Health Applications of AIRS Data: Environmental Connection and Prediction
of Influenza and Dengue Fever
Heidar'Thor'Thrastarson1,2'1.'Joint'Ins4tute'For'Regional'Earth'System'Science'and'Engineering,''
University'of'California'Los'Angeles'
Joao'Teixeira2'2.'NASA'Jet'Propulsion'Laboratory,'California'Ins4tute'of'Technology'
Emily'Serman3,'Anna'Zeng4,'Anish'Parekh3,'Enoch'Yeo5'3.'University'of'Southern'California,'4.'Stanford'University,'5.'Harvey'Mudd'College'
''
NASA'Sounder'Science'Team'Mee4ng,'October,'2017'
Influenza Background • Seasonal'influenza'epidemics'are'a'major'public'health'concern'
– Millions'of'cases'of'severe'illness'worldwide'each'year'– 250,000'to'500,000'deaths'worldwide'each'year'– Large'economic'toll'
• In'temperate'regions'influenza'incidence'generally'has'pronounced'peaks'''''''''''''''''in'the'winter.'
– But'specific'4ming,'magnitude'and'dura4on'of'individual'local'outbreaks'''''''''''''''''''''''''''''''''''''''''''in'any'given'year'are'variable'and'not'well'explained''
• If'the'4ming'and'intensity'of'seasonal'influenza'outbreaks'can'be'forecast,'''''''''''''this'would'be'of'great'value'for'public'health'response'efforts.''
– Could'guide'both'mi4ga4on'and'response'efforts'– Planning'and'stockpiling'of'vaccines'and'drugs'– Management'of'hospital'resources'– Focusing'of'efforts'to'areas'with'more'urgent'need'
©'2017.'All'rights'reserved.'
Humidity & Influenza • Recent'studies'have'highlighted'a'role'of'absolute'(or'specific)'humidity'condi4ons'
as'a'leading'explana4on'for'the'seasonal'behavior'of'influenza'outbreaks'^'Decreasing'absolute'humidity'associated'with'increased'influenza'ac4vity'
• Lab'experiments:'^'Absolute'humidity'strongly'modulates'the'airborne'survival'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''and'transmission'of'the'influenza'virus.''
• Climate'&'influenza'data'records:'^'Increased'winter4me'influenza^related'mortality'in'the'US'''''''''''''''''''''''''''''''''''''''''''''''''''''''''associated'with'anomalously'low'absolute'humidity'levels'
• Humidity^driven'epidemiological'models'yielding'promising'results''• The'reason'for'the'humidity–influenza'rela4onship'is'not'well'established''''''''''''''
but'a'few'mechanisms'are'proposed'^ Drying'of'mucous'membranes'^ Humidity'effects'on'droplet'sizes'and'travel'range'^ Increased'survival'4mes'for'the'virus''
Humidity & Influenza
-0.00025
-0.0002
-0.00015
-0.0001
-0.00005
0
0.00005
0.0001
0.00015
-40 -37 -34 -31 -28 -25 -22 -19 -16 -13 -10 -7 -4 -1 2 5 8 11 14 17 20
H2O
MM
R A
nom
aly
(kg/
kg)
Days from Onset
GFT Plus State Level: Days from Onset vs Humidity Anomaly (Smoothed)
Onset 150 Onset 200 Onset 250 Onset 300 -1
-0.8 -0.6 -0.4 -0.2
0 0.2 0.4 0.6 0.8
1 1.2
-40
-37
-34
-31
-28
-25
-22
-19
-16
-13
-10 -7
-4
-1 2 5 8 11
14
17
20
Tem
p A
nom
aly
(K)
Days from Onset
State Level: Onset vs Temperature Anomaly (Smoothed)
Onset 150 Onset 200 Onset 250 Onset 300
Credit:'Emily'Serman'
• Shaman'et'al.'(2010)'used'30'years'of'flu^related'mortality'data'and'humidity'data'
• Anomalously'low'humidity'preceding'the'onset'of'influenza'seasons'
• We'have'used'AIRS'and'Google'Flu'Trends'data'at'city,'state'and'regional'scales'in'the'US'to'give'further'support'for'the'role'of'humidity'(and'temperature)'in'driving'influenza'seasonality'
Model of Influenza Outbreaks
• We'have'developed'and'implemented'a'numerical'predic4on'system'that'is'driven'by'specific'humidity'to'predict'influenza'outbreaks.'
• Standard'compartmental'epidemiological'model,'SIRS'(Suscep4ble^Infec4ous^Recovered^Suscep4ble)'type'
• Two'coupled'first^order'Ordinary'Differen4al'Equa4ons'numerically'solved'for'the'number'of'suscep4ble'and'infected/infec4ous'people'in'a'given'popula4on.'
• Rate'of'infec4ons'and'recoveries'parameterized'in'terms'of'average'length'of'immunity'and'mean'infec4ous'period'and'assumed'to'have'a'simple'dependence'on'input'specific'humidity.''
SIRS Model
SIRS-q: Humidity-Driven Influenza Model
Heidar Thor Thrastarson
August 3, 2015
1 SIRS Model
Model of influenza transmission dynamics. SIRS (Susceptible-Infectious-Recovered-Susceptible) typemodel. Similar to Shaman & Karspeck (2012). Two coupled first-order ODEs for S and I:
dS
dt=
N � I � S
L� �IS
N� ↵ (1)
dI
dt=
�IS
N� I
D+ ↵ (2)
where
� =R
0
D=
1
D[R
0min
+ (R0max
�R0min
)eaq] (3)
Some assumptions/limitations:
- Fractional persons
- Only a single influenza strain
- Partial and cross-immunities not represented
- Perfectly mixed - clustering and structured interactions of sub-populations not included
- Variations in host infectiousness not modeled
- Other potential drivers than humidity not included, e.g., school calendar
- No age stratification
Table 1: Variables/Parameters
N Population sizeS Susceptible personsI Infectious (= infected) personsL Average immunity duration [days]D Mean infectious period [days]↵ Rate of (travel-related) import of virus into model domain [(infections)/day]� Contact rate [(contacts)/day]R
0
(Daily) basic reproductive numbera (Negative) coe�cient in contact rate exponentialq Specific humidity [kg/kg]
1
SIRS-q: Humidity-Driven Influenza Model
Heidar Thor Thrastarson
August 3, 2015
1 SIRS Model
Model of influenza transmission dynamics. SIRS (Susceptible-Infectious-Recovered-Susceptible) typemodel. Similar to Shaman & Karspeck (2012). Two coupled first-order ODEs for S and I:
dS
dt=
N � I � S
L� �IS
N� ↵ (1)
dI
dt=
�IS
N� I
D+ ↵ (2)
where
� =R
0
D=
1
D[R
0min
+ (R0max
�R0min
)eaq] (3)
Some assumptions/limitations:
- Fractional persons
- Only a single influenza strain
- Partial and cross-immunities not represented
- Perfectly mixed - clustering and structured interactions of sub-populations not included
- Variations in host infectiousness not modeled
- Other potential drivers than humidity not included, e.g., school calendar
- No age stratification
Table 1: Variables/Parameters
N Population sizeS Susceptible personsI Infectious (= infected) personsL Average immunity duration [days]D Mean infectious period [days]↵ Rate of (travel-related) import of virus into model domain [(infections)/day]� Contact rate [(contacts)/day]R
0
(Daily) basic reproductive numbera (Negative) coe�cient in contact rate exponentialq Specific humidity [kg/kg]
1
SIRS-q: Humidity-Driven Influenza Model
Heidar Thor Thrastarson
August 3, 2015
1 SIRS Model
Model of influenza transmission dynamics. SIRS (Susceptible-Infectious-Recovered-Susceptible) typemodel. Similar to Shaman & Karspeck (2012). Two coupled first-order ODEs for S and I:
dS
dt=
N � I � S
L� �IS
N� ↵ (1)
dI
dt=
�IS
N� I
D+ ↵ (2)
where
� =R
0
D=
1
D[R
0min
+ (R0max
�R0min
)eaq] (3)
Some assumptions/limitations:
- Fractional persons
- Only a single influenza strain
- Partial and cross-immunities not represented
- Perfectly mixed - clustering and structured interactions of sub-populations not included
- Variations in host infectiousness not modeled
- Other potential drivers than humidity not included, e.g., school calendar
- No age stratification
Table 1: Variables/Parameters
N Population sizeS Susceptible personsI Infectious (= infected) personsL Average immunity duration [days]D Mean infectious period [days]↵ Rate of (travel-related) import of virus into model domain [(infections)/day]� Contact rate [(contacts)/day]R
0
(Daily) basic reproductive numbera (Negative) coe�cient in contact rate exponentialq Specific humidity [kg/kg]
1
Specific'humidity'
Quasi-Operational Forecasting System
Humidity'Observa4ons'
Humidity'Forecasts'
Influenza'Observa4ons'
SIRS'Model'
Influenza'Forecasts'
• Daily'upda4ng'most'recent'values'for'near^surface'H2O'mixing'ra4o,'AIRS'level'3'data'(v6)''
• NCEP'forecasts'for'near^surface'humidity'
• Influenza'data'assimilated'
'• Center'for'Disease'
Control'(CDC):'^ 'Regional,'weekly'surveillance'records'for'the'propor4on'of'doctor’s'visits'for'influenza^like'illness'(ILI)'^ 'Combined'with'lab'virology'results'for'the'percentage'of'influenza'posi4ve'samples'
• Google'Flu'Trends:''^'Near'real^4me'es4mates'of'influenza'infec4on'rates'based'on'online^search'queries'
• The'output'is'the'number'of'infected'and'suscep4ble'people'in'a'popula4on'(city/state/region)'
Retrospective Simulations ! Pseudo-forecasts performed for 21 US cities (in hindsight with
observed humidity) for 2005-2015 seasons
! 5-day and 10-day influenza forecasts (with AIRS humidity) are often quite accurate (example below for LA, 2013/14 winter)
Thrastarson & Teixeira, 2017
Quasi-Operational Prediction System applied to US cities
• The'system'has'been'running'quasi^opera4onally'for'several'US'ci4es'
• AIRS'near^surface'humidity'is'key'component'of'a'quasi^opera4onal'(produced'daily)'influenza'predic4on'system'
'• Most'recent'values'of'AIRS'near^
surface'specific'humidity'as'well'as'NCEP'humidity'predic4ons'regularly'incorporated'into'the'model'
• ‘Observa4onal’'data'for'influenza'incidence'from'CDC/Google'assimilated'to'make'analysis'and're^ini4alize'model'
• Ensembles'of'forecasts'run'with'different'model'parameter'values'drawn'from'distribu4ons'reflec4ng'limited'constraints'
Example'results'for'the'2016^2017'season'
• Timing'and'rela4ve'behavior'of'influenza'outbreaks'generally'captured'fairly'well,'while'gejng'absolute'numbers'of'affected'people'is'more'challenging'
AIRS and Influenza in South Africa ! Near-surface humidity plays critical role in influenza epidemics
! AIRS near-surface humidity correlates well with influenza cases
! Mathematical relations depend on regional climate (provinces)
! These relations can be used to monitor and predict influenza
4500
3500
2500
15000.004 0.013
Casesp
er100
,000
AverageHumidity[kg/kg]
GautengR2 =0.84705
y =3E+07x2 - 814548x+6944.6
WesternCapeR2 =0.70196
y =1.0196x-1.599
Kwazulu-NatalR2 =0.90757
y =4E+07x2 - 1E+06x+9018
0.004 0.013 0.004 0.013
CorrelationofFluCasesandHumidity
Leastsquaresregressionmodelscomparingaveragenear-surfacehumiditywithaverageflucountsfromtheGoogleFluTrendsdatabase.
Gauteng Kwazulu-Natal Western Cape
Credit:'Anish'Parekh'
Dengue Fever and Environment Dependence • Dengue'Fever'is'the'most'common'
mosquito^borne'virus'in'the'world''• Carried'by'Aedes'aegyp4'mosquitos'
(same'as'Zika,'Chikungunya'and'Yellow'Fever)'–'strongly'affected'by'environmental'condi4ons'
• Temperature'affects'mosquito'development'and'reproduc4on,'frequency'of'feeding,'virus'incuba4on'period'and'geographical'range'of'the'vector'(tropics'and'sub^tropics)'
• Precipita4on'provides'breeding'sites'and's4mulates'egg'hatching,'but'can'also'hurt'habitats'through'flooding'and'humidity'has'also'been'iden4fied'as'a'substan4al'factor'affec4ng'favorable'condi4ons'for'the'vector'
• Typically'expect'effects'of'temperature'and'humidity'to'take'6^8'weeks'
AIRS – Dengue • We'are'exploring'the'applica4on'of'AIRS'climate'data'to'the'predic4on'of'dengue'fever'incidence.'
• An'ul4mate'goal'is'to'create'and'implement'an'improved'predic4on'model'for'Dengue.'
• We'have'done'a'focused'study'on'Dengue'fever'in'Mexico,'from'2003^2015'
• Mexico'has'significant'Dengue'incidence'in'varied'climate'condi4ons,'with'weekly'Google'Dengue'Trends'data'available'at'state'level'as'an'es4mate'of'disease'ac4vity'
• AIRS'variables:'surface'air'temperature,'specific'humidity,'rela4ve'humidity'
• Examined'trends,'palerns'and'4me'lags,'regression'models'of'varying'complexity'
Den
gue
case
s (n
orm
aliz
ed)
AIRS and Dengue Fever in Mexico
These relations are being tested to monitor and predict Dengue incidence during its decay phase
Dengue cases correlate particularly well with AIRS humidity during decay phase H
umid
ity (g
/kg)
Time (week)
Red – Dengue cases
Blue - humidity
Dengue and Climate in Mexico Comparisons'of'year^averaged'Dengue'ac4vity'(Google'Dengue'Trends,'top)'and'AIRS'near^surface'temperature'and'humidity'(bolom)'show'climate^related'regional'differences''
Tropical'
Arid'
Dengue'Ac4vity'(GDT,'year^averaged)'
Atmospheric'Variables'(AIRS,'year^averaged)'
Abs.'Hum
idity
'Re
l.'Hu
midity
'
Different'states'
More%tropical%regions%
More%arid%%%%regions%
More'extended'and'elevated'dengue'seasons'
Dengue'peaks'more'sharply'and'only'once'
In'correla4on'with'extended'and'elevated'temperatures'and'humidity'
In'correla4on'with'steeper'rises'and'falls'in'temperature'and'humidity''
Shorter'4me'lags'in'the'climate^dengue'rela4onship'
Longer'4me'lags'in'the'climate^dengue'rela4onship'''
K'
%'
''
Credit:'Anna'Zeng'
Temp'
g/kg'
Summary
• Results'using'AIRS'and'influenza'data'support'role'of'humidity'and'temperature'in'driving'the'seasonality'of'influenza'outbreaks'
• We'have'developed'a'humidity'driven'influenza'model'opera4ng'on'a'city,'state'or'regional'scale'
• AIRS'near^surface'humidity'is'key'component'of'the'quasi^opera4onal'(produced'daily)'influenza'predic4on'system'
'• For'Dengue'Fever,'we'have'done'a'focused'study'in'Mexico,'exploring'
trends,'palerns'and'4me'lags'for'Dengue'and'environmental'variables,'using'regression'models'of'varying'complexity.''
• Promising'correla4ons'and'climate^related'regional'differences'have'been'iden4fied.'
Future Work '• Further'valida4on'of'the'predic4on'system'
• Obtaining'and'incorpora4ng'more'specific'influenza'incidence'data'from'medical'networks'and'authori4es'and'internet'sources.''
• Developing'confidence'and'uncertainty'measures'(including'effects'of'AIRS'humidity'data'uncertainty)'
• Beler'constraining'of'parameters'and'methods'that'account'for'their'variability'and'uncertainty'
'• Implemen4ng'and'assessing'longer'term'seasonal'predic4ons'(using'climatology,'maybe'
longer'term'humidity'predic4ons'in'the'future)'
• Exploring'different'types'of'models,'virus'subtypes,'popula4on'age'structure,'geographical'spread'
• More'engagement'with'poten4al'end'users'