Stochastic Modeling of Infectious Diseases The 34 th Quality and - PowerPoint PPT Presentation

Stochastic Modeling of Infectious Diseases The 34 th Quality and Productivity Research Conference - 2017 Volodymyr Serhiyenko vserhiyenko@metabiota.com June 15 th 2017

Agenda Hi Historical cal Examp xample • Metabiota Overview • Disease Spread Modeling • Preparedness Index and • Coronavirus Risk Model

Outbreak Starts… • On February 21 st (2003), 64-year-old doctor who was treating “atypical pneumonia” in Guangdong province (China) arrived in Hong Kong to attend a wedding and stayed in Hot Hotel l Me Metr tropole • Ne Next da day he felt ill and was admitted to the intensive care unit • On March 4 he died from a mysterious respiratory disease of unk unkno nown n ori rigin n

Outbreak Spreads… • 20 20 cases ses were associated with the transmission on 9 th floor started from the index patient who spent only on one night in the hotel • 7 out of 20 cases were responsible for consequent large outbreaks in Canada , Vi Ca Vietnam , Si Sing ngapore , 9th floor layout of the Ho Hotel l Metropole le in Hong Kong Source: Christopher R. Braden, Scott F. Dowell, Daniel B. Jernigan, and Ho Hong Kong itself and James M. Hughes - Emerging Infectious Diseases Journal, Volume 19, Number 6—June 2013 • In Vietnam, Dr Dr. Ca Carlo Urbani , a WHO physician, recognized a new and highly contagious disease. He later became infected and died, but his ea early warni ning ng started a massive response worldwide

Outbreak Aftermath… • Mysterious disease was eventually named as Sev Sever ere e ac acute respirat atory syndrome me (SARS) • SARS outbreak started in Guangdong, China, on 16 16 2002 and ended in Taiwan on 5 No November 2002 5 Jul uly 2003 2003 (spreading to 27 es ) 27 count untries Co Count untry Cases Ca es Fa Fatal • Numerous Sup Super er-Sp Sprea eading ng Eve vents ts , like one in China 5327 349 Metropolitan Hotel, had been recorded Hong Kong 1755 299 Taiwan 346 73 CoV spread to humans from wild pa SA SARS-Co palm civet • Canada 251 43 Singapore 238 33 ts that are valued for their meat and are sold in ca cats Vietnam 63 5 Chinese markets. USA 27 0 Philippines 14 2 bats are the na • It is also believed that ba natur ural Other 75 6 reservoirs of SARS-like coronaviruses. re TO TOTAL 8096 810

Lessons learned for modeling… exible to consider different di • We must be fl flex disease s spe pecific cs like Super-Spreading Events, availability of ch charact cteristics vaccines, vaccination strategies (mass or ring), etc. vity and tra el patterns play • Gl Globa bal co connect ctivi travel a crucial role in the spread and magnitude of the disease epidemic • We must be prepared for new newly emer em erging ng infectious diseases

Agenda Historical Example • Me Metabiot ota Ov Overvie iew • Disease Spread Modeling • Preparedness Index and • Coronavirus Risk Model

Metabiota Mission $54B SARS $23B In the last decade, there (Global, 2003) have been over 470 470 MERS $11B human disease (Korea, 2015) outbreaks Foot & Mouth (UK, 2001) $3.3B Avian Flu $2B (US Midwest, 2015) Ebola (W. Africa, 2015) $900M Dengue Fever (Brazil, 2013)

Where is Metabiota? • Founded in 2008 with offices in San Francisco, Canada, Ukraine, Democratic Republic of Congo, and Cameroon and operations in 20 countries

Metabiota Team • Multidisciplinary team • Collaboration with academic partners • Perform epidemiological, statistical, and actuarial modeling

Agenda Historical Example • Metabiota Overview • Di Disease Spr pread d Mode deling • Preparedness Index and • Coronavirus Risk Model

Milestones in Epidemic Modeling

Modeling Cooperation • Me Metabiot ota closely cooperates with Al Alessand essandro Ve Vespignani and his colleagues from Northeastern University's Laboratory for the Mod Modeling of of Bi Biological and Socio-tec techni hnical Sy System stems • Together we are developing disease spread models to realistically simulate disease spark, spread, and duration of epidemics • Our main framework is Gl Global al E Epidemic an and Mo Mobility ty mode model l (GLE LEaM) that stochastically simulates the spread of epidemics at the worldwide scale

GLEaM framework S E I R Exposed Infected Removed Susceptible Underlying Compartmental Model • Global population is divided into basins around transportation hubs (i.e. airports). The resulting network consists of 3, 3,362 362 geographic subpopulations airline transportation data + Full ai nge mobility network + Sho Short rt-ra rang

Individual based model • Model probabilistically progresses ls through each compartment by in indiv ivid iduals ng values from stocha stoc hasti stically si simul ulati ting binomial and multinomial distributions 𝑁𝑣𝑚𝑢𝑗𝑜𝑝𝑛𝑗𝑏𝑚 𝐹 + 𝑢 , 𝑞 . / →1 / 2 , 𝑞 . / →1 / 3 , 𝑞 . / →1 / 43 𝐶𝑗𝑜𝑝𝑛𝑗𝑏𝑚 𝑇 + 𝑢 , 𝑞 7 / →. / • More details can be found in: Balcan, D., Gonçalves, B., Hu, H., Ramasco, J. J., Colizza, V., & Vespignani, A. (2010). Modeling the spatial spread of infectious diseases: The GLobal Epidemic and Mobility computational model. Journal of computational science , 1(3), 132-145.

Agenda Historical Example • Metabiota Overview • Disease Spread Modeling • Pr Preparedness Index and • Co Coronavirus Risk Model

Coronavirus Outbreaks 2003 outbreak: 8096 8096 cases, 810 810 deaths, 27 27 countries effected • SA SARS 2003 iratory Syndrome (MERS) 2013 outbreak: 1980 1980 cases, 699 699 deaths, 15 15 • Mid Middle le East Respir countries effected (as of June 6, 2017) – cased by novel MERS-CoV • First case reported in Saudi Arabia April 2012, still on on-go going • Saudi Arabia is the most affected country (80% of total cases) • Notable event: South Korea 2015 MERS outbreak Caused by on one index patient Source: de Wit et al., SARS and 182 182 cases with 37 37 deaths MERS: recent insights into emerging coronaviruses, 2016

Model Design – Compartments • Main model parameters: R 0 - basic reproductive number (number of secondary infections) 𝜗 9: - incubation period 𝜈 9: - infectious period Travel Reduction (%) Transmissibility reduction time, etc. • Super-Spreading Events: 𝑂𝑣𝑛𝑐𝑓𝑠 ~ 𝑂𝑓𝑕𝐶𝑗𝑜𝑝𝑛𝑗𝑏𝑚(𝑆 J , 𝑙) 𝑝𝑔 𝑡𝑓𝑑𝑝𝑜𝑒𝑏𝑠𝑧 𝑑𝑏𝑡𝑓𝑡

Differences among countries If outbreak starts in US USA , is it Hospital beds per capita going to be different from Ch Chin ina or Sierra Le Leone outbreak of the same disease? Country-level differences in • Outbreak surveillance • Outbreak reporting time Source: World Bank • Timing of intervention measures • … How do we capture these differences?

Epidemic Preparedness Index (1 (1=most p prepar ared, 4 , 4=leas ast p prepar ared) PHI: Public Health Infrastructure PI: Physical and Communications Infrastructure IC: Institutional Capacity EF: Economic Factors PHC: Public Health Communications

EPI influence on CFR On average, improving country’s Epidemic Preparedness by one unit is decreasing odds of dying by 28%

Be the Trusted Disease Model Li Library Pr Proprietary Da Data Set Source for 1M year stochastic event catalog • • 1,200+ Outbreaks Best in Class 18M stochastic realizations with • • 150+ Pathogens Models weekly resolution informed the • 240+ Data Sources event catalog • 230+ Countries / Territories 180K simulations evaluated • • Over 48M Cases 117K distinct demographic • subpopulations • Over 6M Deaths 88K+ AWS Compute optimized • • Curated, cleansed, hours to date continuously updated 100+TB of data • Largest in the industry •

Thank you! Questions?

Model Design – Spark Map Data Layers Bioclimatic Data Number of shared human-bat viruses Zoonotic mammal species Proximity to large cities Human Density Bat (Taphozous sp.) Dromedary Camel abundance PREDICT (Metabiota) data