AI Techniques for Forecasting Epidemic Dynamics: Theory and Practice

Forecasting epidemic dynamics has been an active area of research for at least two decades. The importance of the topic is evident: policy makers, citizens, and scientists would all like to get accurate and timely forecasts. In contrast to physical systems, the co-evolution of epidemics, individual and collective behavior, viral dynamics, and public policies make epidemic forecasting a problematic task. The situation is even more challenging during a pandemic as has become amply clear during the ongoing COVID-19 pandemic. Researchers worldwide have put in extraordinary efforts to try to forecast the time-varying evolution of the pandemic;despite their best efforts, it is fair to say that the results have been mixed. Several teams have done well on average but failed to forecast upsurges in the cases. In this chapter, we describe the state-of-the-art in epidemic forecasting, with a particular emphasis on forecasting during an ongoing pandemic. We describe a range of methods that have been developed and discuss the experience of our team in this context. We also summarize several challenges in producing accurate and timely forecasts. © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022.

Data-driven scalable pipeline using national agent-based models for real-time pandemic response and decision support

This paper describes an integrated, data-driven operational pipeline based on national agent-based models to support federal and state-level pandemic planning and response. The pipeline consists of (i) an automatic semantic-aware scheduling method that coordinates jobs across two separate high performance computing systems;(ii) a data pipeline to collect, integrate and organize national and county-level disaggregated data for initialization and post-simulation analysis;(iii) a digital twin of national social contact networks made up of 288 Million individuals and 12.6 Billion time-varying interactions covering the US states and DC;(iv) an extension of a parallel agent-based simulation model to study epidemic dynamics and associated interventions. This pipeline can run 400 replicates of national runs in less than 33 h, and reduces the need for human intervention, resulting in faster turnaround times and higher reliability and accuracy of the results. Scientifically, the work has led to significant advances in real-time epidemic sciences. © The Author(s) 2022.

Controlling Epidemic Spread using Probabilistic Diffusion Models on Networks

The spread of an epidemic is often modeled by an SIR random process on a social network graph. The MININFEDGE problem for optimal social distancing involves minimizing the expected number of infections, when we are allowed to break at most B edges;similarly the MININFNODE problem involves removing at most B vertices. These are fundamental problems in epidemiology and network science. While a number of heuristics have been considered, the complexity of these problems remains generally open. In this paper, we present two bicriteria approximation algorithms for MININFEDGE, which give the first non-trivial approximations for this problem. The first is based on the cut sparsification result of Karger (1999), and works when the transmission probabilities are not too small. The second is a Sample Average Approximation (SAA) based algorithm, which we analyze for the Chung-Lu random graph model. We also extend some of our results to tackle the MININFNODE problem.

epiDAMIK 5.0: The 5th International Workshop on Epidemiology meets Data Mining and Knowledge Discovery

Similar to previous iterations, the epiDAMIK@KDD workshop is a forum to promote data driven approaches in epidemiology and public health research. Even after the devastating impact of COVID-19 pandemic, data driven approaches are not as widely studied in epidemiology, as they are in other spaces. We aim to promote and raise the profile of the emerging research area of data-driven and computational epidemiology, and create a venue for presenting state-of-the-art and in-progress results-in particular, results that would otherwise be difficult to present at a major data mining conference, including lessons learnt in the 'trenches'. The current COVID-19 pandemic has only showcased the urgency and importance of this area. Our target audience consists of data mining and machine learning researchers from both academia and industry who are interested in epidemiological and public-health applications of their work, and practitioners from the areas of mathematical epidemiology and public health. Homepage: https://epidamik.github.io/. © 2022 Owner/Author.

Effective Social Network-Based Allocation of COVID-19 Vaccines

We study allocation of COVID-19 vaccines to individuals based on the structural properties of their underlying social contact network. Using a realistic representation of a social contact network for the Commonwealth of Virginia, we study how a limited number of vaccine doses can be strategically distributed to individuals to reduce the overall burden of the pandemic. We show that allocation of vaccines based on individuals' degree (number of social contacts) and total social proximity time is significantly more effective than the usually used age-based allocation strategy in reducing the number of infections, hospitalizations and deaths. The overall strategy is robust even: (i) if the social contacts are not estimated correctly;(ii) if the vaccine efficacy is lower than expected or only a single dose is given;(iii) if there is a delay in vaccine production and deployment;and (iv) whether or not non-pharmaceutical interventions continue as vaccines are deployed. For reasons of implementability, we have used degree, which is a simple structural measure and can be easily estimated using several methods, including the digital technology available today. These results are significant, especially for resource-poor countries, where vaccines are less available, have lower efficacy, and are more slowly distributed. © 2022 Owner/Author.

Deploying Vaccine Distribution Sites for Improved Accessibility and Equity to Support Pandemic Response

In response to COVID-19, many countries have mandated social distancing and banned large group gatherings in order to slow down the spread of SARS-CoV-2. These social interventions along with vaccines remain the best way forward to reduce the spread of SARS CoV-2. In order to increase vaccine accessibility, states such as Virginia have deployed mobile vaccination centers to distribute vaccines across the state. When choosing where to place these sites, there are two important factors to take into account: accessibility and equity. We formulate a combinatorial problem that captures these factors and then develop efficient algorithms with theoretical guarantees on both of these aspects. Furthermore, we study the inherent hardness of the problem, and demonstrate strong impossibility results. Finally, we run computational experiments on real-world data to show the efficacy of our methods. © 2022 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved

Theoretical Models and Preliminary Results for Contact Tracing and Isolation

Efficient contact tracing and isolation is an effective strategy to control epidemics, as seen in the Ebola epidemic and COVID-19 pandemic. An important consideration in contact tracing is the budget on the number of individuals asked to quarantine-the budget is limited for socioeconomic reasons (e.g., having a limited number of contact tracers). Here, we present a Markov Decision Process (MDP) framework to formulate the problem of using contact tracing to reduce the size of an outbreak while limiting the number of people quarantined. We formulate each step of the MDP as a combinatorial problem, MinExposed, which we demonstrate is NP-Hard. Next, we develop two approximation algorithms, one based on rounding the solutions of a linear program and another (greedy algorithm) based on choosing nodes with a high (weighted) degree. A key feature of the greedy algorithm is that it does not need complete information of the underlying social contact network, making it implementable in practice. Using simulations over realistic networks, we show how the algorithms can help in bending the epidemic curve with a limited number of isolated individuals. © 2022 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved.

Drivers and Predictors of COVID-19 Vaccine Hesitancy in Virginia

This research uses the COVID-19 Trends and Impact Survey provided by Carnegie Mellon University in partnership with Facebook to study predictors and drivers of COVID-19 vaccine hesitancy in Virginia’s adult population. It estimates vaccine hesitancy rates at a zip code level in Virginia by applying multilevel statistical models. Our analysis identifies the demographic features of zip codes that are associated with vaccine hesitancy. It also examines the drivers of COVID-19 vaccine hesitancy across Virginia. Results show the presence of a larger percentage of Black and White population and a lower percentage of Hispanic population are predictors of higher vaccine hesitancy within a zip code in Virginia. Among these drivers, the biggest is system distrust, where individuals either do not trust the government or believe that the vaccine is not efficacious. Finally, it provides policy insights and tailored outreach programs for improving COVID-19 vaccination acceptability in different regions in Virginia. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Deciphering COVID-19-Associated Effects on Hospital MRSA Transmission and Social Networks

Background. The COVID-19 pandemic was associated with a significant (28%) reduction of methicillin-resistant Staphylococcus aureus (MRSA) acquisition at UVA Hospital (P=0.016). This "natural experiment" allowed us to analyze 3 key mechanisms by which the pandemic may have influenced nosocomial transmission: 1) enhanced infection control measures (i.e., barrier precautions and hand hygiene), 2) patient-level risk factors, and 3) networks of healthcare personnel (HCP)-mediated contacts. Hospital MRSA acquisition was defined as a new clinical or surveillance positive in patients with prior unknown or negative MRSA status occurring >72h after admission. 10 month time periods pre- (5/6/2019 to 2/23/2020) and post-COVID-19 (5/4/2020 to 2/28/2021) were chosen to mitigate the effects of seasonality. A 6-week wash-in period was utilized coinciding with the onset of several major hospital-wide infection control measures (opening of 2 special pathogen units with universal contact/airborne precautions on 4/1/21 and 5/1/21, universal mask 4/10/21 and eye protection 4/20/20 policies instituted along with staff education efforts including the importance of standard precautions). Box and whisker plots depict quartile ranges, median (dotted line), and mean values. Mean MRSA acquisition rates pre- (0.92 events per 1,000 patient days) significantly declined post-COVD-19 (to 0.66;P=0.016). Independent-samples t tests were used (2-tailed) for statistical comparisons except for variables without a normal distribution (Shorr Scores), for which a Mann-Whitney U test was used. Methods. Census-adjusted hospital-acquired MRSA acquisition events were analyzed over 10 months pre- (5/6/2019 to 2/23/2020) and post-COVD-19 (5/4/2020 to 2/28/2021), with a 6-week wash-in period coinciding with hospital-wide intensification of infection control measures (e.g., universal masking). HCP hand hygiene compliance rates were examined to reflect adherence to infection control practices. To examine impacts of non-infection control measures on MRSA transmission, we analyzed pre/post-COVD-19 differences in individual risk profiles for MRSA acquisition as well as a broad suite of properties of the hospital social network using person-location and person-person interactions inferred from the electronic medical record. Figure 2. Social Network Construction We constructed a contact network of hospitalized patients and staff at University of Virginia Hospital to analyze the properties of both person-location and person-person networks and their changes pre- and post-COVID-19. Colocation data (inferred from shared patient rooms and healthcare personnel (HCP)-patient interactions recorded in the electronic health record, e.g., medication administration) were used to construct contact networks, with nodes representing patients and HCP, and edges representing contacts. The above schematic shows how the temporal networks are inferred. In the figure, circles represent patients and the small filled squares represent HCP, while the larger rectangles represent patient rooms. The first room is a shared room with two patients. At each time step, co-location is inferred from the EMR data, which specifies interactions between HCP and patients. This can be represented as the temporal network (t) at the bottom. Results. Hand hygiene compliance significantly improved post-COVD-19, in parallel with other infection control measures. Patient Shorr Scores (an index of individual MRSA risk) were statistically similar pre-/post-COVD-19. Analysis of various network properties demonstrated no trends to suggest a reduced outbreak threshold post-COVD-19. Figure 3. Hand Hygiene Compliance Rates Analysis of hospital-wide hand hygiene auditing data (anonymous auditors deployed to various units across UVA Hospital with an average 1,710 observations per month (range 340 - 7,187)) demonstrated a statistically significant (6%) improvement in average monthly hand hygiene compliance (86.9% pre- versus 93.1% post-COVD-19;P=0.008). Figure 4. Individual MRSA Risk Factors We calculated the Shorr Score (a validated tool to estimate individual risk for MRSA carriage in hospitalized patients;Shorr et al. Arch Intern Med. 2008;168(20):2205-10) for patients using data from the electronic health record to test the hypothesis that individual risk factors in aggregate did not change significantly in the post-COVD-19 period to explain changes in MRSA acquisition. Values for this score ranged from 0 to 10 with the following criteria: recent hospitalization (4), nursing home residence (3), hemodialysis (2), ICU admission (1). Pictured are frequency distributions of Shorr scores in the pre-COVID-19 and post-COVID-19 periods. The Mann-Whitney effect size (E), 0.53 (P=0.51), indicated that pre- and post-COVD-19 distributions were very similar. We analyzed three major types of network properties for this analysis: (1) Node properties of the pre- and post-COVID-19 networks consisted of all the edges in the pre- and post-COVID-19 periods, respectively. We considered a number of standard properties used in social network analysis to quantify opportunities for patient-patient transmission: degree centrality (links held by each node), betweenness centrality (times each node acts as the shortest 'bridge' between two other nodes), closeness centrality (how close each node is to other nodes in network), Eigenvector centrality (node's relative influence on the network), and clustering coefficient (degree to which nodes cluster together) in the first five panels (left to right, top to bottom);(Newman, Networks: An Introduction, 2010). Each panel shows the frequency distributions of these properties. These properties generally did not have a normal distribution and therefore we used a Mann Whitney U test on random subsets of nodes in these networks to compare pre- and post-COVID properties. The mean effect size (E) and P-values are shown for each metric in parenthesis. We concluded that all of these pre- versus post-COVID-19 network properties were statistically similar. (2) Properties of the ego networks (networks induced by each node and its 'one-hop' neighbors). We considered density (average number of neighbors for each node;higher density generally favors lower outbreak threshold) and degree centrality (number of links held by each node) of ego networks (middle right and bottom left panels). The mean effect size and p-values using the Mann Whitney test are shown in parenthesis;there were no statistically significant differences in these properties in the pre- and post-COVID networks. (3) Aggregate properties of the weekly networks, consisting of all the interactions within a week. We considered modularity (measure of how the community structure differs from a random network;higher modularity means a stronger community structure and lower likelihood of transmission) and density (average number of neighbors each node;higher density generally favors lower outbreak threshold) of the weekly networks (bottom middle and bottom right panels). The modularity in the post-COVID weekly networks was slightly lower (i.e., it has a weaker community structure, and the network is more well mixed), while density was slightly higher, the differences of which were statistically significant;a caveat is that these are relatively small datasets (about 40 weeks). These differences (higher density, and better connectivity) both increase the risk of transmission in the post-COVID networks. In summary, the post-COVID networks either have similar properties as the pre-COVID networks, or had changes which are unlikely to have played a role in reducing MRSA transmission. Conclusion. A significant reduction in post-COVD-19 MRSA transmission may have been an unintended positive effect of enhanced infection control measures, particularly hand hygiene and increased mask use. A modest (11.6%) post-COVD-19 reduction in surveillance testing may have also played a role. Despite pandemic-related cohorting and census fluctuations, most network properties were not significantly different post-COVID-19, except for aggregate density and modularity which varied in a directio that instead favored transmission;therefore, HCP-based networks did not play a significant role in reducing MRSA transmission. Multivariate modeling to isolate relative contributions of these factors is underway. Figure 6. Surveillance Testing and Clinical Culturing Post-COVD-19, there was a modest (11.6%) but statistically significant reduction in surveillance PCR testing (42.4 mean tests per 1,000 patient days pre- versus 37.5 post-COVD-19;P<0.002). There was not a statistically significant difference in rates of clinical cultures sent (2.48 cultures per 1,000 patient days pre- versus 2.23 post-COVD-19;P=0.288).

AI-Driven Agent-Based Models to Study the Role of Vaccine Acceptance in Controlling COVID-19 Spread in the US

We study the role of vaccine acceptance in controlling the spread of COVID-19 in the US using AI-driven agent-based models. Our study uses a 288 million node social contact network spanning all 50 US states plus Washington DC, comprised of 3300 counties, with 12.59 billion daily interactions. The highly-resolved agent-based models use realistic information about disease progression, vaccine uptake, production schedules, acceptance trends, prevalence, and social distancing guidelines. Developing a national model at this resolution that is driven by realistic data requires a complex scalable workflow, model calibration, simulation, and analytics components. Our workflow optimizes the total execution time and helps in improving overall human productivity.This work develops a pipeline that can execute US-scale models and associated workflows that typically present significant big data challenges. Our results show that, when compared to faster and accelerating vaccinations, slower vaccination rates due to vaccine hesitancy cause averted infections to drop from 6.7M to 4.5M, and averted total deaths to drop from 39.4K to 28.2K nationwide. This occurs despite the fact that the final vaccine coverage is the same in both scenarios. Improving vaccine acceptance by 10% in all states increases averted infections from 4.5M to 4.7M (a 4.4% improvement) and total deaths from 28.2K to 29.9K (a 6% increase) nationwide. The analysis also reveals interesting spatio-temporal differences in COVID-19 dynamics as a result of vaccine acceptance. To our knowledge, this is the first national-scale analysis of the effect of vaccine acceptance on the spread of COVID-19, using detailed and realistic agent-based models. © 2021 IEEE.

The 4th International Workshop on Epidemiology meets Data Mining and Knowledge Discovery (epiDAMIK 4.0 @ KDD2021)

The 4th epiDAMIK@SIGKDD workshop is a forum to discuss new insights into how data mining can play a bigger role in epidemiology and public health research. While the integration of data science methods into epidemiology has significant potential, it remains under studied. We aim to raise the profile of this emerging research area of data-driven and computational epidemiology, and create a venue for presenting state-of-the-art and in-progress results-in particular, results that would otherwise be difficult to present at a major data mining conference, including lessons learnt in the 'trenches'. The current COVID-19 pandemic has only showcased the urgency and importance of this area. Our target audience consists of data mining and machine learning researchers from both academia and industry who are interested in epidemiological and public-health applications of their work, and practitioners from the areas of mathematical epidemiology and public health. © 2021 Owner/Author.

PREEMPT: Scalable Epidemic Interventions Using Submodular Optimization on Multi-GPU Systems

Preventing and slowing the spread of epidemics is achieved through techniques such as vaccination and social distancing. Given practical limitations on the number of vaccines and cost of administration, optimization becomes a necessity. Previous approaches using mathematical programming methods have shown to be effective but are limited by computational costs. In this work, we present PREEMPT, a new approach for intervention via maximizing the influence of vaccinated nodes on the network. We prove submodular properties associated with the objective function of our method so that it aids in construction of an efficient greedy approximation strategy. Consequently, we present a new parallel algorithm based on greedy hill climbing for PREEMPT, and present an efficient parallel implementation for distributed CPU-GPU heterogeneous platforms. Our results demonstrate that PREEMPT is able to achieve a significant reduction (up to 6.75x) in the percentage of people infected and up to 98% reduction in the peak of the infection on a city scale network. We also show strong scaling results of PREEMPT on up to 128 nodes of the Summit supercomputer. Our parallel implementation is able to significantly reduce time to solution, from hours to minutes on large networks. This work represents a first-of-its -kind effort in parallelizing greedy hill climbing and applying it toward devising effective interventions for epidemics.