Healthcare Associated Infections (HAIs) like MRSA are a major threat to our hospitals and public health systems that significantly
affect lives and resources. Unlike many common transmissible diseases, HAIs spread not only via direct person to person contacts but also indirectly through infected surfaces. Consequently, many of the standard epidemiological models like SIS, SEIR etc. cannot be used here. As a result, recently, many 2 Mode models were developed where people and surfaces behave differently allowing the pathogen to spread both via people and surfaces. However, these 2-Mode models are linear and cannot model non-linear contagions and group interactions effectively. In this paper, we present an agent based 2-Mode Hypergraph based Model and show that it is more expressive than the graph based benchmark both theoretically and experimentally.

Modelling Healthcare Associated Infections with Hypergraphs

Real-time forecasting of disease outbreaks requires standardized outputs generated in a timely manner. Development of pipelines to automate infectious disease forecasts can ensure that parameterization and software dependencies are common to any execution of the forecasting code. Here we present our implementation of an automated cloud computing pipeline to forecast infectious disease outcomes, with examples of usage to forecast  COVID-19 and influenza targets. We also offer our perspective on the limits of automation and importance of human-in-the-loop automated infectious disease forecasting.

Automated Infectious Disease Forecasting: Use-cases and Practical Considerations for Pipeline Implementation

Data collected in the Global COVID-19 Trends and Impact Surveys (UMD Global CTIS), and data on variants sequencing from GISAID, are used to evaluate the impact of the Omicron variant (in South Africa and other countries) on the prevalence of COVID-19 among unvaccinated and vaccinated population, in general and discriminating by the number of doses. In South Africa, we observe that the prevalence of COVID-19 in December (with strong presence of Omicron) among the unvaccinated population is comparable to the prevalence during the previous wave (in August-September), in which Delta was the variant with the largest presence. However, among vaccinated, the prevalence of COVID-19 in December is much higher than in the previous wave. In fact, a significant reduction of the vaccine efficacy is observed from August-September to December. For instance, the efficacy drops from $0.81$ to $0.30$ for those vaccinated with 2 doses, and from $0.51$ to $0.09$ for those vaccinated with one dose. The study is then extended to other countries in which Omicron has been detected, comparing the situation in October (before Omicron) with that of December. While the reduction measured is smaller than in South Africa, we still found, for instance, an average drop in vaccine efficacy from $0.53$ to $0.45$ among those vaccinated with two doses. Moreover, we found a significant negative (Pearson) correlation of around $-0.6$ between the measured prevalence of Omicron and the vaccine efficacy.

Using Survey Data to Estimate the Impact of the Omicron Variant on Vaccine Efficacy against COVID-19 Infection

A key challenge with data collection for time-series data for forecasting is detecting anomalies in real-time. Anomalies can negatively affect the performance of forecasting models, if not accounted for. It is a relatively simple process to retrospectively identify an anomaly in a time-series and correct it, however the problem lies with identifying a particular time-step as anomalous when there is no further information to distinguish it from a change in trend. In the context of the COVID-19 pandemic, accurate forecasts are critical to policymakers, and yet, most data is recorded manually, susceptible to human errors or other technical constraints. While most errors are corrected, it often takes several weeks for these corrections to be made, leading to real-time forecasting models generating predictions on data that is not necessarily accurate. In order to prevent this, we utilize a neural framework called Back2Future and build a simple classifier on top of it to propose a real-time anomaly detection system in the context of detecting anomalies in COVID-19 reporting data.

Real-time Anomaly Detection in Epidemic Data Streams

We consider the problem of diffusion control via interventions that change network topologies. We study this problem for general weighted networks and present an iterative algorithm, Frank-Wolfe-EdgeCentrality, to reduce the spread of a diffusion process by shrinking the network's top singular values. Given an edge-weight reduction budget, our algorithm identifies the near-optimal edge-weight reduction strategy to minimize the sum of the largest $r$ eigenvalues of $W^{\top}W$, where $W$ is the network weight matrix. Our algorithm provably converges to the optimum at a rate of $O(t^{-1})$ after $t$ iterations; each iteration only requires a nearly-linear runtime in the number of edges.

We perform a detailed empirical study of our algorithm on a wide range of weighted networks. In particular, we apply our approach to reduce edge weights on mobility networks (between points of interest and census block groups), which have been used to model the spread of COVID-19. In SEIR model simulations, our algorithm reduces the number of infections by 25.70% more than existing approaches, averaged over three weighted graphs and eight mobility networks. Meanwhile, the largest singular value of the weight matrix $W$ decreases by 25.48% more than existing approaches on these networks. An extension of our algorithm to temporal mobility networks also shows an effective reduction in the number of infected nodes.

Optimal Intervention on Weighted Networks via Edge Centrality

Detecting risky behavior using smartphones and other mobile devices may help mitigate the spread of infectious diseases. However, the privacy concerns introduced by individualized activity recognition may counteract potential benefits. As an example, consider a public health official gauging their message's efficacy. Machine learning and data mining methods may help them understand how their local population's behavior changes, but aggressive surveillance can severely hinder individual privacy and cause disproportionate harm to disadvantaged groups. In this work, we benchmark a series of machine learning algorithms predicting high-risk behaviors---going to bars and gyms, attending parties, and riding on buses---given only low-level smartphone sensor data. We find that models trained to perform these challenging tasks are largely unreliable and should be avoided in practice, though their predictions are significantly better than random.

On Detecting COVID-Risky Behavior from Smartphones

Metapopulation models capture the spatial interactions in disease dynamics through mobility and mixing matrices among regions of interest. However, the impact of seeding (initialization) in such networks is not well understood. We have constructed and open-sourced metapopulation networks for countries around the world (Patch-Flow) and we use them to study the effect of seeding using an extension of a discrete-time SEIR simulator, PatchSim. The impact of initialization is studied by looking at the resulting epidemic curves.
We use various metrics to characterize the epidemic curves, including those based on epidemic intensity entropy. Using these, we study the impact of levels of connectivity, skewness in seeding, and spatial resolution at the national scale. We find that these effects vary across countries and are more pronounced at certain transmissibility levels. This study provides early insights into the impact of model initialization and demonstrates the use of PatchFlow networks.

Impact of Seeding and Spatial Heterogeneity on Metapopulation Disease Dynamics

With healthcare being critical aspect, health insurance has become an important scheme in minimizing medical expenses. Medicare is an example of such a healthcare insurance initiative in the United States. Following this, the healthcare industry has seen a significant increase in fraudulent activities owing to increased insurance, and fraud has become a significant contributor to rising medical care expenses, although its impact can be mitigated using fraud detection techniques. To detect fraud, machine learning techniques are used. The Centers for Medicaid and Medicare Services (CMS) of the United States federal government released "Medicare Part D" insurance claims are utilized in this study to develop fraud detection system. Employing machine learning algorithms on a class-imbalanced and high-dimensional medicare dataset is a challenging task. To compact such challenges, the present work aims to perform feature extraction following data sampling, afterward applying various classification algorithms, to get better performance. Feature extraction is a dimensionality reduction approach that converts attributes into linear or non-linear combinations of the actual attributes, generating a smaller and more diversified set of attributes and thus reducing the dimensions. Data sampling is commonly used to address the class imbalance either by expanding the frequency of minority class or reducing the frequency of majority class to obtain approximately equal numbers of occurrences for both classes. The proposed approach is evaluated through standard performance metrics such as F-measure and AUC score. Thus, to detect fraud efficiently, this study applies autoencoder as a feature extraction technique, synthetic minority oversampling technique (SMOTE) as a data sampling technique, and various gradient boosted decision tree-based classifiers as a classification algorithm. The experimental results show the combination of autoencoders followed by SMOTE on the LightGBM (short for, Light Gradient Boosting Machine) classifier achieved best results.

Impact of the composition of feature extraction and class sampling in medicare fraud detection

In light of the COVID-19 pandemic, it is an open challenge and critical practical problem to find a optimal way to dynamically prescribe the best policies that balance both the governmental resources and epidemic control in different countries and regions. To solve this multi-dimensional tradeoff of exploitation and exploration, we formulate this technical challenge as a contextual combinatorial bandit problem that jointly optimizes a multi-criteria reward function. Given the historical daily cases in a region and the past intervention plans in place, the agent should generate useful intervention plans that policy makers can implement in real time to minimizing both the number of daily COVID-19 cases and the stringency of the recommended interventions. We prove this concept with simulations of multiple realistic policy making scenarios and demonstrate a clear advantage in providing a pareto optimal solution in the epidemic intervention problem. 

Optimal Epidemic Control as a Contextual Combinatorial Bandit with Budget

Social media plays a pivotal role in acquiring, exchanging and expressing public opinions and perceptions on a unprecedented scale
in these pandemic times. In this paper, we develop an end-to-end knowledge extraction and management framework named as EVADE.
This framework is used to automatically extract information consistent and inconsistent with scientific evidence regarding vaccination. Additionally, we seek to explore public opinion towards vaccination resistance proposing novel natural language processing methods. The knowledge extraction pipeline consists of three major modules, namely, knowledge-base construction, categorization of vaccine dissenting tweets, and effective analyses of discourses in those tweets effectively. Our major contributions lie in the fact that (i) the proposed knowledge extraction framework does not require huge amounts of labelled tweets of different categories and (ii) our module outperformed baselines by a significant margin of ≈ 8% to ≈ 14% in the classification tasks, and effectively analyze vaccine dissenting discourse

EVADE: Exploring Vaccine Dissenting Discourse on Twitter

$\textbf{Background:}$ At the onset of a pandemic, such as COVID-19, data with proper labeling/attributes corresponding to the new disease might be unavailable or sparse. Machine Learning (ML) models trained with the available data, which is limited in quantity and poor in diversity, will often be biased and inaccurate. At the same time, ML algorithms designed to fight pandemics must have good performance and be developed in a time-sensitive manner. To tackle the challenges of limited data, and label scarcity in the available data, we propose generating conditional synthetic data, to be used alongside real data for developing robust ML models. $\textbf{Methods:}$ We present a  hybrid model consisting of a conditional generative flow and a classifier for conditional synthetic data generation. $\textbf{Results:}$ We performed conditional synthetic generation for chest computed tomography (CT) scans corresponding to normal, COVID-19, and pneumonia afflicted patients. We show that our method significantly outperforms existing models both on qualitative and quantitative performance. As an example of downstream use of synthetic data, we show improvement in COVID-19 detection from CT scans with conditional synthetic data augmentation.

ACM SIGKDD 2022 Workshop epiDAMIK Submissions