Program

Last updated on 14 July 2017



Time

Activity

08:00-08:30

Registration

08:30-08:35

Opening Remarks

Joseph Wu
The University of Hong Kong

08:35-08:40

Welcome Remarks

Keiji Fukuda
The University of Hong Kong

Moderator: Joseph Wu, The University of Hong Kong

08:40-09:20

Keynote

Precision Public Health

Scott F. Dowell
Bill & Melinda Gates Foundation

09:20-10:00

Keynote

Harnessing Multinational Data Sources in Global Models to Inform Vaccine Policy

Mark Jit
London School of Hygiene and Tropical Medicine

10:00-10:40

Keynote

Using Genetic Data at Multiple Scales to Understand Constraints on Viral Adaptation

Katia Koelle
Duke University

10:40-10:55

Refreshment break

Moderator: Tommy Lam, The University of Hong Kong

10:55-11:15

Young Investigator
Award

Virus Genomes Reveal the Factors that Spread and Sustained the West African Ebola Epidemic

Gytis Dudas
Fred Hutchinson Cancer Research Center

11:15-11:35

Integrated Analysis of Epidemiological and Genetic Data in Outbreak Settings

Lucy Li
Harvard T.H. Chan School of Public Health

11:35-11:55

The Roles of Selection and Transmission Processes in the Emergence of Drug-resistant Bacteraemia in Thailand: A Modelling Approach

Jiraboon Tosanguan
Mahidol-Oxford Tropical Medicine Research Unit

11:55-12:15

Harnessing Big Data to Quantify Vaccine Hesitancy and the Erosion of Herd Immunity

Shweta Bansal
Georgetown University

12:15-12:35

Monitoring the Fitness of Antiviral-resistant Influenza Strains During an Epidemic: A Mathematical Modelling Study

Kathy Leung
The University of Hong Kong

12:35-14:00

Lunch

Moderator: Steven Riley, Imperial College London

14:00-14:40

Keynote

Towards Digital Pathogen Surveillance: A Very Bumpy and Winding Road

Jennifer Gardy
The University of British Columbia

14:40-15:20

Keynote

Predicting Healthcare Demand during Epidemics in France

Simon Cauchemez
Institut Pasteur

15:20-15:40

Refreshment break

Moderator: Herbert Pang, The University of Hong Kong

15:40-16:00

Young Investigator
Award

Joint Genetic, Digital And Epidemiological Inference to Contain Viral Epidemics. The Examples of Zika and Ebola Virus

Moritz U.G. Kraemer
Harvard Medical School

16:00-16:20

Resistance Gene Selection in the Human Gut Microbiome

Rene Niehus
University of Oxford

16:20-16:40

Characterizing the Dynamics Underlying Global Spread of Emerging Infectious Diseases

Lin Wang
The University of Hong Kong

16:40-17:00

Harnessing Medical Claims Data for Influenza Surveillance and Inference

Elizabeth Lee
Georgetown University

17:00-17:20

Using Epidemiological and Genetic Data to Understand the Spread of Multi Drug-resistant Bacteria in Healthcare Settings

Ben Cooper
Mahidol-Oxford Tropical Medicine Research Unit




Time

Activity

Moderator: Elad Yom-Tov, Microsoft Research

08:30-09:10

Keynote

Influenza Immunity and Transmission in Time, Age and Space

Steven Riley
Imperial College London

09:10-09:50

Keynote

Cura Te Ipsum: Health Search at Google

Evgeniy Gabrilovich
Google

09:50-10:30

Keynote

Quantifying the Global Variation in Risk of Influenza Virus Emergence

Colin Russell
University of Cambridge

10:30-10:50

Refreshment break

Moderator: Eric Lau, The University of Hong Kong

10:50-11:10

Young Investigator
Award

Mobile Real-time Surveillance and Genomic Sequencing of Zika Virus in Brazil

Nuno R. Faria
University of Oxford

11:10-11:30

The Use of Serological Data in Understanding Past and Predicting Future Transmission for Arboviruses in Vietnam

Hannah Clapham
University of Oxford

11:30-11:50

Climatic Factors Driving the Transmission of Human Influenza and Other Respiratory Viruses

Tommy Lam
The University of Hong Kong

11:50-12:10

Physician versus Computer Coding of Verbal Autopsies, A Randomised Control Trial of 12,500 Deaths in 5 Districts of India

Mireille Gomes
St. Michael’s Hospital and University of Toronto

12:10-12:30

Metagenomics Study of Individuals with Seasonal Influenza

Herbert Pang
The University of Hong Kong

12:30-14:00

Lunch

Moderator: Mark Jit, London School of Hygiene and Tropical Medicine

14:00-14:40

Keynote

Using Multiple Infectious Disease Models for Local Decision Making to Control Neglected Tropical Diseases

Deirdre Hollingsworth
University of Warwick

14:40-15:20

Keynote

Digital Health & Epidemiology

Marcel Salathe
École Polytechnique Fédérale De Lausanne

15:20-15:40

Refreshment break

Moderator: Mark Jit, London School of Hygiene and Tropical Medicine

15:40-16:00

Spread of Zika Virus in the Americas

Kaiyuan Sun
Northeastern University

16:00-16:20

Contrasting the Geographic Transmission Patterns of Three Recent Influenza Outbreaks in the United States

Stephen Kissler
University of Cambridge

16:20-16:40

High-resolution Contact Network Measurements in Developing Country Settings

Ciro Cattuto
ISI Foundation

16:40-17:00

Modeling and Analysis of 2016 Yellow Fever Outbreak in Luanda, Angola

Daihai He
Hong Kong Polytechnic University

17:00-17:20

Establishment of Lowly Pathogenic H5N2 Influenza Lineage in Domestic Ducks in Taiwan

Peiyu Huang
The University of Hong Kong

17:45

Shuttle bus to Kowloon Shangri-La

18:30-21:30

Conference Dinner at Cafe Kool of Kowloon Shangri-La


Time

Activity

Moderator: Ben Cooper, Mahidol-Oxford Tropical Medicine Research Unit

09:00-09:40

Keynote

Analysis of Emerging Epidemics in the Era of Real-time Pathogen Genomics

Oliver Pybus
University of Oxford

09:40-10:00

Contribution of Population Structure and Mobility Assumptions to Variation in Intrinsic Infection Attack Rate

David Haw
Imperial College London

10:00-10:20

Antibiotic Resistance in a Spatial and Temporal Model

Mathupanee Oonsivilai
Mahidol-Oxford Tropical Medicine Research Unit

10:20-10:40

The Molecular Phylogenetic Analyses of Imported Zika Virus Genomes in China Suggest Its Local Transmission in Venezuela and Samoa

Jiufeng Sun
Guangdong Provicial Center for Disease Control And Prevention

10:40-11:00

Refreshment break

Moderator: Joseph Wu, The University of Hong Kong

11:00-11:40

Keynote

From Disease Tracking to Disease Understanding Using Internet Data

Elad Yom-Tov
Microsoft Research

11:40-12:00

Epidemic Dynamics Attributable to Day-of-week Mobility Patterns: A Shanghai City Simulation

Zhanwei Du
The University of Texas at Austin

12:00-12:20

Identifying Artemisinin Resistance from Parasite Clearance Half-life Data with A Shiny Web Application

Sai Thein Than Tun
Mahidol Oxford Tropical Medicine Research Unit

12:20-12:40

Long Term Evolution and Transmission Dynamics of H3 Subtype of Influenza Virus in Pigs

Ziying Jin
The University of Hong Kong

12:40-13:00

Synthesis of Pathogen Genomic Sequencing Data and Clinical Records for Efficient Control of Nosocomial Infections

Rosemarie Sadsad
Westmead Hospital

Virus Genomes Reveal the Factors that Spread and Sustained the West African Ebola Epidemic
Gytis Dudas, Luiz Max Carvalho, Trevor Bedford, Andrew J Tatem, Marc Suchard, Philippe Lemey, Andrew Rambaut
The 2013-2016 epidemic of Ebola virus disease in West Africa was of unprecedented magnitude, duration and impact. Extensive collaborative sequencing projects have produced a large collection of over 1600 Ebola virus genomes, representing over 5% of known cases, unmatched for any single human epidemic. In this comprehensive analysis of this entire dataset, we reconstruct in detail the history of migration, proliferation and decline of Ebola virus throughout the region. We test the association of geography, climate, administrative boundaries, demography and culture with viral movement among 56 administrative regions. Our results show that during the outbreak viral lineages moved according to a classic 'gravity' model, with more intense migration between larger and more proximate population centres. Notably, we find that despite a strong attenuation of international dispersal after border closures, localised cross-border transmission beforehand had already set the seeds for an international epidemic, rendering these measures relatively ineffective in curbing the epidemic. We use this empirical evidence to address why the epidemic did not spread into neighbouring countries, showing that although these regions were susceptible to developing significant outbreaks, they were also at lower risk of viral introductions. Finally, viral genome sequence data uniquely reveals this large epidemic to be a heterogeneous and spatially dissociated collection of transmission clusters of varying size, duration and connectivity. These insights will help inform approaches to intervention in such epidemics in the future.

The Molecular Phylogenetic Analyses of Imported Zika Virus Genomes in China Suggest Its Local Transmission in Venezuela and Samoa
Jiufeng Sun, De Wu, Changwen Ke
Zika virus (ZIKV) is an arbovirus (Flaviviridae, Flavivirus), firstly isolated from a sentinel rhesus monkey from Uganda in 1947.The rapid transmission and potential link to severe neurologic complications and birth defects have made ZIKV infection a serious threat to global public health. However, the initial ancestor of ZIKV and the potential transmission route remain in doubt. In 2016, 28 confirmed ZIKV infection cases were imported into China. In total, thirteen ZIKV genomes were successfully amplified from clinical samples and ZIKV isolates. The molecular clock phylogenic analyses estimated the evolutionary rate of the Asian lineage to be 8.86 × 10-4 substitutions/site/year (95% highest posterior density interval, HPD = 7.74 × 10-4 - 1.01 × 10-3 substitutions/site/year) The estimated time-scaled phylogeny contains a well-supported cluster of ZIKV strains (posterior probability, PP = 1.0) that share a common ancestor with the French Polynesia lineage in 2013 (clusters a and b, PP = 0.99) ,which indicates a common ancestor likely transmitted from French Polynesia to South America, particularly to Brazil. Interestingly, all ZIKV cases imported into China were located in two independent clusters, c and d (PP=1.0). Cluster c contained the isolates imported into China from Venezuela. The common ancestor of this clade descended from an ancestral lineage that can be dated to approximately 2013 (2011-2015, 95% HPD); therefore, the imported China isolates from Venezuela were probably already endemic locally before 2015. In cluster c, all the isolates were imported into China from the South Pacific island Samoa. The common ancestor of cluster c can be dated to approximately 2012 (2009-2016, 95% HPD). The estimated time-scaled phylogeny indicates that it also shared a common ancestor with French Polynesia in 2013 (PP = 0.99), thus, it was unlikely to be transmitted from Brazil in 2015, alternatively, an independent local transmission of ZIKV appears to have been previously established in Samoa after 2013. In addition, we detected 7 extra unique nonsynonymous mutations in clusters c and d compared to the global data set for ZIKV.5 None of these mutations were predicted to significantly affect the interactions between host cells and ZIKV due to the limited X-ray structural information of ZIKV proteins; however, the mutations may indicate that clusters c and d went through independent evolution histories that were not caused by sudden adaption to Chinese hosts, thus, it suggest a potential local transmission of ZIKV in Venezuela and Samoa before 2015.

Joint Genetic, Digital And Epidemiological Inference to Contain Viral Epidemics. The Examples of Zika and Ebola Virus
Moritz U.G. Kraemer
Despite many successes in the control of human infectious diseases they continue to pose a considerable risk to human health. Today, the rapid spatial spread of novel pathogens such as Ebola virus in West Africa and Zika in the Americas is unprecedented in their speed and magnitude. The main drivers are rapidly changing ecological and demographic processes. Traditionally, epidemiologists relied on passive disease case reporting and contact tracing data to inform and direct public health responses. However, to accommodate the increased speed of change in recent years, digital epidemiology allowed for more rapid case detection and advances in high-resolution genomic epidemiology enabled the reconstruction of virus spread and the detection and quantification of importation events that lead to local outbreaks and persistence in human populations. With the right tools, such information can aid epidemiological inference and a unifying formal integration of genetics into epidemiology would ultimately help contain and eradicate disease faster. Here, I propose a novel analytical and inferential framework that meaningfully connects genetic information about the virus importation process with high-resolution disease mapping tools, in order to better predict where and when deadly viruses emerge and spread and how they can be contained. I use the examples of Ebola and Zika to illustrate the need for a joint integration of all three methods: 1. I show how digital epidemiology, specifically participatory methods can complement epidemiological data especially facilitating the early detection of the virus at high spatial resolution. 2. I illustrate how genetic inference tools can be used to infer the likely importation patterns of disease, 3. How epidemiological methods can constrain the inferences made from genetic data in regards to the size and composition of transmission clusters. In the context of the Ebola virus disease outbreak in West Africa continued re-importation of disease into large cities driven by human mobility triggered the spread of the outbreak from the origin. I show how mobility was important in the first phase of the outbreak (expansion phase) and then declined in importance once the peak of the epidemic was reached. In the context of ZIKV there were two important distinctive processed that could be observed: 1. During the invasion process into the Americas the epidemic quickly became endemic in the continent causing large local outbreaks after initial importation. This process was driven by the coinciding nature of the travel related importations and local ecological conditions conducive to arbovirus transmission. 2. In Florida, however, ZIKV was repeatedly introduced and high-resolution genomic data reveal the likely number of clusters generated by the outbreak. In addition, epidemiological analysis was able to complement the genetic analyses by estimating the reproduction number of the outbreak in Florida.

From Disease Tracking to Disease Understanding Using Internet Data
Elad Yom-Tov
Over the past decade, infectious diseases, especially influenza, have been tracked using Internet data. These data include search engine queries and social media, as well as other forms of user generated data. The mechanisms developed to track these diseases allow us to take a further step and achieve deeper understanding of these diseases. I will demonstrate this approach beginning with the modeling of Respiratory Syncytial Virus (RSV), the leading cause of hospitalization in children less than 1 year of age in the United States. Using limited surveillance data from 5 US states we were able to estimate disease load for the entire US through domain adaptation. Moreover, this estimate enables tracking the spread of RSV across the US by observing the time of peak use of the search term in different states, providing deeper insights to the seasonal progression of the disease. Similarly, I will show that the application of novel signal-separation methodology can help in estimating the SIR model parameters for individual pathogens from a high resolution estimate of the rate of influenza-like illness derived from Twitter data.
Elad Yom-Tov is a Principal Researcher at Microsoft Research and a Visiting Assistant Professor at the Faculty of Industrial Engineering and Management, Technion. Before joining Microsoft he was with Yahoo Research, IBM Research, and Rafael. Dr. Yom-Tov studied at Tel-Aviv University and the Technion. He has published four books, over 100 papers (of which 4 were awarded prizes), and was granted more than 20 patents. His primary research interests are in using Machine Learning and Information Retrieval to improve health. His latest book is “Crowdsourced Health: How What You Do on the Internet Will Improve Medicine” (MIT Press, 2016).

Mathematical and Cost-effective Modelling for Elimination of Tuberculosis (TB) in Bangladesh
Md Abdul Kuddus, Michael Meeham, Emma McBryde
Mycobacterium tuberculosis (MTB) kills more people each year than any other infectious disease including HIV and malaria. Tuberculosis has been a major public health concern in Bangladesh for decades. According to the World Health Organization (WHO) there are 22 high TB burden countries and Bangladesh ranks sixth among them. Currently, multi-drug resistant (MDR) TB is emerging as the greatest threat to TB control globally. Novel agents have emerged for treatment and human trials in the last few years after a long hiatus, hence it is crucial to deploy these judiciously and to examine ways in which any pipeline drugs could be introduced in to new regimens. In this paper, we have developed reliable mathematical and cost-effective models equipped by stability analysis of the corresponding numerical technique to support the design and characterization of the TB disease modelling in Bangladesh. Results reveal factors -such as antibiotic pressure, poor use of antibiotics, weak health systems with poor follow-up and incomplete antibiotic courses - that were highly influential in increasing risk of resistance. This cost effective modelling approach also provides the prediction of incidence in Bangladesh in 2020 and recommends some elimination strategies for TB. Key Words: Mathematical modelling, Reproduction number, TB, Infected population, Prediction, Influencing factors, Bangladesh

Mobile Real-time Surveillance and Genomic Sequencing of Zika Virus in Brazil
Nuno R. Faria, Josh Quick, Moritz Kraemer, Simon Cauchemez, Marcio Nunes, Luiz Alcantara, Ester Sabino, Nick Loman, Oliver Pybus
Zika virus (ZIKV) transmission in the Americas was first confirmed in May 2015 in Northeast Brazil. Brazil has the highest number of reported ZIKV cases worldwide as well as the greatest number of cases associated with microcephaly and other birth defects. Following the initial detection of ZIKV in Brazil, 47 countries and territories in the Americas have reported local ZIKV transmission, with 24 of these reporting ZIKV-associated severe disease. Yet the origin and epidemic history of ZIKV in Brazil and the Americas remain poorly understood, despite the value of such information for interpreting past and future trends in reported microcephaly. To address this we generated >50 complete or partial ZIKV genomes, mostly from Brazil, and report data generated by the ZiBRA project – a mobile genomics lab that travelled across Northeast (NE) Brazil in 2016. One sequence represents the earliest confirmed ZIKV infection in Brazil. Joint analyses of viral genomes with ecological and epidemiological data estimate that ZIKV epidemic was present in NE Brazil by March 2014 and likely disseminated from there, both nationally and internationally, before the first detection of ZIKV in the Americas. Estimated dates of the international spread of ZIKV from Brazil indicate the duration of pre-detection cryptic transmission in recipient regions. NE Brazil’s role in the establishment of ZIKV in the Americas is further supported by geographic analysis of ZIKV transmission potential and by estimates of the virus’ basic reproduction number.

Antibiotic Resistance in a Spatial and Temporal Model
Mathupanee Oonsivilai, Ben Cooper
Antibiotic resistance is amongst the world’s most important public health problems: it threatens the safety of patients that undergo medical procedures, is associated with worse patient outcomes, and increases economic costs of treatment. Much research has been done to model the transmission dynamics of resistance, but existing models often struggle to sufficiently explain coexistence between antibiotic-sensitive and resistant phenotypes and our current understanding of how antibiotic usage patterns affect these is limited. A potentially important limitation of most models of spread of antibiotic resistance in the community is that they overlook spatial and temporal clustering of antibiotic treatment. We aimed to investigate the impact of spatial and temporal correlation in patterns of antibiotic usage on the spread and coexistence of susceptible and resistant bacteria strains using a simple model. We developed a spatial model for a bacterial pathogen following susceptible-infected-susceptible (SIS) dynamics. The model allows for two strains of this pathogen, one fully sensitive to antibiotics and one fully resistant. We use the model to explore how changing both the frequency of antibiotic use and the extent of spatial and temporal clustering affect the system dynamics and stability. We explore the sensitivity of our results to assumed contact patterns, comparing models with only local contacts between individuals, fully mass action models, and models with small-world like connection topologies. While the aim of this work was to generate insight rather than to directly represent a real-world data-set, our work is informed by the analysis of detailed data (including whole genome sequences of 3085 Streptococcus pneumoniae isolates) from a study of acute respiratory infection in a long-term refugee camp in Mae La, northwest Thailand. We argue that our simulator can help us better understand how antibiotics select for resistant phenotypes, while also providing a framework for thinking about potential interventions to combat the spread of resistance.

Project Tycho 2.0: New Release of Open Access Infectious Disease Data with a Global Scope, Improved Inter-operability, and Easier Access
Wilbert Van Panhuis
In 2013 we released the first version of Project Tycho, an open access resource for infectious disease data containing weekly disease case counts for 50 notifiable conditions reported by health agencies in the United States for 50 states and 1284 cities between 1888 and 2014. Over the past 3.5 years, over 3000 users have registered to use Project Tycho data and we have had 36,500 online visitors resulting in a total of 226,500 page views of our Project Tycho website. The value of releasing this dataset has become clear as 16 peer-reviewed papers have used these data so far and 235 media articles mentioned Project Tycho. The first version of Project Tycho data was limited to US data, only partially standardized and excluded information that we could not yet standardize. Furthermore, users could only download data through our online user interface or through the application programming interface (API). Now, we released Project Tycho 2.0, a new iteration of Project Tycho that has a global scope, includes more data, is more extensively standardized, and can be downloaded in various ways. Project Tycho 2.0 includes case counts for 28 additional notifiable conditions for the US and dengue case count data for 100 countries between 1955 and 2010, obtained from the World Health Organization and national health agencies. Most of these data are at a subannual and subnational spatiotemporal resolution. Project Tycho 2.0 data are standardized using external codes for diseases (SNOMED-CT and ICD-10), locations (ISO and GADM), and pathogen species (NCBI TaxonID). These standards greatly increased interoperability between Project Tycho data and other datasets. Project Tycho 2.0 data can be accessed in various ways, including a redesigned online query interface, a Project Tycho R-package, GitHub, and a new API. We have standardized our data curation workflow and in the near future, will be able to accept data submissions from investigators and others that would like to include their data in Project Tycho. We will also integrate highly curated datasets that have been released by investigators as part of peer-reviewed papers, so that these data can be centrally queried and downloaded. It is our aim to continue expanding Project Tycho towards a global resource of infectious disease, and related data to catalyze biomedical research and innovation.

Long Term Evolution and Transmission Dynamics of H3 Subtype of Influenza Virus in Pigs
Ziying Jin, Xiaohui Fan, Yiu-Man Cheung, Leo Lit-Man Poon, Malik Peiris, Yi Guan, Huachen Zhun
Introduction: The bidirectional transmission of influenza A viruses between human and swine and the sporadic introduction of viruses from birds poses a continuous threat to public health and agriculture, as indicated by the emergence and development of 2009 pandemic H1N1 (pdm/09) virus. H3N2 human influenza viruses (huIVs) repeatedly transferred to pigs and led to the establishment of several major swine influenza virus (SIV) lineages. Avian H3 viruses also occasionally cross the species barrier to infect pigs. This study aims to depict the interspecies transmission and evolution of H3 subtype of viruses in pigs . Methods: Based on our systematic surveillance of SIVs in China from 1998-2015, all H3 subtype of SIVs were subjected to whole genome sequencing. Phylogenies of these and the related virus sequences retrieved from GenBank and GISAID were inferred using RAxML v8, Phyml v3.1 and BEAST v1.8.6. Results: From 1998 to 2015, 343 H3 SIVs were isolated, with three closely related to local duck H3N2 viruses, suggesting a recent introduction of Eurasian gene pool-like viruses to the local pigs, without further establishment. All the other H3 viruses were derived from the huIV lineage. In 1998-2004, H3N2 huIVs were regularly introduced into pigs and occasionally reassorted with the prevailing SIVs in southern China. Among the 76 H3N2 isolated from our surveillance program, 46 were phylogenetically dispersed with circulating huIVs in each gene segment, indicating frequent introduction of huIVs into the pig population. The remaining were reassortant viruses (n=30) with all the internal genes obtained from the Eurasia Avian-like (EA) SIVs. This reassortant had been established in Europe before it spread to China and caused a zoonotic infection in a child. From 2005 to late 2010, H3 subtype disappeared in our swine surveillance system until the emergence of A/swine/Guangxi/NS2783/2010(H3N2), a novel reassortant virus with A/swine/Binh Duong /03_06/2010-like surface genes and pdm/09-like internal genes. These NS2783-like viruses (n=264) became established and further developed into 7 genotypes in the subsequent years. Phylogenetic analyses on global SIVs revealed that all the five established H3 SIVs harbored internal genes from enzootic swine lineages (EA, Triple Reassortant, Classical Swine or pdm/09) and surface genes from the huIVs prevailing in different decades. Reassortments with local SIVs frequently occurred, generating multiple transient variants with diverse genotypes. Sporadic human infections have been reports in different countries, with most occurring in the United States contracting the H3N2v (an H3N2 Triple Reassortant variant with M gene from pdm/09). Conclusion: H3N2 influenza viruses are long-term threat to public health and pig farming, warranting a close monitoring of their evolution and development. Understanding the viral determinants and agricultural practices that facilitate the bidirectional transmission and adaptation of influenza viruses is crucial for disease control.

Contrasting the Geographic Transmission Patterns of Three Recent Influenza Outbreaks in the United States
Stephen Kissler, Julia Gog, Cécile Viboud, Vivek Charu, Ottar Bjørnstad, Lone Simonsen, Bryan Grenfell
High-volume medical claims records offer powerful insight into the geographic transmission patterns of infectious diseases. In this paper, a mechanistic transmission model is fit to geo-tagged influenza-like illness data, gathered from ICD9-coded outpatient medical insurance claims records from ~800 cities across the United States. The geographic transmission patterns of the 2003-04 A/H3N2 ('Fujian') influenza outbreak, the 2006-07 mixed-strain outbreak, and the 2009 A/H1N1pdm pandemic influenza outbreak are contrasted. Then, for each season, the transmission model is reverse-engineered to identify the most likely sites of introduction followed by onward spread, or the 'transmission hubs'. Tracing infection forward again from these hubs yields geographic basins of infection, or clusters of cities where infection can be traced back with high probability to one of the hubs. These basins provide a testable hypothesis of each outbreak's phylogeography, and broadly agree with previous phylogeographic studies and available antigenic data.

Integrated Analysis of Epidemiological and Genetic Data in Outbreak Settings
Lucy Li, Nicholas Grassly, Christophe Fraser
Background: Epidemiological parameters are often estimated by fitting transmission models to data. In addition to epidemiological data such as incidence time series, there has been growing interest in phylodynamic approaches in which models are fit to the phylogeny of pathogen sequences. For infectious diseases with low reporting rates, incorporating phylodynamic analyses can help to quantify the total number of infections. Because the same transmission process underlies both types of data, fitting models simultaneously to both types of data should yield more accurate parameter estimates. An important characteristic to model in outbreak settings is the offspring distribution, often parameterized by negative binomial distribution with mean equal to the reproductive number and dispersion parameter k. Mis-specifying k could affect estimates of other parameters because it modulates the degree of unpredictability of an epidemic and it affects the statistical relationship between prevalence of infection and the pathogen phylogeny. We developed a joint inference framework that can be used to estimate epidemiological parameters including the dispersion parameter k from both epidemiological data and pathogen phylogeny. To demonstrate the value of a joint analysis, we tested our approach on both simulated data and real data from a poliovirus outbreak. Method: We developed a particle Markov Chain Monte Carlo (PMCMC) approach to estimate posterior densities, using a particle filter to estimate the marginal likelihood over stochastic realizations of a transmission model. We generated simulated data using a modified stochastic compartmental model with negative binomial offspring distribution, and used our framework to estimate the model parameters. For the polio analysis, we fit an age-structured model to data from the 2010 Tajikistan outbreak of poliovirus. Results: Fitting to simulated data, we found that more accurate and less biased estimates of the reproductive number were obtained by combining epidemiological and phylogenetic analyses. Incorporating pathogen phylogeny was necessary for However, k was most accurately estimated using pathogen phylogeny alone. Accurately estimating k was necessary for unbiased estimates of the reproductive number, but it did not affect the accuracy of reporting probability and epidemic start date estimates. We further demonstrated that inference was possible in the presence of phylogenetic uncertainty by sampling from the posterior distribution of phylogenies. When we applied our method to poliovirus data, the case-to-infection ratio was more precisely estimated by incorporating phylogenetic data. However, there was not evidence of overdispersion in the offspring distribution. Conclusions: Instead of relying on separate inference pipelines, we demonstrated that a joint inference approach using both epidemiological and phylogenetic data improved the accuracy of parameter estimates especially k. The ability to estimate k will improve estimates of the reproductive number and the accuracy of predictive modeling. Furthermore, a PMCMC framework allows integration over stochastic realizations of the model, which is especially important for emerging epidemics when infection numbers are low. Finally, we highlighted the role of phylodynamic analysis in understanding polio epidemiology. As the number of polio cases continues to decrease, the ability to jointly analyze multiple sources of data will be invaluable in elucidating transmission dynamics of polio.

The Roles of Selection and Transmission Processes in the Emergence of Drug-resistant Bacteraemia in Thailand: A Modelling Approach
Jiraboon Tosanguan, Esther Van Kleef, Maliwan Hongsuwan, Cherry Lim, Direk Limmathurotsakul, Ben Cooper
Introduction & Objectives: Antimicrobial resistance (AMR) is a major public health issue. Approximately 700,000 deaths per year are caused by antimicrobial-resistant bacteria globally; the majority of these occurring in developing countries. In Thailand alone, an estimated 19,000 excessive deaths occur annually as a result of multidrug-resistant bacterial infections in hospitalised patients. Major uncertainties exist regarding the drivers of the dissemination of clinically relevant multi-drug resistant bacteria. We aimed to analyse a large volume of routinely collected hospital data from Northeast Thailand using a mechanistic transmission model to determine the relative importance of transmission within the hospital environment versus selection of pre-existing resistant strains as a result of antimicrobial therapy for different pathogens. Methods: Data on 1,803,506 admission records (1,255,571 patients) between 2004 and 2010 from nine provincial hospitals (median bed number = 450 beds) in Thailand were obtained. A total of 8,886 cases of adult bacteraemia with the target bacterial species (Escherichia coli (EC), Acinetobacter spp. (AS), Klebsiella pneumoniae (KP), and Staphylococcus aureus (SA)) were identified. A population-based transmission model was developed to describe the spread of sensitive and resistant strains of bacteria within a hospital, incorporating both selection and transmission processes. Maximum likelihood estimation was used for parameter estimation. Confidence intervals were derived using the profile likelihood method and likelihood ratio tests were used to compare nested models. Results: We found very strong temporal signals whereby the proportion of bacteraemias caused by multiply drug resistant bacteria was an increasing function of the time from admission to infection. The shape of these functions, however, showed substantial variation between bacterial species. Analysis with the dynamic model suggested that both the force of infection and selection periods varied between the bacterial species. Gram-negative bacteria (EC, AS & KP) generally had lower forces of infection and shorter selection periods than SA. This suggests that selection may have played a more dominant role. There was also evidence of an increasing trend of carriage of resistant bacteria on admission to the hospital. Likelihood ratio tests indicated that the models with both selection and transmission components gave better fits to the AS and SA data when compared to models with just selection or just transmission components only. Conclusions: This study has demonstrated the utility of routine hospital surveillance for making inferences about the drivers of the increasing prevalence of multiply-drug resistant bacteria in Thailand, and has highlighted important differences between bacterial species. These findings could potentially be used for informing future studies to evaluate the impact of interventions to reduce the burden of AMR.

Physician versus Computer Coding of Verbal Autopsies, A Randomised Control Trial of 12,500 Deaths in 5 Districts of India
Mireille Gomes, Rehana Begum, Atul Budukh, Dinesh Kumar, Abhishek Singh, Zehang Richard Li, Prabha Sati, Prabhat Jha
BACKGROUND: Cause of death (COD) information for about 40 million out of the 60 million deaths worldwide remain unknown. The largest gap in information is in low and middle-income countries, where most deaths occur at home rather than in health facilities. COD information derived from verbal autopsies (VA), which are structured interviews with family members of the deceased, are a crucial source of data for public health planning. A VA characteristically comprises of a checklist of key symptoms leading up to death and a free-text narrative in local language. Typically, this information is reviewed and cause of death diagnosis is retrospectively made by two physicians independently. Accurate machine learning-based COD assignment to replace (or complement) physician diagnosis could improve mortality statistics. The choice of physician versus computer coding of VA has been hotly debated, as both methods have limitations. Physician diagnosis may suffer from inter and intra-observer differences. A literature search yielded only one study that systematically assessed the performance of several leading computer-coded verbal autopsy algorithms compared with physician diagnosis on 24,000 deaths. Hence, there is a need for direct randomized evidence. METHODS: We conducted the first ever randomised control trial assessing physician versus computer coding of VA. In five districts of India, we allocated 12,500 deaths in the same number of households randomly to dual physician diagnosis (which included a half-page narrative in local language) or to a checklist of questions (without a narrative) for use in current leading computer algorithms. The primary outcome was agreement in the COD distribution at the population level, with the expectation that well performing algorithms would provide a comparable COD distribution to that from physician-coded VAs. This trial is registered at Trials.gov (NCT02810366). RESULTS: The two arms of the trial achieved good overall balance from randomisation. The average population level agreement of computer algorithms with physician coding was 60% for adult deaths, 56% for child deaths (1 month-12 years) and 50% for deaths in the first month of life. No single computer method performed best for deaths at different ages. The highest agreement between computer algorithms and physician assigned COD distribution was ~80%, which is insufficient for public health decision making. A sub-study of 250 deaths found good concordance in narrative information and final COD between VA interviews done by non-medical survey staff and those done by doctors. INTERPRETATION: Computer algorithms cannot yet replace physician coding of verbal autopsies. The use of non-medical interviewers for VAs with central medical coding remains the best standard. Further research focusing on use of combination of algorithms, and use of narrative information (through natural language processing) is required.

Establishment of Lowly Pathogenic H5N2 Influenza Lineage in Domestic Ducks in Taiwan
Pei-Yu Huang, Chang-Chun David Lee, Ziying Jin, Zhihua Ou, Chun-Hung Yip, Chung-Lam Cheung, Guangchuang Yu, Tommy Tsan-Yuk Lam, David Keith Smith, Chwan-Chuen King, Huachen Zhu, Yi Guan
Background: Influenza A viruses (IAVs) of most subtypes can perpetuate and reassort without having any significant fitness cost or lineage establishment in the natural reservoirs, specifically migratory ducks and waders. Domestic ducks are regarded as a part of this reservoir, where recurrent genetic exchange occurs among viruses from varying wild or domestic birds. Despite the fact that domestic ducks have been long hypothesised as a facilitator between natural and poultry gene pools of IAVs, the evolutionary behaviour of IAVs in domestic ducks is still largely unknown. Taiwan accommodates large aggregations of migratory birds as well as high density of terrestrial and aquatic poultry, which provides a suitable platform for the examination of such matter. Materials and Methods: In Dec 2012-May 2014, we conducted a systematic surveillance in a live poultry market in northern Taiwan and identified 586 duck and 141 chicken influenza viruses from 6,585 duck and 14,602 chicken samples. Full genomes of all the isolates and additional 86 archived viruses were deep sequenced with MiSeq platform (Illumina). Maximum likelihood phylogeny of each viral gene was inferred using RAxML v8.1.16 and their temporal phylodynamic was further analysed using Bayesian MCMC method with SRD06 substitution model and uncorrelated lognormal relaxed molecular clock implemented in BEAST v.1.8.4. The amino acid sites under natural selection were detected using SLAC, FEL, IFEL, MEME and FUBAR incorporated in Datamonkey web server as well as the M8+BEB site model of CodeML integrated in PAML 4.9d. Substitutions detected under natural selection were mapped to three-dimensional protein structure and visualised using PyMOL to estimate their biological functions. Results: In this study, we have identified a remarkably high prevalence of H5 subtype (51%) among all other IAV subtypes in domestic ducks in Taiwan. These H5 influenza viruses were mainly H5N2 subtype, of which 94% formed a unique monophyletic group that approximately diverged from Eurasian gene pool in 1900s in H5 gene phylogeny. Unlike most influenza viruses in ducks, the gene reassortment was constrained in N2 and M genes of this novel duck H5N2 lineage (NovH5N2). Elevated rate of nucleotide substitution in H5, N2 and M genes of NovH5N2 was also observed when compared to other duck viruses, significant on the long branches that connect NovH5N2 to its gene pool ancestors. Both mammalian- and avian-like amino acid substitutions at site 151 and 154 (H5 numbering) were observed, indicating the broad spectrum of receptor binding affinity obtained by this virus. The negative selection at site 31S of NovH5N2 M2 protein suggested that the susceptibility to amantadine and rimantadine was retained in this virus. Conclusions: Our results suggested that potentially due to the advent of modern intensive farming practices, influenza viruses in domestic ducks can evolve similarly to those in aberrant hosts and play a role of bridging influenza gene pools in the natural reservoirs and land-based poultry.

Modeling and Analysis of 2016 Yellow Fever Outbreak in Luanda, Angola
Daihai He
Background: Yellow fever (YF) is a life-threatening viral disease related to Ebola that is endemic to tropical regions of Africa and South America and transmitted via bites of infected mosquitoes. YF has largely been controlled by widespread national vaccination. However, between December 2015 and August 2016, YF resurged in Angola and quickly gave rise to the largest YF outbreak of the last 30 years. Only recently YF resurged again in Brazil (December 2016). The present study provides refined mathematical models to assesses and reconstruct important epidemiological processes underlying Angola’s YF outbreak. This includes the outbreak’s attack rate, the reproductive number R0, the role of the mosquito vector, the influence of climatic variables and the unusual (unnoticed) appearance of two-waves in the YF outbreak. The model explores actual and hypothetical vaccination strategies, and the impacts of possible human reactive behavior changes (eg., response to media precautions). The model helps untangle the many complex epidemiological processes characteristic to YF, a contribution that should be of benefit in mitigating the spread and impact of YF outbreaks in the future. Methods and Findings Data: We studied the YF outbreak in Luanda, which is a province and capital city of Angola, and has a population size of 6,543,000 (2016). The WHO published weekly data for Luanda specifying 952 confirmed and suspected weekly YF cases, and 62 deaths over the study period 05 December 2015 to 18 August 2016. Data of the population vaccinated in Angola was obtained from WHO reports from February 2 2016 when coverage was 38% to August 2016 when coverage was 93%. A mathematical epidemiological model was developed for the Angola YF outbreak that includes vector-host dynamics between the human population and mosquito carriers. The model was parameterized from prior knowledge of YF, and accepted parameter values. The YF outbreak was modeled as a Partially Observed Markov Process and makes use of Iterated Filtering and plug-and-play likelihood-based inference frameworks to fit the data. This is the most modern state-of-the-art statistical methodology for fitting complex epidemiological datasets. The model was first fitted to the observed YF cases and deaths, given knowledge of the actual vaccination coverage. The Basic Reproductive Number R0 of our model was considered to be time-dependent. Results were reported for two different scenarios depending on strong or weak asymptomatic transmission from the host population. The model reproduced the observed YF outbreak patterns in Luanda before and after the national vaccination campaign, and provides the first estimate of vaccination effectiveness. While the observed data reports 952 confirmed and suspected cases and 62 deaths over the 37-week period, the model found that in the absence of vaccination campaign there would have been 19,347 cases and 1117 deaths. A 30(/90)-day delay to the vaccination roll-out, would have resulted in 40(/316) extra deaths. The analysis estimated a mean R0 =2.33 and an estimated YF attack rate of 0.02-0.03% (% population infected by YF). Our estimated initial R0 and upper bound R0 are in line with a previous study (4.8).

Harnessing Medical Claims Data for Influenza Surveillance and Inference
Elizabeth Lee, Shweta Bansal
Traditional infectious disease epidemiology is built on the foundation of high quality and high accuracy data on disease and behavior. While these data are usually characterized by small size, they benefit from designed sampling schemes that make it possible to make population-level inferences. On the other hand, digital infectious disease epidemiology uses existing digital traces, re-purposing them to identify patterns in health-related processes. Medical claims are a digital source of epidemiological information that capture patient-level data across an entire population of healthcare seekers, and have the benefits of medical accuracy through reporting by physicians, and fine spatial and temporal resolution in near real-time. Our work harnesses the large volume and high specificity of medical diagnosis codes to improve our understanding of the ecology of influenza, a non-reportable disease with non-specific symptoms that afflicts millions with severe illness annually. The determinants hypothesized to drive these patterns are as varied as: the circulation of different influenza subtypes, environmental factors affecting transmission or virus survival, travel flows between different populations, population age structure, and socioeconomic factors linked to healthcare access and quality of life. In addition, the nature of surveillance data collection must be factored into our understanding of observed epidemiological patterns in order to perform robust inference about drivers of disease. Using aggregated medical claims data for influenza-like illness (ILI) from 2001 to 2009, we developed a Bayesian hierarchical modeling framework to estimate the importance of both socio-environmental and measurement factors on observed spatial variation of influenza disease burden across all counties in the United States. We sought to determine the strength, directionality, and consistency of these factors over multiple flu seasons and measures of disease burden such as total seasonal intensity and epidemic duration. Due to the high resolution of county-level data and the multiplicity of models in our analyses, we used Integrated Nested Laplace Approximation (INLA) techniques for Bayesian inference to render our questions computationally tractable. Our results suggest that both measurement and socio-environmental factors contribute to the spatial variation in reported influenza burden, and that different determinants explain that heterogeneity at different spatial scales. We also found that digital surveillance with fixed reporting locations may produce more robust inference, and that high volumes of data can offset measurement biases in opportunistic digital data samples. Our work tackles open questions in mathematical infectious disease modeling through the synthesis of interdisciplinary hypotheses and non-traditional epidemiological data sources and the integration of data of differing spatial scales. The results of our study improve our understanding of the spatial distribution of influenza and the underlying reporting and data aggregation processes that drive these patterns. Finally, our findings may inform the data collection and spatial aggregation in digital disease surveillance efforts beyond the use of medical claims.

Resistance Gene Selection in the Human Gut Microbiome
Rene Niehus, Ben Cooper, Esther van Kleef
Bacterial resistance to antibiotics is a major emerging threat to global public health, with new resistances emerging and spreading at an accelerating speed. This trend is linked to the global increase in antibiotic use. With the goal to better understand how antibiotics affect humans and bacteria, a lot of funding money has gone into research that studies human associated bacterial communities, especially the largest of them all: the human gut microbiota. With the advent of high-throughput sequencing methods–including shotgun sequencing and 16s pyrosequencing–it has become possible to get a much better understanding of the diversity of this complex community which contains thousands of different bacterial species.Recent studies of our microbiota under the effect of antibiotic treatment show a rapid loss of bacterial diversity.However, we know little about how antibiotic treatment selects for resistance genes within the human gut microbiota community. Here we study the effect of antibiotics on the abundance of the CTX-m gene, the gene that most commonly confers ESBL resistance in Enterobacteriaceae. In our study, the abundance of the CTX-m gene was determined using quantitative PCR (qPCR) performed on stool samples from single patients over the time of hospital stay. Because DNA yield from stool samples can vary, we focus our analysis on relative abundance of resistance gene, which is calculated as CTX-m gene abundance divided by abundance of the conserved bacterial 16s gene region. The study was performed in European hospitals on adult patients who were carriers of ESBL-producing Enterobacteriaceae at the time of hospital admission. We ask, can we predict distinct types of resistance gene abundance patterns from antibiotic exposure of the patients? To address this question we use an array of different statistical and computational methods including hierarchical Bayesian modelling, elastic-net regularized regression, clustering and random forests. Our results show that there is a high diversity of dynamical patterns of resistance gene abundance. Surprisingly, none of the approaches used found any evidence of association between patient antibiotic use and dynamical changes in CTX-M abundance. We argue that to understand the spread of resistance, we must consider a more detailed ecology of the human microbiome, including species interactions and the resulting dynamics. Following this logic, we use parameters from mouse experiments to simulate human microbiota dynamics and we evaluate statistical methods that infer microbial interactions with and without antibiotic treatment. A key result is that for studying these interactions it is better to obtain a larger number of samples for each patient over time, rather than sampling many patients for only a few time points.

Harnessing Big Data to Quantify Vaccine Hesitancy and the Erosion of Herd Immunity
Sandra Goldlust, Shweta Bansal
Despite advances in sanitation and immunization, vaccine-preventable diseases remain a significant cause of morbidity and mortality worldwide. In high-income countries such as the United States, coverage rates for vaccination against childhood infections remain high. However, the phenomenon of vaccine hesitancy makes maintenance of herd immunity difficult, impeding global disease eradication efforts. Reaching the `last mile' will require early detection of vaccine hesitancy (driven by philosophical or religious choices), identifying pockets of susceptibility due to underimmunization (driven by vaccine unavailability, costs ineligibility), determining the factors associated with the behavior and developing targeted strategies to ameliorate the concerns. Towards this goal, we harness high-resolution medical claims data to geographically localize vaccine refusal and underimmunization in the United States and identify the socio-economic determinants of the behaviors. Our study represents the first large-scale ‘big data’ effort for vaccination behavior surveillance and has the potential to aid in the development of targeted public health strategies for optimizing vaccine uptake.

The Use of Serological Data in Understanding Past and Predicting Future Transmission for Arboviruses in Vietnam
Quan Tran Minh, Lam Ha Minh, Thao Tran Thi Nhu, Vy Nguyen Ha Thao, Phuong Huynh Thi , Thanh Nguyen Thi Le, Maciej Boni, Hannah Clapham
Measuring the antibody response to pathogens in a population can be informative about past transmission or vaccination, as well as susceptibility of the population. For infections such as measles with little or no cross-reactivity with other viruses and long-lived immunity, understanding both of these can be straightforward. A serosurvey can tell you what proportion of the population have been infected or vaccinated and are therefore immune, and what proportion of the population is still susceptible. Understanding the size of this susceptible population can help in making predictions about future transmission. Chikungunya is another example of a pathogen with little cross-reactivity and long-lived immunity. There is no vaccine for chikungunya, so interpretation of the serological results gives us information about past transmission and the susceptible pool in the population. We report here on results from testing serum from 550 individuals of 1-80 years old in 4 locations across Vietnam with a chikungunya IgG ELISA. The results suggest that there has not been transmission of chikungunya in Vietnam for the last 30 years, but there was sustained transmission prior to this time. This is consistent with the few reports of chikungunya transmission in the literature 30-50 years ago. With these serological results we are able to estimate 1) the R0 of chikungunya when it was previously spreading in Vietnam and 2) the proportion of the population that is susceptible to chikungunya which can help us understand the potential of chikungunya to spread in Vietnam currently given importation. We contrast chikungunya with dengue, a pathogen also without a vaccine currently in use, but one with much cross-reactivity both within dengue serotypes and across flaviviruses. This cross-reactivity is not only seen in the results of assays to measure antibody response, but also in the epidemiology of the virus: infection with one dengue serotype first protects to all serotypes and then leaves you more likely to have a severe outcome upon infection with another serotype. There are suggestions that this relationship also holds for dengue and other flaviviruses such as Zika. We discuss preliminary work using two methods: ELISA and a novel flavivirus protein microarray to test for population flavivirus antibody response in Vietnam. In this context we also discuss the more complex methods needed to use flavivirus serological data to understand past transmission and population susceptibility for flaviviruses.

Spread of Zika Virus in the Americas
Qian Zhang, Kaiyuan Sun, Matteo Chinazzi, Ana Pastore y Piontti, Natalie E. Dean, Diana, Patricia Rojas, Stefano Merler, Dina Mistry, Piero Poletti, Luca Rossi, Margaret Bray, M., Elizabeth Halloran, Ira M. Longini, Jr., Alessandro Vespignani
The Zika Virus (ZIKV) outbreaks in the Americas, as Public Health Emergency of International Concerns declared by WHO, has generated a call to arms toward gathering data to improve our understanding of the disease and its spreading mechanisms. However, despite the effort of the research community, much uncertainty is still surrounding the spreading of ZIKV. Indeed, little is known about the circulation of the virus prior to the WHO alert, as well as its time of introduction in Brazil. Furthermore the large fraction of asymptomatic infections (~80%) is almost entirely invisible from surveillance system. For this reason it is extremely difficult to evaluate the impact of the disease and assess quantitatively the likely magnitude and timing of the epidemic. In response to the challenge of understanding ZIKV outbreak, we provide a study to analyze the spatial and temporal dynamics of the ZIKV epidemic in the Americas with a microsimulation approach informed by high definition demographic, mobility and epidemic data. Our research is multidisciplinary, combining methodologies from epidemiology, statistics, economics, phylogenetics, computational & data science. Our study provide information of ZIKV outbreaks with high spatiotemporal resolution in terms of time of introduction in Brazil, timing and magnitudes of epidemics as well as zika-affected pregnancies across American countries and importation of ZIKV cases in the US and European countries. The results of our study are critical for the current debate on the ZIKV epidemic as well as preparation and analysis of contingency plans aimed at its mitigation.

Contribution of Population Structure and Mobility Assumptions to Variation in Intrinsic Infection Attack Rate
David Haw, Steven Riley, Derek Cummings
Traditional spatial transmission models of infectious diseases are designed to gain insight into infection data aggregated over time and space. However, a feature of next generation digital epidemiology is that data are often gathered about specific events in time and space and are not necessarily aggregated. Therefore, there is a need to understand how the resolution of spatial models and the data with which they are parameterized affects key model outputs such as cumulative infection attack rates. For example, studies of influenza and other directly transmitted infectious diseases show heterogeneity in attack rates at small spatial scales. These differences may be explained by the interaction of human mobility and variable population density.

We investigate this phenomenon with a scalable spatial model of influenza transmission, including alternative formulations of the force of infection. The model represents scenarios in which only susceptible individuals are mobile (S-mobility), only infectious individuals are mobile (I-mobility), or all agents are mobile (dual mobility). We compare the attack rates in suburban populations with their population density and with the local gradient of population density.

Our results for dual mobility are uniform in space, consistent with theoretical results. However, both S-mobility and I-mobility exhibit substantial variability, correlated with population density and with the gradient of population density. The S-mobility model produces attack rates that are positively correlated with population density and the gradient of population density. Moreover, this result is maintained with partial mobility of infectious agents. The I-mobility model yields negative correlations.

Sensitivity analysis on kernel parameters show that these correlations are reversed for large enough population power. This observation allows us to understand the behavioural mechanisms behind patterns in attack rates at small spatial scales.

Epidemic Dynamics Attributable to Day-of-week Mobility Patterns: A Shanghai City Simulation
Zhanwei Du, Lauren Meyers, Jiming Liu
Population mobility and connectivity patterns drive epidemic progression, but most common epidemic approaches assume static population connectivity. Temporal variation in these patterns has important consequences for disease emergence and subsequent transmission, with even daily variation potentially influencing disease dynamics (e.g. weekend versus weekday). Using a metapopulation model of Shanghai, informed by hourly subway transportation routes, we investigated the location and timing effects of the introduction pathogen on disease emergence. In the simulations, we found the obvious disparity of epidemic fluctuations over locations, which were connected to their populations and mobility with other locations through theoretical and numerical analyses. In terms of the introduction location and time, we found that the average probability and size of a subsequent epidemic was not influenced by them during the week, but they could significantly change the course of an epidemic (such as the peak time). Specifically the epidemic growth rate differed substantially between the introduction locations in city center having higher epidemic growth rates, or the introduction days with Friday’s connectivity mixture leading to the quickest disease dispersal. Finally we find theoretical explanations for these results based on the changing network connectivity. To better understand the connectivity changes, we further explore the mobility flows by community patterns. Unsurprisingly we found that daily community patterns show either a workday or weekday signal, with Friday connectivity showing weekday signals in the morning and weekend at night. The specialty of Friday in epidemic growth rate, as a connectivity mixture,isfurtherexplainedfromtheviewofmobilitychanges. Ourresultssuggestthathourlychangesinconnectivitypatterns can influence the rate of epidemic growth. This information could prove useful to public health policy makers when scheduling responses to potential epidemic threats.

Monitoring the Fitness of Antiviral-resistant Influenza Strains During an Epidemic: A Mathematical Modelling Study
Kathy Leung, Marc Lipsitch, Kwok Yung Yuen, Joseph Wu
Background: Antivirals (eg, oseltamivir) are important for mitigating influenza epidemics. In 2007, an oseltamivir-resistant influenza seasonal A H1N1 strain emerged and spread to global fixation within 1 year. This event showed that antiviral-resistant (AVR) strains can be intrinsically more transmissible than their contemporaneous antiviral-sensitive (AVS) counterpart. Surveillance of AVR fitness is therefore essential. Our objective was to develop a simple method for estimating AVR fitness from surveillance data. Methods: We defined the fitness of AVR strains as their reproductive number relative to their co-circulating AVS counterparts. We developed a simple method for real-time estimation of AVR fitness from surveillance data. This method requires only information on generation time without other specific details regarding transmission dynamics. We first used simulations to validate this method by showing that it yields unbiased and robust fitness estimates in most epidemic scenarios. We then applied this method to two retrospective case studies and one hypothetical case study. Findings: We estimated that the oseltamivir-resistant A H1N1 strain that emerged in 2007 was 4% (95% credible interval [CrI] 3–5) more transmissible than its oseltamivir-sensitive predecessor and the oseltamivir-resistant pandemic A H1N1 strain that emerged and circulated in Japan during 2013–14 was 24% (95% CrI 17–30) less transmissible than its oseltamivir-sensitive counterpart. We show that in the event of large-scale antiviral interventions during a pandemic with co-circulation of AVS and AVR strains, our method can be used to inform optimal use of antivirals by monitoring intrinsic AVR fitness and drug pressure on the AVS strain. Interpretation: We developed a simple method that can be easily integrated into contemporary influenza surveillance systems to provide reliable estimates of AVR fitness in real time.

Using Epidemiological and Genetic Data to Understand the Spread of Multi Drug-resistant Bacteria in Healthcare Settings
Ben Cooper
The diminishing treatment options for many multi-drug-resistant bacterial pathogens make them one of the most important global health threats. For many of the most concerning pathogens healthcare settings are thought to play a key role in transmission. If the infections caused by these organisms can’t be treated, can we stop them spreading though hospital infection control measures such as antibiotic stewardship, hygiene or isolation measures? Can we learn more about how well different control measures work by combining traditional epidemiological information with sequencing data? Whole genome sequencing (WGS) technology has enabled the study of transmission networks for a number of important pathogens with unprecedented resolution. The last few years have also seen a small epidemic in statistical methods to infer transmission networks using WGS data. In most cases these methods have been motivated by applications to specific pathogens, and many require restrictive assumptions such as a well-defined generation interval, a known time of infection or disease onset, a single sequence per infected individual, transmission bottlenecks of size one, and, in many cases, an epidemic with a single source. Our focus is the analysis of transmission networks of multiply antibiotic-resistant bacteria, particularly in hospital settings. Such pathogens are characterised by largely asymptomatic carriage, with symptomatic disease appearing in only a small proportion of those infected. The generation interval is not well-defined. Lower resolution typing methods have shown that endemic transmission in such settings is often characterised by multiple introductions of common pathogens and the simultaneous propagation of distinct lineages rather than the expansion of a single clone traceable to a known source patient in the sample. Whole genome sequencing studies have confirmed this picture and also highlighted a surprisingly high degree of within-host genetic diversity. In this talk I provide an overview of what we have learnt about the epidemiology of these pathogens from WGS studies, highlight the difficulties in accurately reconstructing transmission networks even with WGS data, discuss statistical methods that have been developed in an attempt to overcome these problems, and illustrate the application of these methods to datasets concerning the nosocomial transmission of multi-drug-resistant bacteria in Southeast Asia. I will also discuss potential problems with inferences which have been made about infection control measures based on WGS data and potential solutions to these problems.

Influenza Immunity and Transmission in Time, Age and Space
Steven Riley
There is a widely accepted standard conceptual model influenza transmission and immunity in humans: new strains emerge from animal reservoirs into a fully immune human population; they spread easily via aerosols and droplets as their hosts make social interactions, causing a pandemic; after which, population immunity forces the strain to continually evolve away from existing antigens. Although not wrong, the obvious mathematical implementations of this model cannot explain many important observations being made with next generation surveillance. In this talk, I will describe some current studies that are attempting to refine the standard model with respect to time, age and space. Within the context of these studies, I will argue that there are substantial scientific and public health gains to be made by developing better conceptual models of influenza transmission and immunity. These conceptual models, when implemented mathematically, need to be capable of both reproducing and forecasting modern digital and biological data streams.
Steven Riley, Professor of Infectious Disease Dynamics at Imperial College London
Steven studies the transmission of human pathogens. He conducts field studies, analyses data and uses mathematical models to look at scientific questions that are relevant to public health: how does our pattern of social contacts affect the transmission of respiratory infections? How much more severe is one strain of influenza than another? How far does influenza penetrate into rural areas after it sweeps through cities?

Predicting Healthcare Demand during Epidemics in France
Simon Cauchemez
In this presentation, I will discuss how mathematical modelling and new types of infectious disease surveillance may enhance our ability to anticipate healthcare demand during infectious disease epidemics. The talk will be illustrated with two case studies, looking respectively at the care of Guillain-Barré Syndromes during a recent Zika epidemic in Martinique, a French island in the Caribbean, and at influenza-like-illness hospitalizations in France.
Simon Cauchemez, Mathematical Modelling of Infectious Diseases Unit, Institut Pasteur, Paris, France

Characterizing the Dynamics Underlying Global Spread of Emerging Infectious Diseases
Lin Wang, Joseph Wu
Global metapopulation epidemic simulations built with worldwide air-transportation data have been the main tool for studying how epidemics spread from the origin to other parts of the world (e.g. for pandemic influenza, SARS, MERS-CoV and Ebola). Despite the long history and widespread use of these models, it remains unclear how disease epidemiology and the air-transportation network structure determine epidemic arrivals for different populations around the globe. In this study, we fill this knowledge gap by developing and validating a novel theory that requires only basic analytics from stochastic processes. We apply this theory retrospectively to the 2009 influenza pandemic and 2014 Ebola epidemic to show that key epidemic parameters could have been robustly estimated in real-time from public data on the local and global spread. Our innovative theory not only elucidates the dynamics underlying global spread of epidemics but also advances our capability in nowcasting and forecasting epidemics.

Using Genetic Data at Multiple Scales to Understand Constraints on Viral Adaptation
Katia Koelle
RNA viruses, such as influenza, dengue, and zika, generally have high mutation rates and large population sizes, enabling these viruses to often times rapidly adapt to new host populations and environments. Yet, several processes also exist that can limit or constrain the ability of these viruses to rapidly adapt. Here, I will discuss two of such processes: transmission bottlenecks between donor hosts and recipient hosts and genetic linkage between viral sites. I will first show recent work from my group that uses influenza A virus deep-sequencing data to estimate transmission bottleneck sizes for this virus by transmission pair. While we find that influenza’s transmission bottleneck is rather loose overall, we also find appreciable variation in estimated bottleneck sizes between transmission pairs that can in part be explained by donor temperature.

This indicates that hosts may differ in the degree of constraints they place on virus evolution. I will then scale up to the population level to argue that genetic linkage in influenza, within and in part between gene segments, acts to constrain viral evolution. I will focus specifically on how genetic linkage can slow influenza’s antigenic evolution in humans and how it can help us understand the virus’s characteristic pattern of relatively rare and punctuated cluster transitions. I end with how the field of phylodynamics has started to consider these important processes, particularly in phylodynamic inference, and future work that remain to be done.
Katia Koelle is an Associate Professor in the Department of Biology at Duke University. She earned her PhD in 2005 from the University of Michigan and then held a Center for Infectious Disease Dynamics post-doctoral fellowship at Penn State prior to joining the faculty at Duke. Her interests include the development of mathematical models to better understand patterns of viral evolution and disease dynamics between and within human hosts. She is further interested in the development and application of statistical approaches to characterize disease spread from viral sequence data. She works primarily on the ‘phylodynamics’ of RNA viruses, most notably influenza and dengue.

Using Multiple Infectious Disease Models for Local Decision Making to Control Neglected Tropical Diseases
Deirdre Hollingsworth
Infectious disease modelling is increasingly being used to inform local-level decision making for the control of tropical diseases in low-income populations. Neglected tropical diseases (NTDs) are a group of infections which predominantly affect the poorest populations in the world. They are responsible for chronic suffering as well as mortality in these hard to reach populations. In recent years, there has been a drive to reduce the burden of these diseases through an international effort to roll out interventions at a global scale.

To improve decision making for disease control at a local scale, we need high resolution spatial models that can be used to predict the outcome of alternative control strategies. Once these models are developed, communication becomes very important, as the knowledge needs to be transferred to audiences, such as policy makers, in an accessible format. Graphical user interfaces (GUIs) are a tool that can be used intuitively by the general public, making modelling outputs more approachable.
T. Déirdre Hollingsworth is deputy director of the Zeeman Institute for Systems Biology and Warwick Infectious Disease Epidemiology Research (SBIDER) which works across mathematics, statistics, medicine and life sciences at the University of Warwick. She leads the neglected tropical disease (NTD) modelling consortium, a large international network of infectious disease epidemiologists working to improve the design of public health interventions to reduce the burden of these diseases. Within NTDs, her research foci are lymphatic filariasis (elephantiasis), visceral leishmaniasis (kala azar), and soil-transmitted helminths.

Professor Hollingsworth studied mathematics and mathematical modelling at the University of Oxford (undergraduate and masters degrees) and the University of Cambridge (PhD) in the UK. She completed her post-doctoral training at Imperial College London, working on influenza, SARS and HIV. She then was awarded a prestigious personal Imperial College Research Fellowship at the Medical Research Council Centre for Outbreak Analysis and Modelling, during which she began her research in NTDs. She was then appointed as an assistant professor in a joint position between the University of Warwick and Liverpool School of Tropical Medicine. She has been awarded a personal professorship at the University of Warwick and holds honorary positions at Liverpool School of Tropical Medicine and Imperial College London.

Cura Te Ipsum: Health Search at Google
Evgeniy Gabrilovich
Approximately 1 percent of all Google searches are symptom-related, as users are conducting online research on pertinent medical conditions. In this talk we will discuss the recently launched symptom search experience on Google. We use machine learning methods to identify queries with condition-seeking intent. We extract relevant health conditions by analyzing the web search results as well as by consulting the Knowledge Graph. Finally, we learn a ranker for ordering the list of relevant conditions, and evaluate the system performance with medical doctors.
Dr. Evgeniy Gabrilovich is a senior staff research scientist at Google, where he works on improving healthcare. Prior to joining Google in 2012, he was a director of research and head of the natural language processing and information retrieval group at Yahoo! Research. Evgeniy is an ACM Distinguished Scientist, and is a recipient of the 2014 IJCAI-JAIR Best Paper Prize. He is also a recipient of the 2010 Karen Sparck Jones Award for his contributions to natural language processing and information retrieval. Evgeniy has served as a program chair for WWW 2017 and WSDM 2015. He earned his PhD in computer science from the Technion - Israel Institute of Technology. Recently, he graduated (with extra credit) from the Executive MD training program at Harvard Medical School.

Towards Digital Pathogen Surveillance: A Very Bumpy and Winding Road
Jennifer Gardy
Genomic epidemiology – using pathogen genome sequences to understand outbreak/epidemic dynamics – is an exciting new field. The recent Ebola and Zika epidemics have taken viral genomic epidemiology to new places, while work in the bacterial domain, including several foundational papers from our group, has shown that reconstructing transmission for pathogens like TB is not only possible, but that it can influence public health practice in real time. As a community, we have a tremendous opportunity before us – combining genomic epidemiology with digital epidemiology and the OneHealth ethos to create a truly digital model for global pathogen surveillance – but realizing this vision won’t happen unless we overcome some substantial barriers, both technical and societal. In this talk, I’ll review some of the exciting proof-of-concept work in genomic epidemiology from our lab and others, and highlight what I believe to be some of the critical issues that we must address if we are to operationalize genomic epidemiology.
Dr. Jennifer Gardy holds the Canada Research Chair in Public Health Genomics at the University of British Columbia in Vancouver, Canada, and she is also the Senior Scientist, Genomics at the British Columbia Centre for Disease Control. Her laboratory uses genome sequencing as a tool to understand pathogen transmission patterns and, in turn, develop new public health policies. She is a member of the National Academies of Science, Engineering, and Medicine’s Forum on Microbial Threats, a Michael Smith Foundation for Health Research Scholar, and a Senior Editor at Microbial Genomics.

Digital Health & Epidemiology
Marcel Salathe
Online, mobile, global – the ongoing digital revolution affects all aspects of life. Massive amounts of data are now shared by billions of people around the globe through mobile phones, social media services, and other outlets, on any issue imaginable, including issues of health. These data sources can be mined for epidemiological purposes, giving rise to digital epidemiology. Of equal importance, but less discussed, is the fact that these large data sets (big data) provide the raw material for new machine learning algorithms to train on (e.g., "deep learning"), resulting in software that in various domains is close to achieving, or already has achieved, human performance. As human expertise, specifically also white collar expertise, can increasingly be replaced by artificial intelligence, a huge disruptive potential will be unleashed. The health domain in particular will be deeply affected. In this seminar, I will discuss opportunities and challenges in these turbulent times.
Marcel Salathé is a digital epidemiologist working at the interface of population biology, computational sciences, and the social sciences. He obtained his PhD at ETH Zurich and spent two years as a postdoc in Stanford before joining the faculty at Penn State in 2010 at the Center for Infectious Disease Dynamics. In 2014, he spent half a year at Stanford as visiting assistant professor. In the summer of 2015, Marcel became an Associate Professor at EPFL where he heads the Digital Epidemiology Lab at the new Campus Biotech. In 2016, he has also been appointed Academic Director of EPFL Extension School, whose mission is to provide high quality online education in digital technology.

He published numerous papers in the biological, medical, and computational fields, and wrote a book called "Nature, in Code". He led the development of the MOOC “Epidemics - The Dynamics of Infectious Disease”, a popular large-scale online course and has just recently launched a new EPFL MOOC « Nature, in Code: Biology in JavaScript. » He’s the co-founder of PlantVillage, a knowledge exchange platform on crop diseases, and the founder of opendfood.ch, an open food data API designed to foster an ecosystem of applications around food and nutrition data. He also founded CrowdAI, an open data challenge platform whose goal is to accelerate research on big data across multiple scientific domains. He is also Deputy Editor of PLOS Computational Biology, and Editor at EPJ Data Science.

Marcel spent a few years in the tech industry as web application developer. He was part of the renowned Y Combinator startup accelerator’s class of Winter 2014.

Data and Decision- Bridging Research and Health Policy
Muhammad Mamdani
Recent advances in technology have ushered in the era of computational medicine. Large health data repositories are increasingly prominent and advanced analytical approaches such as machine learning are growing in popularity. However, there often remains a gap between the data, the analyses, and consequent actions that affect the patients we serve. This session will present experiences, challenges, and successes in bridging the knowledge-to-action gap when using large health data repositories to influence population and individual level health.
Muhammad Mamdani, PharmD, MA, MPH
Director – Li Ka Shing Centre for Healthcare Analytics Research and Training (CHART)
Li Ka Shing Knowledge Institute of St. Michael’s Hospital; Professor, University of Toronto

Dr. Mamdani is the Director of the Li Ka Shing Centre for Healthcare Analytics Research and Training (CHART) of the Li Ka Shing Knowledge Institute of St. Michael’s Hospital in Toronto. He is also Professor in the Leslie Dan Faculty of Pharmacy, the Department of Medicine of the Faculty of Medicine, and the Institute of Health Policy, Management and Evaluation of the Dalla Lana Faculty of Public Health. He is also adjunct Senior Scientist at the Institute for Clinical Evaluative Sciences (ICES). Dr. Mamdani also is a member of the Human Drug Advisory Panel of the Patented Medicine Prices Review Board (PMPRB) and is a co-Principal Investigator of the Ontario Drug Policy Research Network. In 2010, Dr. Mamdani was named among Canada’s Top 40 under 40. Prior to joining the Li Ka Shing Knowledge Institute and St. Michael’s Hospital, Dr. Mamdani was a Director of Outcomes Research at Pfizer Global Pharmaceuticals in New York. Dr. Mamdani’s research interests include pharmacoepidemiology, pharmacoeconomics, and drug policy. He has published approximately 400 research studies in peer-reviewed medical journals, including leading journals such as the New England Journal of Medicine, the Lancet, the Journal of the American Medical Association, the British Medical Journal, and Annals of Internal Medicine.

Dr. Mamdani obtained a Doctor of Pharmacy degree (PharmD) from the University of Michigan (Ann Arbor) in 1995 and subsequently completed a fellowship in pharmacoeconomics and outcomes research at the Detroit Medical Center in 1997. During his fellowship, Dr. Mamdani obtained a Master of Arts degree in Economics from Wayne State University in Detroit, Michigan. He then completed a Master of Public Health degree from Harvard University in 1998 with a concentration in quantitative methods, focusing on biostatistics and epidemiological principles.

Analysis of Emerging Epidemics in the Era of Real-time Pathogen Genomics
Oliver Pybus
Pathogen genome sequences contain a remarkable amount of information about epidemiological processes. With appropriate analysis, they can reveal where and when an outbreak initiated, estimate transmission rates, quantify routes and rates of spatial spread, and inform studies of pathogenicity. Yet the contribution that genomics can make to infectious disease surveillance and outbreak control is only beginning to be appreciated by public health agencies. Faster, cheaper and more portable sequencing technologies mean that genomics can now take place alongside field epidemiology investigations. I will outline the opportunities and challenges ahead as we try to formally integrate genomic, spatial, and epidemiological data. Can the culture of openly sharing influenza virus genomes be replicated by other research communities? I will present results from recent epidemics, such as highly pathogenic H5N8 avian influenza virus in Europe, Ebola virus in west Africa, and Zika virus in the Americas. I’ll also introduce the ZiBRA project, a mobile sequencing lab that travelled across north-east Brazil to study Zika virus.
Oliver Pybus is Professor of Evolution & Infectious Disease at the University of Oxford, UK and a Professorial Fellow of New College, Oxford. He is a Principal Investigator of the Oxford Martin School for Emerging Infections and Chief Editor of the journal Virus Evolution. He investigates the evolution, epidemiology, and phylodynamics of infectious diseases, particularly pathogenic RNA viruses. His research includes the development of new techniques for inferring population processes from genetic sequence data.

Metagenomics Study of Individuals with Seasonal Influenza
Yinpeng A. Wang, Benjamin Cowling, Joseph Wu, Malik Peiris, Herbert Pang
Background: Advancement of metagenomic technologies has given us an unprecedented opportunity for understanding complex infectious diseases like influenza. The research community has recently embarked on studying microbiome and its relation with viral infection. The objective of this study is to understand the underlying microbiome profiles of patients of seasonal influenza type A. Methods: Nasal swabs were collected from ten healthy subjects at baseline. These subjects might become infected at the second time point. Each sample was subjected to whole-genome shotgun sequencing. The microbiota of subjects who remained healthy at time point two were compared with those who got infected. Results: At the second visit, five subjects remained healthy and five got infected. The nasal bacterial communities were dominated by Actinobacteria, Bacteroidetes, Firmicutes, and Proteobacteria. We observed that the change of the abundance of Proteobacteria in infected subjects were significantly different from healthy subjects. Summary: Our study represents the first survey of the nasal microbiota before and during seasonal influenza infection.

Harnessing Multinational Data Sources in Global Models to Inform Vaccine Policy
Mark Jit
Global multilateral organisations such as the World Health Organization and Gavi, the Vaccine Alliance, make policy recommendations and funding decisions around vaccination and other global health investments. These decisions usually take into account the needs of a range of countries in different geographical regions and with socioeconomic and cultural situations. This has given rise to a generation of “global models” which seek to be representative of a range of countries. Such models have limited ability to take into account special data sources of particular countries, but instead harness large global datasets such as the Global Health Observatory (from WHO), World Development Indicators (from the World Bank) and Demographic and Health Surveys. Conducting these global analyses gives rise to unique opportunities and pitfalls. These will be discussed in the talk, using examples from “global models” of human papillomavirus, pneumococcal, measles and dengue vaccines.
Mark Jit is professor of vaccine epidemiology at the London School of Hygiene & Tropical Medicine (LSHTM) and senior scientist in the Modelling and Economics Unit of Public Health England (PHE). His research focuses on epidemiological and economic modelling of infectious disease control interventions such as vaccination, to support evidence-based public health decision making. He has published over 100 papers covering a range of vaccine antigens including measles, HPV, pneumococcus, rotavirus, influenza, dengue and EV71, as well as methodological papers advancing the ways vaccines are evaluated. This work has influenced many of the major changes to immunisation policy in the UK and globally over the past ten years.

Quantifying the global variation in risk of influenza virus emergence
Colin Russell
Avian and swine influenza viruses circulate worldwide and even moderately-sized animal populations are capable of harbouring viruses that threaten human and animal health. For over 20 years, influenza pandemic research and surveillance has largely focused on avian H5 and H7 subtype viruses due to their ability to develop high pathogenicity phenotypes and the high mortality rates of human infections with these viruses. Despite all of this attention, an H1N1 virus originating from swine in Mexico caused the most recent human influenza pandemic and there is no evidence that viruses of the H5 or H7 subtype will be the etiological agent of the next influenza pandemic. The design of comprehensive, risk-based influenza virus surveillance strategies is hampered by a lack of information on the global variation of risk of virus emergence and the geographic distribution of current surveillance efforts. This talk describe the development od influenza virus subtype-agnostic estimates of the global variation in risk of virus emergence from poultry and swine based on the distribution of human and animal populations and factors likely to mediate their interactions. These estimates combined with a global model of onward transmission risk can be used to identify areas in greatest need of human and animal influenza surveillance. Systematically comparing these risk estimates to current surveillance efforts reveals that poultry surveillance is concentrated in high-risk areas. However, for every high-risk area with multiple years of poultry surveillance there are ~3 areas of similarly high risk with no evidence for surveillance. For swine, this ratio of areas with surveillance to areas at risk is ~1:10.
Colin Russell is an evolutionary biologist at the University of Cambridge. He received his undergraduate degree in biology from Emory University in Atlanta, USA (2001) and his PhD from University of Cambridge (2006). After finishing his PhD he worked as postdoc with Derek Smith from 2006-2008 and subsequently received an NIH Fogarty Guest Fellowship (2007), a Junior Research Fellowship from Clare College Cambridge (2008), and a Royal Society University Research Fellowship (2009) – all held at the University of Cambridge. In 2013, he joined the faculty of the Cambridge Department of Veterinary Medicine as a Reader. Beyond his basic research on influenza virus evolution, Colin has worked extensively with the World Health Organisation on both pandemic preparedness and influenza vaccine strain selection, including serving on the WHO Influenza Vaccine Strain Selection Committee from 2009-2015. He is also one of the chief scientists of the Cambridge World Health Organization Collaborating Center for Modelling, Evolution and Control of Emerging Infectious Diseases. In October 2017 he will join the faculty of the Academic Medical Center of the University of Amsterdam as Professor and Head of the Laboratory of Applied Evolutionary Biology in the Department of Medical Microbiology.

Precision Public Health
Scott F. Dowell
By more precisely defining the geospatial distribution of disease, incorporating whole genome sequencing of pathogens into epidemic responses, and pathologically confirming causes of death in mortality surveillance, the practice of public health is made more efficient and more effective. By analogy with precision medicine, which aims to bring the right treatment to the right patient at the right time to minimize side effects and maximize efficacy, precision public health aims to bring the right intervention to the right population. Of importance to the Bill and Melinda Gates foundation such approaches also promote equity, by making the best of public health available to more people, including those in poor or marginalized areas. We are enthusiastic about the rapid advances in disease modeling, phylodynamics, and digital disease detection and the potential opportunities they open for more impactful and equitable public health programs.
Scott F. Dowell, a pediatric infectious disease specialist by training, now focuses on tracking the causes of global childhood mortality for the Bill and Melinda Gates Foundation. He joined the foundation in 2014 after 21 years at the U.S. Centers for Disease Control and Prevention (CDC), where he studied viral and bacterial pneumonia, responded to outbreaks of Ebola and other pathogens, and established global disease surveillance and outbreak response programs. From 2001 to 2005 he established and directed the International Emerging Infections Program in Thailand, a collaboration between the CDC and the Thai Ministry of Public Health. The program received accolades from both the Thai and U.S. governments for its prominent role in responding to the SARS crisis, and for its leadership in defining the response to avian influenza A (H5N1) in Southeast Asia. Building on the success of the Thailand work on emerging infections, Dr. Dowell returned to Atlanta to help develop the Global Disease Detection program, which became CDC’s principal means of identifying and containing emerging infections around the world, with GDD Regional Centers in 10 countries. In 2009 the program was formally recognized by the World Health Organization as a Collaborating Center and Dr. Dowell was named as its first director. He led CDC’s response to the earthquake and cholera epidemic in Haiti during 2010-11, helping to rebuild the public health infrastructure and contributing to the saving of an estimated 7,000 lives. Dr. Dowell served as the Director of the Division of Global Disease Detection and Emergency Response from 2009-12, and led the agency’s Global Health Security Agenda from 2012-14. In 2014 he retired from the US Public Health Service at the rank of Rear Admiral and Assistant Surgeon General. Dr. Dowell has co-authored more than 170 publications and has a special interest in targeted reductions in childhood mortality.

Climatic Factors Driving the Transmission of Human Influenza and Other Respiratory Viruses
Tommy Lam
Communicable human respiratory diseases are caused by a range of fast evolving RNA viruses including influenza A (FluA) and B (FluB) virus, respiratory syncytial virus (RSV) and parainfluenza virus (PIV), etc. Previous studies have shown that some climatic factors may play roles in driving the disease transmission and prevalence. In this study, we used statistical models to investigate the climatic drivers for FluA, FluB, RSV and PIV prevalence among various geographical locations (in North America, Europe, Middle East, Africa, Asia and Oceania), collected through a global research network. We also analysed the available gene sequences of these RNA viruses to gain insights into their global transmission.

Identifying Artemisinin Resistance from Parasite Clearance Half-life Data with A Shiny Web Application
Sai Thein Than Tun
Malaria elimination is under threat due to the emergence of parasites resistant to artemisinin derivatives, the most efficacious drugs against Plasmodium falciparum malaria. Novel tools for supporting the surveillance of artemisinin resistance are critical for malaria control and elimination strategies. We developed a modelling tool to analyze data on parasite half-life as distributions of artemisinin-sensitive and artemisinin-resistant populations, which could help in the surveillance of artemisinin resistance. The model itself was validated using parasite half-life data from studies on the Thai-Myanmar border and in Western Cambodia. The tool can be accessed at bit.ly/id_artemisinin_resistance. The model can be used to analyze parasite clearance half-life data after treatment with artemisinin derivatives and could also be used alongside other methods such as the identification of molecular markers.

High-resolution Contact Network Measurements in Developing Country Settings
Ciro Cattuto
Personal electronic devices and wearable sensors are regarded as an increasingly important data source for digital epidemiology. This talk will focus on the technical challenges and preliminary results of ongoing studies that deploy wearable proximity sensors to map human and animal contact networks in low resource settings, such as rural households and schools in developing countries. I will illustrate the general features of time-resolved contact network data, discuss sampling, and show the results of comparisons with other methods for measuring contact networks.

Synthesis of Pathogen Genomic Sequencing Data and Clinical Records for Efficient Control of Nosocomial Infections
Rosemarie Sadsad, Matthew OSullivan, Vitali Sintchenko, Suyin Hor, Mary Wyer, Lyn Gilbert
Prospective genomic-based surveillance can revolutionise the management and control of infectious diseases, however challenges in its implementation have limited its routine use in clinical settings. We demonstrate the feasibility for implementing a prospective genomic-based surveillance system for multidrug-resistant hospital acquired infections in a 975-bed tertiary public hospital in Sydney, Australia. We have integrated different ‘Big Data’ sources including pathogen molecular typing and whole genome sequencing (WGS) data, electronic health records, and pathology records and have applied automatic cluster detection models to this genetic, clinical, and epidemiological data to rapidly detect cases of nosocomial infection and to help identify sources of transmission. This information will soon be reported to clinicians and infection control practitioners through a clinical web-based application. We describe how this information can be used to detect, describe, and manage outbreaks by using the example of a highly prevalent methicillin-resistant Staphylococcus aureus (MRSA) sequence type (ST) 239 outbreak that occurred in a surgical ward over a one year period. We found MRSA-ST239 was transmitted by multiple point sources and also propagated between patients and surfaces. Our results suggest the outbreak was persistent, but fluctuated throughout the study period during which several infection control interventions were implemented. 63 of 112 isolates contained a type II arginine catabolic mobile element (ACME II), which is linked to fitness and colonisation and reported to be emerging in ST239 strains. Prospective genomic-based surveillance can rapidly detect suspected nosocomial outbreaks and be used to inform and evaluate infection control interventions in high endemic settings. In future, (almost) real-time digital surveillance that feeds back to clinicians should lead to interventions that will interrupt or reduce transmission more effectively. Our next steps are to implement real-time notification of transmission events to clinicians caring for patients and use video-reflexive ethnography to evaluate the utility of this new information for improving patient care and infection prevention and control practice.