SLIDE 1
Dengue Disease Outbreak Detection
Sep 2, 2014
Pankaj Dayama Kameshwaran S IBM Research - INDIA
MIE 2014
Outline
- Background
- Dengue Transmission Dynamics
- Proposed Approach
- Results
- Future Directions
MIE 2014
Dengue Disease Outbreak Detection Pankaj Dayama Kameshwaran S IBM - - PDF document
Dengue Disease Outbreak Detection Pankaj Dayama Kameshwaran S IBM - - PDF document
Sep 2, 2014 Dengue Disease Outbreak Detection Pankaj Dayama Kameshwaran S IBM Research - INDIA MIE 2014 Outline Background Dengue Transmission Dynamics Proposed Approach Results Future Directions MIE 2014 Dengue Dengue
SLIDE 2
SLIDE 3
Dengue
- Dengue is a major international public health concern – endemic in more 100 countries in
Asia, Latin America, Africa, and Western Pacific.
- Around 40% of the world population is at risk from Dengue: 50 – 100 million cases of
infections annually
- Dengue has annual seasonality and the epidemics follow a quasi-periodicity of 4-7 years for
large epidemic outbreaks
- There is NO treatment and NO vaccination
MIE 2014
SLIDE 4
Transmission Dynamics
Intrinsic Incubation 4-7 Days Recovery 3-10 Days Latent Infectious 5 Days
Onset&of Symptoms
Extrinsic Incubation 8-12 Days Infectious 2 – 4 Weeks
Infec/ous&Mosquito&Bite Infec/ous Mosquito&Bite Suscep/ble&Mosquito&Bite
Intrinsic Incubation 4-7 Days Recovery 3-10 Days Latent Infectious 5 Days
Symptoms Stop Suscep/ble&Mosquito&Bite Pa/ent&2
Dengue&is&a&mosquito&borne&infec2ous&disease
Pa/ent&1
SLIDE 5
Transmission Dynamics
- Four similar but distinct serotypes of Dengue
viruses
- Immunity and Cross-immunity:
– Life-long immunity from serotype of primary infection (usually DF) – Partial cross-immunity to other three serotypes – Secondary infection results in DHF and DSS (high mortality rate)
Prevention & Control
- Source reduction
– Killing of mosquitoes or larvae or eggs
- Isolating infected humans
HOST
- Seroprevalence
- Cross immunity
VECTOR
- Aedes burden
- Biting rate
VIRUS
- Serotype
- Multiplication
- Shedding
- Weather influences vector and viral
characteristics (short term influences)
- Host features and serotype change over a
long period of time
- Vector and viral characteristics cannot be
directly observed (Hidden)
- Host characteristics can only be statistically
inferred with sampling (Incomplete)
Week 1 Week 2 Week 3
Egg hatches
- n
contact with water Larvae Pupa Mosquito 1007Eggs 1007Eggs 1007Eggs 1007Eggs
Aedes%Life%Cycle
MIE 2014
SLIDE 6
Periodicity of Outbreaks and Serotype Predominance
- Co-circulation of multiple virus serotypes in Singapore - hyperendemicity
- Dengue characterized by cycle of 3 - 4 years, increased incidence within a cycle
followed by lull of 1-2 years.
- Quasi-periodicity attributed to intrinsic epidemic dynamics and not on climatic factors
MIE 2014
SLIDE 7
Dengue Incidence – Yearly
MIE 2014
SLIDE 8
Existing Approach
MIE 2014
Source: MOH (Singapore). Weekly Infectious Diseases Bulletin. Singapore: Ministry of Health.
- Compute moving average and standard deviation of DI (past 5
years)
- Warning Threshold = µ + 1σ
- Epidemic Threshold = µ + 2σ
Shortcomings:
- Seasonal Fluctuation and correlation in DI series ignored
- Dengue cases reporting delays ignored
SLIDE 9
Proposed Approach
MIE 2014
Percen&le(Based Threshold Generalized Shewart(Chart Robust(Control Charts
Seasonality captured by fitting weekly distributions Non-parametric, weekly threshold set to M percentile for DI based on historical data Time Series filtering procedure (SARIMA) + Statistical surveillance method (EWMA)
Majority (YES) Unanimity (NO) At least One (YES)
Aggregation Methods:
SLIDE 10
Results
MIE 2014
Outbreak Prediction: Test Period (2005-2006) Outbreak Prediction: Test Period (2011-2013)
SLIDE 11
Current Work: Alternate Perspective What is the current level of epidemic severity?
Traditional
- 2 Levels
Hidden Markov Models
- User defined Multiple Levels
- Mean and SD of Incidence for
each Level
SLIDE 12
Using Hidden Markov Models
- Severity levels for 2014
- HMM is trained on weekly incidence from 2000 - 2013
SLIDE 13
Thanks
MIE 2014