An update on the google- funded UCAR Meningitis Weather Project Abudulai Adams-Forgor, Patricia Akweongo, Anaïs Columbini, Vanja Dukic, Mary Hayden, Abraham.

An update on the google-funded UCAR Meningitis Weather Project

Abudulai Adams-Forgor, Patricia Akweongo, Anaïs Columbini, Vanja Dukic, Mary Hayden, Abraham Hodgson, Thomas Hopson, Benjamin Lamptey, Jeff Lazo, Roberto Mera, Raj Pandya, Jennie Rice, Fred Semazzi, Madeleine Thomson, Sylwia Trazka, Tom Warner, Tom Yoksas

NC STATE UNIVERSITY1

Outline: Short-term weather forecasts to help allocate scarce meningitis vaccine

• Project goals:1. Minimize meningitis incidence by providing 1-14 day weather

forecasts to target dissemination of scarce vaccine2. Contribute to better understanding of disease transmission with a

focus on intervenable factors

• Activities: 1. Predict district level onset of high humidity, a factor that may

contribute to the end of an epidemic2. Verify and quantify the historical relationship between weather and

meningitis3. Build an information system to support vaccination decisions in real

time 4. Examine human-environmental factors that influence meningitis 5. Evaluate the economic benefit of improved weather prediction

Humidity and meningitis

• In April 2009, the Kano epidemic stopped after relative humidity crossed above a 40% threshold

• Attack rates fell in D’jamena and Gaya when average relative humidity for the week rose above 40%.

Slide from Roberto Mera

Modeling meningitis-weather dependence• Uses a differential equation-based model of MRSA• Adds physical insight into meningitis transmission• Numerous assumptions:

– Number of cases small compared to overall population– District population is constant– Carriage is proportional to population– Proximity to neighboring districts with cases influences the chances of

having a case– Same mechanisms determine transmission and infection across belt– The disease cycle is less than two weeks– Weather in the centroid of the district is representative of district-

wide weather

Slide from Vanja Dukic and Tom Hopson

Data from Clement Lingani (via Stéphane Hugonnet)

Vapor pressure (current, lagged by 1 and 2 weeks) correlated with probability of case occurrence. Other variables such as temp, wind or wind from the NE not significantly correlated with probability of cases (stochastic data set)

Forecasting the end of an epidemic

1. Use relationship between (current and lagged) VP and probability of epidemic :

• To determine which districts show historic variance in epidemic end time as predicted by vapor pressure

• For those districts, to predict a vapor pressure at which the epidemic typically declines

2. Predict vapor pressure using quantile regression and global models

3. Use those forecasts of vapor pressure to predict the probable end of epidemic

Using ‘Quantile Regression’ to better predict vapor pressure from global ensembles

Without Quantile Regression:Observations outside range of ensembles

With Quantile Regression: Ensembles bracket observations

From Tom Hopson

Surveys• KN district – upper East Region of Ghana• Administered in preferred language• Goal

– 100 cases 2007-present– 300 age-, gender- , location-matched controls

• So Far– 66 cases, 134 control surveys completed

• To Do– Geo-code all surveyed households

• Temperature and humidity measured hourly along a N-S transect in 20 households– 10 cases, 10 controls– each site has one inside and one inside

Knowledge, Attitudes, and Practices Survey• Administered to all cases and controls• Part I: KAP

– Knowledge of meningitis– Personal and household experience with meningitis– Customs and practices– Attitudes about diseases

• Part II: Socio-demographics– Education-literacy– Occupation (travel)– Housing (ventilation, sleeping arrangements)– Cooking, water, waste, etc. – Household assets; food security

Cost of Illness Survey

• Administered only to cases– Costs of the case

• Medicine, transport to and from hospital, provision of meals, cost of treatment

– Costs in terms of missed work (either directly or for caregiver)

– Costs due to sequelae– Limitations – recall bias in earlier cases

• To do– Estimate costs borne by government

Thank you!

• [email protected]• [email protected]• [email protected]• [email protected]

Fitting all Weather Variables together …

Step-wise forward selection used logistic regression and cross-validation with Brier score cost function

RH VP AIRT VP/T TOTWIND NEWINDcurrent const 0 1 0 0 0 0lag1 const 0 1 0 0 0 0lag2 const 1 1 0 0 0 0

current P 0 0 0 0 0 0lag1 P 0 0 0 0 0 0lag2 P 0 0 0 0 0 0

current P/A 0 0 0 0 0 0lag1 P/A 0 0 0 0 0 0lag2 P/A 0 0 0 0 0 0

current Pr/r 0 0 0 1 0 0lag1 Pr/r 0 0 0 1 0 0lag2 Pr/r 0 0 0 0 0 0

current Pr/r/Ar 0 0 0 0 0 0lag1 Pr/r/Ar 0 0 0 0 0 0lag2 Pr/r/Ar 0 0 0 0 0 0