whose effects are assumed to be varying through time
is the effect of the covariate
for the
th
station
is the effect of
for the
th
time point.
Slide20
Proposed model
The error term is assumed to follow an autoregressive process of order 1, given by
<1
The model postulates that there are variables whose effects are isolated for a fixed station or whose effect will vary for different stations, captured by
; and there are variables whose effects are constant for all stations for a fixed time point, captured by
.
Slide21
Proposed estimation procedureSlide22
Step 1: Cochranne-orcutt
procedure
For a fixed
, estimate the model
using regression with
autocorrelated
errors via the
Cochranne-Orcutt
Procedure using Maximum Likelihood Estimation.
Store the estimates and obtain the residuals,
. The residuals will contain information explained by
Slide23
Step 2: forward search
Fix
and perform the following:
Fit
Obtain the residuals.
A subset of size
, corresponding to the stations with the smallest residuals will be chosen. The
n
observations are ideal and outlier-free.
Fit
using the
observations from (2).
Using the fitted model in (3), compute the fitted response on the
left-out observations and obtain the residuals.
The observation with the smallest residual from (4) will be included in the subset of observations from (2).
Fit the model
using the
observations from (5).
Iterate from (4), adding one observation at a time until the model is behaving wildly based on the Cook’s D or until all
locations have been included in the model.
Slide24
Step 3: robust estimation of
Suppose
is the number of stations in the final subset. For a fix
:
Draw a simple random sample of size
with replacement from the subset.
Fit the model
.
Repeat (1) and (2) B times, yielding B
.
If we denote by
the
th
bootstrap estimate of
the
montecarlo
estimates of the mean and standard deviation
are given, respectively, by
, S.E.(
.
Slide25
Step 4: updating
A new dependent variable will be computed as
.
The algorithm then iterates back to
Step 1
using
as the new dependent variable. In updating the dependent variable, the original values of
will be used.
The iteration converges when there are minimal changes in the values of the parameter estimates.
Slide26
Testing for Significance and Evidence of Structural ChangeSlide27
A non-parametric bootstrap-based approach will be used to test for structural change:
After obtaining the final parameter estimates via the proposed estimation procedure, compute the MSE.
Recreate the data by performing the following:
Generate random numbers from
. These random numbers are the innovations
.
Using the final parameter estimates obtained from the proposed estimation procedure, compute for new values of the response variable using the structural form of the postulated model.
Apply the estimation procedure using the recreated dataset from (b).
Repeat (2) to (3) B times and store all the parameter estimates
To obtain the 95% bootstrap confidence intervals for a particular parameter, compute for the 2.5
th
percentile and 97.5
th
percentile of the B parameter estimates.’
Slide28
data and variablesSlide29
Location:
16 PAGASA stations nationwide
Time:
Monthly data on incidence of dengue from December 2007 to December 2013 for 16 provinces.
The provinces or station locations considered are the
ff
:
Data
Ilocos Norte
Albay
Cagayan
Leyte
Isabela
Bohol
Pangasinan
Surigao del Norte
Benguet
Cebu
Nueva
Ecija
Surigao
del Sur
Manila
Zamboanga del Sur
Quezon City
South
CotabatoSlide30
Response:
Incidence of Dengue (count)
Covariates:
Pressure (
HPa
)
Temperature (C)
Temperature before
vaporation
(C)
Precipitaiton
(mm)
Maximum Temperature (C)
Minimum Temperature (C)
Maximum Precipitation (mm)
Maximum Wind (km/h)
Southern Oscillation Index
variablesSlide31
Results and discussionSlide32
<1
and
Final modelSlide33
Parameter estimatesSlide34
Parameter estimates of min temperature, precipitation and
soi
Station
MinTemp
Precipitation
SOI
rho
Ilocos Norte
-1.39
1.9136
-1.0425
0.0435
Cagayan
2.038
-0.1321
-1.2384
0.1302
Isabela
7.544
1.7649
-1.0944
0.0197
Pangasinan
-0.056
0.2677
-0.7384
-0.0441
Benguet
0.119
0.1084
-0.4792
-0.1446
Nueva Ecija
1.626
-0.1277
-0.5426
-0.4553
Manila
6.182
0.4746
-1.4144
0.1276
Quezon City
14.748
-1.2394
-0.6249
0.2069
Albay
4.769
-0.0412
-0.8126
-0.0053
Leyte
4.194
-0.9108
-0.8536
-0.2496
Bohol
0.368
-0.685
-0.6596
-0.1578
Surigao del Norte
0.861
-0.6817
-0.5603
-0.1390
Cebu
-0.807
-1.401
-0.6896
-0.0115
Surigao del Sur
2.434
0.0581
-0.8218
-0.1482
Zamboanga del Sur
1.139
0.2983
-0.7798
-0.3687
South Cotabato
2.711
-0.4655-0.8582-0.4939
The effect of the Southern Oscillation Index is consistently negative for all stations. Moreover, the magnitude of the estimates does not vary much from each other.
For minimum temperature, the effect highly varies across station. Majority of the estimates are positive, although for 3 stations, the estimates are negative. Slide35
Parameter estimates of temperature
The effect of temperature is consistently positive for all time points. The magnitude of the estimates varies across time which supports the postulated dynamic spatial-temporal model. Slide36
Parameter estimates of temperature before
vaporation
The estimates vary highly in sign and magnitude.Slide37
Parameter estimates of max precipitation and max wind
Majority of the estimates of the effect of maximum precipitation and maximum wind are negative. Slide38
Significance of estimatesSlide39
95% Bootstrap confidence intervals
MinTemp
Precipitation
SOI
rho
Station
LB
UB
LB
UB
LB
UB
LB
UB
Ilocos Norte
-20.03%
23.38%
0.98%
7.50%
-4.58%
0.50%
-6.86%
26.99%
Cagayan
-4.00%
17.59%
-1.62%
3.17%
-3.54%
1.02%
-33.18%
-4.67%
Isabela
5.49%
31.53%
-0.42%
4.56%
-4.49%
-0.35%
-21.76%
5.52%
Pangasinan
-8.83%
10.71%
-0.91%
1.32%
-4.11%
0.10%
-4.60%
23.93%
Benguet
-12.49%
10.79%
-0.25%
2.52%
-3.35%
1.70%
-0.21%
26.37%
Nueva Ecija
0.49%
8.65%
-2.07%
2.39%
-3.01%
0.70%
40.43%
63.93%
Manila
-16.56%
29.47%
-3.43%
2.84%
-5.50%
-0.27%
-31.98%
0.95%
Quezon City
6.72%
34.31%
-4.21%
-0.67%
-3.36%
-0.49%
-24.49%
5.22%
Albay
-4.76%
16.99%
-2.04%
3.79%
-3.47%
-0.21%
0.21%
23.66%
Leyte
2.82%
37.78%
-6.69%2.84%-3.61%-0.86%3.09%37.66%Bohol-3.84%15.41%-3.03%5.32%-3.27%-1.68%3.33%43.88%Surigao del Norte-26.04%18.90%-4.20%2.26%-3.75%0.33%0.81%31.41%Cebu-32.48%12.24%-6.99%3.87%-3.00%-0.63%-0.88%13.57%Surigao del Sur-22.97%29.04%-9.81%11.82%-4.56%-0.79%15.92%30.08%Zamboanga del Sur-2.85%8.07%-0.86%3.09%-3.44%2.24%21.66%52.14%South Cotabato-19.30%25.49%-4.33%2.06%-3.36%-0.93%43.61%57.83%
SOI and the lag(1) effects are significant for many stations.
Only very few stations have signification effect for minimum temperature and precipitation. Slide40
95% Bootstrap confidence intervals of temperatureSlide41
95% Bootstrap confidence intervals of temperature before
vaporationSlide42
95% Bootstrap confidence intervals of maximum precipitationSlide43
95% Bootstrap confidence intervals of maximum WindSlide44
The Mean Absolute Deviation given by
is equal to 190 with a standard deviation of 200.
Predictive abilitySlide45
On the Robustness of the Parameter EstimatesSlide46
It can be seen that the parameter estimates appear to be stable and controlled despite the erratic movement of the SOI.
The stability of the estimates even during episodes with disturbances in the system is due to the fact that the estimation is done using robust methodsSlide47
Test for structural change
If the parameter estimate is not contained in the robust confidence intervals, then it’s a possible evidence of structural change.Slide48
STATION
RHO_HAT
MINTEMP
PRECIPITATION
SOI
Ilocos Norte
contained
contained
contained
contained
Cagayan
not contained
contained
contained
contained
Isabela
contained
contained
contained
contained
Pangasinan
contained
contained
contained
contained
Benguet
not contained
contained
contained
contained
Nueva Ecija
not contained
contained
contained
contained
Manila
not contained
contained
contained
contained
Quezon City
not contained
contained
contained
contained
Albay
not contained
contained
contained
contained
Leyte
not contained
contained
contained
not contained
Bohol
not contained
contained
contained
not contained
Surigao del Norte
not contained
contained
contained
contained
Cebu
not contained
contained
contained
contained
Surigao del Sur
not contained
contained
contained
contained
Zamboanga del Sur
not contained
contained
contained
contained
South Cotabato
not contained
contained
contained
not contained
All estimates for minimum temperature and precipitation, and majority of the estimates for SOI are contained in the Cis
Many of the estimates of the lag effects are not contained in the CI. Slide49
Many of the estimates are not contained in 95% CI.
The estimates in the years 2012 and 2013 are very wide. These are the same time periods where the incidence of dengue blew up. Slide50
The estimates in the years 2012 and 2013 are very wide. These are the same time periods where the incidence of dengue was very high.Slide51Slide52Slide53
conclusionsSlide54
A dynamic model is appropriate in modelling incidence of dengue using weather variables.
Forward search and bootstrap methods contributed in the stability of the estimates even during high and low episodes of SOI.
For the parameter estimates estimated last in the
backfitting
, many of the estimates are not contained in the bootstrap confidence intervals. This is taken as an evidence for structural change (climate change).
ConclusionsSlide55
Future studiesSlide56
Include higher lags, incorporate seasonality
Consider a dynamic
poisson
autoregression
spatio
-temporal model since the response is a count variable
Include more years and more stations in the study
Consider regime-switching models
Future studiesSlide57
Thank you!