My research is problem driven – I look for solutions to large-scale problems in public health
I strive to develop impactful and implementable solutions to current challenges in public health by utilizing methods from applied probability, network science, and operations research. This is done in collaboration with problem experts and practitioners.
I develop mechanistic models to derive decision and policy insights
Using observation and logic to understand the essential physics of a problem, I postulate mechanistic, and often probabilistic, models of the process. Data is used to calibrate or evaluate, not create, the models.
Click here for a list of my publications and presentations and learn about my current research focus below
Current research focus
My current research focuses on developing models of transmission dynamics in order to better understand how large-scale outbreaks begin, how to quickly identify the source, and how to stop the spread. The questions I am researching can be categorized into two primary areas: i) what is the source location and transmission process, and ii) what are tactical interventions to mitigate the spread.
Click on each research question to learn more
Outbreaks on Networks: Understanding contagion transmission processes and identifying source location
- How can we identify the source of an outbreak of foodborne disease with greater accuracy, certainty, and speed?
The objective of this research is to develop analytical methods to rapidly identify the source large-scale outbreaks of foodborne disease. We have developed a novel network theoretical method for source identification that takes into account the unique complexities of contagion transmission process across food supply networks. This method introduces a random-walk model of contamination transmission and formulates a maximum likelihood estimator for the source location. In evaluation on synthetic networks and real data from a recent outbreak this method has been able to significantly outperform existing methods for source identification in networks. We are currently working with outbreak investigators at the Federal Institute for Risk Assessment (BfR) in Berlin to transfer our models and computational tools for further testing and adaptation to available data in Germany, for their ultimate use in practice during future outbreaks.This was the focus of my doctoral thesisMaterials: Network Source Detection Presentation; Code on Github; Visualization (Created by Elena Polozova)
Locating the source of large-scale outbreaks of foodborne diseaseA. Horn, H. Friedrich (2017). Paper in preparation, draft available upon request
- When does knowing the time evolution of illness data contribute to the ability to identify the outbreak source?
During an large-scale outbreak, data on the time of reported illnesses together with potential transmission times of contagions along a network can in theory provide a signal to investigators. However there are multiple sources of uncertainty in the temporal dimension of how contamination spreads in a network, including the uncertainty accumulated in transmission time between any two nodes, as well the inherent inaccuracy in patient reporting of illness. This research evaluates if and when temporal data might provide meaningful insight for identifying the outbreak source if used in practice. We have developed a method for estimating the outbreak source location and its initiation time in high temporal variance settings, based on a Gaussian diffusion model. We are working on a paper to determine the outbreak scenarios — for example, different food items and pathogen combinations — where the signal from time is great enough to contribute a value for source identification.Materials:Source detection in networks weighted by volume and temporal dimensionsA. Horn, E. Polozova, H. Friedrich (2017). Paper in preparation
- How does network structure determine the ability to identify the source of an outbreak?
While recent work has focused the role of network structure on propagation dynamics, its impact on the ability to identify the propagation source has received less attention. This work develops a novel metric, network traceability entropy (NTE), to measure the intrinsic ability of a network structure to support traceability. We show that this measure, based on the information theoretic definition of entropy, is an efficient means of encoding and transmitting the uncertainty in the source identification problem. In the paper, we demonstrate how NTE can be used as a tool to systematically compare the traceability of various network configurations and yield insights into the influential role of specific parameters. The proposed measure opens possibilities to quantify the traceability of any network involving a diffusion process and is useful in network design or optimization applications where traceability is desirable.Materials:A Universal Measure for Network TraceabilityX. Liu*, A. Horn*, J. Su, J. Jiang. (2017). Management Science, under review
- What is the role of human transportation networks in cholera epidemics?
Cholera is normally thought to spread through environmentally-mediated means, such as contaminated water systems or sanitation infrastructure. However recent outbreak data from Mozambique demonstrates that the infection might be spread through human travel as well. Working with researchers at HealthMap, we are determining whether transport networks in combination with stochastic transmission models of the disease spreading process can explain the data better than environmentally-mediated hypotheses. Understanding this will help us to predict high-risk locations for future outbreaks, identify transmission sources, and direct resources to prevent or stop the spread.
- What are possible methods for identifying food item source of an emerging outbreak?
During the onset of an outbreak of foodborne disease, identifying which food type is causing the outbreak is often the bottleneck in investigations and mitigation processes. In collaboration with the German Federal Institute for Risk Assessment (BfR) and Professor Hanno Friedrich at Kühne Logistics University, we are developing a solution by adapting the source identification methodology into a food-item-network model selection problem.
Outbreak Mitigation & Response: Network based interventions to limit the spread
- During an outbreak, what dynamic strategies can investigators take to maximize the probability of identifying the source?
During the onset of a large-scale outbreak, different response actions can be taken to locate of the source, which can then help mitigate the spread of disease. A major area of research deals with developing network-based interventions to detect that an outbreak is occurring, as well as what actions should be taken actions to limit the spread of the contamination. Much less work has focused on developing interventions with the aim of identifying the outbreak source, especially in a dynamic framework (i.e. in response to an ongoing outbreak).We aim to contribute to the literature on network-based interventions to outbreaks by researching two main problems regarding dynamic response:Where should investigators points be placed in a network in order to optimally identify the outbreak source, given limited investigation resources?When should these investigators be dispatched, aiming to minimize total number of illnesses?These questions are interrelated and we are developing an integrated framework to solve them. Investigators have limited resources and how to best deploy them is complex: The sooner a probable source can be investigated, if successful, the more infections can be averted. At the same time, assessments made at an early stage in an outbreak’s progression will be marked by greater uncertainty.Since the appropriate solution will depend on the network and disease transmission setting, the ultimate goal is to have suite of tactical response strategies for different outbreak scenarios.
- How can network structure be adapted to optimally facilitate traceability?
Network based interventions to identify the source of an outbreak can be tactical – sending out investigators to observe the disease, or structural – altering the network topology. The goal of this research is to identify small changes that can be made proactively to a network’s structure that will improve the ability to identify the source in the event of an outbreak. Building on work to study how network structure influences source detection, this research has three main goals: (i) identifying network features that can, in practice, be modulated, (ii) determining what combinations of those features have the greatest role in determining the probability of source identification, and (iii) specifying how to alter those features to increase the probability of source detection, subject to realistic constraints regarding the cost or feasibility. The network features that can be altered and thus the network design strategies will depend on the problem context. As above, we aim to develop a suite of solutions that are specific to different outbreak settings.
This work also has implications for supply chain and digital network design for traceability.
Case Research: Specific public health cases
- Is it possible to determine the prognosis of critically ill sepsis patients using data from medical records?
This work applied machine learning methods to construct a classification model using patient data on critically ill sepsis patients. It was part of a larger collaboration between MIT, Harvard, and Instituto Superior Técnico (IST) in Lisbon.Multi-Objective Performance Evaluation Using Fuzzy Criteria: Increasing Sensitivity Prediction for Outcome of Septic Shock PatientsAbigail L. Horn, Federico Cismondi, André S. Fialho, Susana M. Vieira, João M.C. Sousa, Shane Reti, Michael Howell, Stan FinkelsteinMulti-stage modeling using fuzzy multi-criteria feature selection to improve survival prediction of ICU septic shock patientsFederico Cismondi, Abigail L. Horn, André S. Fialho, Susana M.Vieira, Shane R. Reti, João M.C. Sousa, Stan Finkelstein