Typically, long-term acute care hospitals (LTACHs) have less experience in and incentives to implementing aggressive infection control for drug-resistant organisms such as carbapenem-resistant Enterobacteriaceae (CRE) than acute care hospitals. Decision makers need to understand how implementing control measures in LTACHs can impact CRE spread regionwide. Using our Chicago metropolitan region agent-based model to simulate CRE spread and control, we estimated that a prevention bundle in only LTACHs decreased prevalence by a relative 4.6%-17.1%, averted 1,090-2,795 new carriers, 273-722 infections and 37-87 deaths over 3 years and saved $30.5-$69.1 million, compared with no CRE control measures. When LTACHs and intensive care units intervened, prevalence decreased by a relative 21.2%. Adding LTACHs averted an additional 1,995 carriers, 513 infections, and 62 deaths, and saved $47.6 million beyond implementation in intensive care units alone. Thus, LTACHs may be more important than other acute care settings for controlling CRE, and regional efforts to control drug-resistant organisms should start with LTACHs as a centerpiece.

Background. When trying to control regional spread of antibiotic-resistant pathogens such as carbapenem-resistant Enterobacteriaceae (CRE), decision makers must choose the highest-yield facilities to target for interventions. The question is, with limited resources, how best to choose these facilities. Methods. Using our Regional Healthcare Ecosystem Analyst-generated agent-based model of all Chicago metropolitan area inpatient facilities, we simulated the spread of CRE and different ways of choosing facilities to apply a prevention bundle (screening, chlorhexidine gluconate bathing, hand hygiene, geographic separation, and patient registry) to a resource-limited 1686 inpatient beds. Results. Randomly selecting facilities did not impact prevalence, but averted 620 new carriers and 175 infections, saving $6.3 million in total costs compared to no intervention. Selecting facilities by type (eg, long-term acute care hospitals) yielded a 16.1% relative prevalence decrease, preventing 1960 cases and 558 infections, saving $62.4 million more than random selection. Choosing the largest facilities was better than random selection, but not better than by type. Selecting by considering connections to other facilities (ie, highest volume of discharge patients) yielded a 9.5% relative prevalence decrease, preventing 1580 cases and 470 infections, and saving $51.6 million more than random selection. Selecting facilities using a combination of these metrics yielded the greatest reduction (19.0% relative prevalence decrease, preventing 1840 cases and 554 infections, saving $59.6 million compared with random selection). Conclusions. While choosing target facilities based on single metrics (eg, most inpatient beds, most connections to other facilities) achieved better control than randomly choosing facilities, more effective targeting occurred when considering how these and other factors (eg, patient length of stay, care for higher-risk patients) interacted as a system.

We describe hyper-parameter optimization and training for a novel convolutional neural network that identifies neural axons in 3D volume electron microscope data for a larval zebrafish. Training a convolutional neural network to detect image components in three dimensional space has a variety of applications. There are many parts of a network that can be tuned to work most effectively in a supercomputing environment. We optimized run parameters including batch sizes, thread counts, and epoch sizes as well as hyper-parameters such as drop-out rates. Once we had improved the speed and accuracy of the network we created a large three dimensional volume to visualize the errors of the network. We added new samples to our training set based on the larger volume and improved the accuracy of the network using those samples. Adapting the network to our supercomputing environment and adjusting the dropout layers increased the accuracy and efficiency of the network. Being able to accurately detect neurons in three dimensional space shows potential for other deep learning applications.

Background: Twitter presents a valuable and relevant social media platform to study the prevalence of information and sentiment on vaping that may be useful for public health surveillance. Machine learning classifiers that identify vaping-relevant tweets and characterize sentiments in them can underpin a Twitter-based vaping surveillance system. Compared with traditional machine learning classifiers that are reliant on annotations that are expensive to obtain, deep learning classifiers offer the advantage of requiring fewer annotated tweets by leveraging the large numbers of readily available unannotated tweets. Objective: This study aims to derive and evaluate traditional and deep learning classifiers that can identify tweets relevant to vaping, tweets of a commercial nature, and tweets with provape sentiments. Methods: We continuously collected tweets that matched vaping-related keywords over 2 months from August 2018 to October 2018. From this data set of tweets, a set of 4000 tweets was selected, and each tweet was manually annotated for relevance (vape relevant or not), commercial nature (commercial or not), and sentiment (provape or not). Using the annotated data, we derived traditional classifiers that included logistic regression, random forest, linear support vector machine, and multinomial naive Bayes. In addition, using the annotated data set and a larger unannotated data set of tweets, we derived deep learning classifiers that included a convolutional neural network (CNN), long short-term memory (LSTM) network, LSTM-CNN network, and bidirectional LSTM (BiLSTM) network. The unannotated tweet data were used to derive word vectors that deep learning classifiers can leverage to improve performance. Results: LSTM-CNN performed the best with the highest area under the receiver operating characteristic curve (AUC) of 0.96 (95% CI 0.93-0.98) for relevance, all deep learning classifiers including LSTM-CNN performed better than the traditional classifiers with an AUC of 0.99 (95% CI 0.98-0.99) for distinguishing commercial from noncommercial tweets, and BiLSTM performed the best with an AUC of 0.83 (95% CI 0.78-0.89) for provape sentiment. Overall, LSTM-CNN performed the best across all 3 classification tasks. Conclusions: We derived and evaluated traditional machine learning and deep learning classifiers to identify vaping-related relevant, commercial, and provape tweets. Overall, deep learning classifiers such as LSTM-CNN had superior performance and had the added advantage of requiring no preprocessing. The performance of these classifiers supports the development of a vaping surveillance system.

Using VisIt, we visualized a 3D volume of neural axons for a larval zebrafish from electron microscope data and locations for those neurons estimated by a deep neural network. Errant points were recorded and used to further train the network. These computations were performed using Tensorflow, and utilized GPU’s to increase performance.

BACKGROUND:Regions are considering the use of electronic registries to track patients who carry antibiotic-resistant bacteria, including carbapenem-resistant Enterobacteriaceae (CRE). Implementing such a registry can be challenging and requires time, effort, and resources; therefore, there is a need to better understand the potential impact.METHODS:We developed an agent-based model of all inpatient healthcare facilities (90 acute care hospitals, 9 long-term acute care hospitals, 351 skilled nursing facilities, and 12 ventilator-capable skilled nursing facilities) in the Chicago metropolitan area, surrounding communities, and patient flow using our Regional Healthcare Ecosystem Analyst software platform. Scenarios explored the impact of a registry that tracked patients carrying CRE to help guide infection prevention and control.RESULTS:When all Illinois facilities participated (n = 402), the registry reduced the number of new carriers by 11.7% and CRE prevalence by 7.6% over a 3-year period. When 75% of the largest Illinois facilities participated (n = 304), registry use resulted in a 11.6% relative reduction in new carriers (16.9% and 1.2% in participating and nonparticipating facilities, respectively) and 5.0% relative reduction in prevalence. When 50% participated (n = 201), there were 10.7% and 5.6% relative reductions in incident carriers and prevalence, respectively. When 25% participated (n = 101), there was a 9.1% relative reduction in incident carriers (20.4% and 1.6% in participating and nonparticipating facilities, respectively) and 2.8% relative reduction in prevalence.CONCLUSIONS:Implementing an extensively drug-resistant organism registry reduced CRE spread, even when only 25% of the largest Illinois facilities participated due to patient sharing. Nonparticipating facilities garnered benefits, with reductions in new carriers.

There is growing interest in conducting public health research using data from social media. In particular, Twitter “infoveillance” has demonstrated utility across health contexts. However, rigorous and reproducible methodologies for using Twitter data in public health are not yet well articulated, particularly those related to content analysis, which is a highly popular approach. In 2014, we gathered an interdisciplinary team of health science researchers, computer scientists, and methodologists to begin implementing an open-source framework for real-time infoveillance of Twitter health messages (RITHM). Through this process, we documented common challenges and novel solutions to inform future work in real-time Twitter data collection and subsequent human coding. The RITHM framework allows researchers and practitioners to use well-planned and reproducible processes in retrieving, storing, filtering, subsampling, and formatting data for health topics of interest. Further considerations for human coding of Twitter data include coder selection and training, data representation, codebook development and refinement, and monitoring coding accuracy and productivity. We illustrate methodological considerations through practical examples from formative work related to hookah tobacco smoking, and we reference essential methods literature related to understanding and using Twitter data.

Introduction: By pairing diluent with vaccines, dual-chamber vaccine injection devices simplify the process of reconstituting vaccines before administration and thus decrease associated open vial wastage and adverse events. However, since these devices are larger than current vaccine vials for lyophilized vaccines, manufacturers need guidance as to how the size of these devices may affect vaccine distribution and delivery. Methods: Using HERMES-generated immunization supply chain models of Benin, Bihar (India), and Mozambique, we replace the routine 10-dose measles-rubella (MR) lyophilized vaccine with single-dose MR dual-chamber injection devices, ranging the volume-per-dose (5.2-26 cm(3)) and price-per-dose ($0.70, $1.40). Results: At a volume-per-dose of 5.2 cm(3), a dual-chamber injection device results in similar vaccine availability, decreased open vial wastage (OVW), and similar total cost per dose administered as compared to baseline in moderately constrained supply chains. Between volumes of 7.5 cm(3) and 26 cm(3), these devices lead to a reduction in vaccine availability between 1% and 14% due to increases in cold chain storage utilization between 1% and 7% and increases in average peak transport utilization between 2% and 44%. At the highest volume-per-dose, 26 cm(3), vaccine availability decreases between 9% and 14%. The total costs per dose administered varied between each scenario, as decreases in vaccine procurement costs were coupled with decreases in doses administered. However, introduction of a dual-chamber injection device only resulted in improved total cost per dose administered for Benin and Mozambique (at 5.2 cm(3) and $0.70-per-dose) when the total number of doses administered changed <1% from baseline. Conclusion: In 3 different country supply chains, a single-dose MR dual-chamber injection device would need to be no larger than 5.2 cm(3) to not significantly impair the flow of other vaccines. (C) 2018 The Authors. Published by Elsevier Ltd.

Although vaccine supply chains in many countries require additional stationary storage and transport capacity to meet current and future needs, international donors tend to donate stationary storage devices far more often than transport equipment. To investigate the impact of only adding stationary storage equipment on the capacity requirements of transport devices and vehicles, we used HERMES (Highly Extensible Resource for Modeling Supply Chains) to construct a discrete event simulation model of the Niger vaccine supply chain. We measured the transport capacity requirement for each mode of transport used in the Niger vaccine cold chain, both before and after adding cold rooms and refrigerators to relieve all stationary storage constraints in the system. With the addition of necessary stationary storage, the average transport capacity requirement increased from 88% to 144% for cold trucks, from 101% to 197% for pickup trucks, and from 366% to 420% for vaccine carriers. Therefore, adding stationary storage alone may worsen or create new transport bottlenecks as more vaccines flow through the system, preventing many vaccines from reaching their target populations. Dynamic modeling can reveal such relationships between stationary storage capacity and transport constraints.

Background: When addressing the urgent task of improving vaccine supply chains, especially to accommodate the introduction of new vaccines, there is often a heavy emphasis on stationary storage. Currently, donations to vaccine supply chains occur largely in the form of storage equipment.Methods: This study utilized a HERMES-generated detailed, dynamic, discrete event simulation model of the Niger vaccine supply chain to compare the impacts on vaccine availability of adding stationary cold storage versus transport capacity at different levels and to determine whether adding stationary storage capacity alone would be enough to relieve potential bottlenecks when pneumococcal and rotavirus vaccines are introduced by 2015.Results: Relieving regional level storage bottlenecks increased vaccine availability (by 4%) more than relieving storage bottlenecks at the district (1% increase), central (no change), and clinic (no change) levels alone. Increasing transport frequency (or capacity) yielded far greater gains (e.g., 15% increase in vaccine availability when doubling transport frequency to the district level and 18% when tripling). In fact, relieving all stationary storage constraints could only increase vaccine availability by 11%, whereas doubling the transport frequency throughout the system led to a 26% increase and tripling the frequency led to a 30% increase. Increasing transport frequency also reduced the amount of stationary storage space needed in the supply chain. The supply chain required an additional 61,269L of storage to relieve constraints with the current transport frequency, 55,255L with transport frequency doubled, and 51,791L with transport frequency tripled.Conclusions: When evaluating vaccine supply chains, it is important to understand the interplay between stationary storage and transport. The HERMES-generated dynamic simulation model showed how augmenting transport can result in greater gains than only augmenting stationary storage and can reduce stationary storage needs.

Objective: Determine the effects on the vaccine cold chain of making different types of World Health Organization (WHO) Expanded Program on Immunizations (EPI) vaccines thermostable.Methods: Utilizing a detailed computational, discrete-event simulation model of the Niger vaccine supply chain, we simulated the impact of making different combinations of the six current EPI vaccines thermostable.Findings: Making any EPI vaccine thermostable relieved existing supply chain bottlenecks (especially at the lowest levels), increased vaccine availability of all EPI vaccines, and decreased cold storage and transport capacity utilization. By far, the most substantial impact came from making the pentavalent vaccine thermostable, increasing its own vaccine availability from 87% to 97% and the vaccine availabilities of all other remaining non-thermostable EPI vaccines to over 93%. By contrast, making each of the other vaccines thermostable had considerably less effect on the remaining vaccines, failing to increase the vaccine availabilities of other vaccines to more than 89%. Making tetanus toxoid vaccine along with the pentavalent thermostable further increased the vaccine availability of all EPI vaccines by at least 1-2%.Conclusion: Our study shows the potential benefits of making any of Niger's EPI vaccines thermostable and therefore supports further development of thermostable vaccines. Eliminating the need for refrigerators and freezers should not necessarily be the only benefit and goal of vaccine thermostability. Rather, making even a single vaccine (or some subset of the vaccines) thermostable could free up significant cold storage space for other vaccines, and thereby help alleviate supply chain bottlenecks that occur throughout the world. (C) 2012 Elsevier Ltd. All rights reserved.

Background: When policymakers make decision about the target populations and timing of influenza vaccination, they may not consider the impact on the vaccine supply chains, which may in turn affect vaccine availability.Purpose: Our goal is to explore the effects on the Thailand vaccine supply chain of introducing influenza vaccines and varying the target populations and immunization time-frames.Methods: We Utilized our custom-designed software HERMES (Highly Extensible Resource for Modeling Supply Chains), we developed a detailed, computational discrete-event simulation model of the Thailand's National Immunization Program (NIP) supply chain in Trang Province, Thailand. A suite of experiments simulated introducing influenza vaccines for different target populations and over different time-frames prior to and during the annual influenza season.Results: Introducing influenza vaccines creates bottlenecks that reduce the availability of both influenza vaccines as well as the other NIP vaccines, with provincial to district transport capacity being the primary constraint. Even covering only 25% of the Advisory Committee on Immunization Practice-recommended population while administering the vaccine over six months hinders overall vaccine availability so that only 62% of arriving patients can receive vaccines. Increasing the target population from 25% to 100% progressively worsens these bottlenecks, while increasing influenza vaccination time-frame from 1 to 6 months decreases these bottlenecks.Conclusion: Since the choice of target populations for influenza vaccination and the time-frame to deliver this vaccine can substantially affect the flow of all vaccines, policy-makers may want to consider supply chain effects when choosing target populations for a vaccine. (c) 2012 Elsevier Ltd. All rights reserved.

BACKGROUND:Many countries, such as Niger, are considering changing their vaccine vial size presentation and may want to evaluate the subsequent impact on their supply chains, the series of steps required to get vaccines from their manufacturers to patients. The measles vaccine is particularly important in Niger, a country prone to measles outbreaks.METHODS:We developed a detailed discrete event simulation model of the vaccine supply chain representing every vaccine, storage location, refrigerator, freezer, and transport device (e.g., cold trucks, 4 × 4 trucks, and vaccine carriers) in the Niger Expanded Programme on Immunization (EPI). Experiments simulated the impact of replacing the 10-dose measles vial size with 5-dose, 2-dose and 1-dose vial sizes.RESULTS:Switching from the 10-dose to the 5-dose, 2-dose and 1-dose vial sizes decreased the average availability of EPI vaccines for arriving patients from 83% to 82%, 81% and 78%, respectively for a 100% target population size. The switches also changed transport vehicle's utilization from a mean of 58% (range: 4-164%) to means of 59% (range: 4-164%), 62% (range: 4-175%), and 67% (range: 5-192%), respectively, between the regional and district stores, and from a mean of 160% (range: 83-300%) to means of 161% (range: 82-322%), 175% (range: 78-344%), and 198% (range: 88-402%), respectively, between the district to integrated health centres (IHC). The switch also changed district level storage utilization from a mean of 65% to means of 64%, 66% and 68% (range for all scenarios: 3-100%). Finally, accounting for vaccine administration, wastage, and disposal, replacing the 10-dose vial with the 5 or 1-dose vials would increase the cost per immunized patient from $0.47US to $0.71US and $1.26US, respectively.CONCLUSIONS:The switch from the 10-dose measles vaccines to smaller vial sizes could overwhelm the capacities of many storage facilities and transport vehicles as well as increase the cost per vaccinated child.

OBJECTIVESWe investigated whether introducing the rotavirus and pneumococcal vaccines, which are greatly needed in West Africa, would overwhelm existing supply chains (i.e., the series of steps required to get a vaccine from the manufacturers to the target population) in Niger.METHODSAs part of the Bill and Melinda Gates Foundation-funded Vaccine Modeling Initiative, we developed a computational model to determine the impact of introducing these new vaccines to Niger's Expanded Program on Immunization vaccine supply chain.RESULTSIntroducing either the rotavirus vaccine or the 7-valent pneumococcal conjugate vaccine could overwhelm available storage and transport refrigerator space, creating bottlenecks that would prevent the flow of vaccines down to the clinics. As a result, the availability of all World Health Organization Expanded Program on Immunization vaccines to patients might decrease from an average of 69% to 28.2% (range = 10%-51%). Addition of refrigerator and transport capacity could alleviate this bottleneck.CONCLUSIONSOur results suggest that the effects on the vaccine supply chain should be considered when introducing a new vaccine and that computational models can help assess evolving needs and prevent problems with vaccine delivery.

Introduced to minimize open vial wastage, single-dose vaccine vials require more storage space and therefore may affect vaccine supply chains (i.e., the series of steps and processes involved in distributing vaccines from manufacturers to patients). We developed a computational model of Thailand's Trang province vaccine supply chain to analyze the effects of switching from a ten-dose measles vaccine presentation to each of the following: a single-dose measles-mumps-rubella vaccine (which Thailand is currently considering) or a single-dose measles vaccine. While the Trang province vaccine supply chain would generally have enough storage and transport capacity to accommodate the switches, the added volume could push some locations’ storage and transport space utilization close to their limits. Single-dose vaccines would allow for more precise ordering and decrease open vial waste, but decrease reserves for unanticipated demand. Moreover, the added disposal and administration costs could far outweigh the costs saved from preventing open vial wastage.

Although the substantial burdens of rotavirus and pneumococcal disease have motivated many countries to consider introducing the rotavirus vaccine (RV) and heptavalent pneumococcal conjugate vaccine (PCV-7) to their National Immunization Programs (EPIs), these new vaccines could affect the countries' vaccine supply chains (i.e., the series of steps required to get a vaccine from their manufacturers to patients). We developed detailed computational models of the Trang Province, Thailand, vaccine supply chain to simulate introducing various RV and PCV-7 vaccine presentations and their combinations. Our results showed that the volumes of these new vaccines in addition to current routine vaccines could meet and even exceed (1) the refrigerator space at the provincial district and sub-district levels and (2) the transport cold space at district and sub-district levels preventing other vaccines from being available to patients who arrive to be immunized. Besides the smallest RV presentation (17.1 cm³/dose), all other vaccine introduction scenarios required added storage capacity at the provincial level (range: 20 L-1151 L per month) for the three largest formulations, and district level (range: 1 L-124 L per month) across all introduction scenarios. Similarly, with the exception of the two smallest RV presentation (17.1 cm³/dose), added transport capacity was required at both district and sub-district levels. Added transport capacity required across introduction scenarios from the provincial to district levels ranged from 1 L-187 L, and district to sub-district levels ranged from 1 L-13 L per shipment. Finally, only the smallest RV vaccine presentation (17.1 cm³/dose) had no appreciable effect on vaccine availability at sub-districts. All other RV and PCV-7 vaccines were too large for the current supply chain to handle without modifications such as increasing storage or transport capacity. Introducing these new vaccines to Thailand could have dynamic effects on the availability of all vaccines that may not be initially apparent to decision-makers.

Algorithms for rotation of 2D images by multiple shearing transformations are well known; algorithms which use as few as four shears to perform an arbitrary 3D rotation on a 3D volume have also been described. By using Fourier transform methods to implement these shears, rotations can be performed completely reversibly and without loss of information. This is of great utility when the rotated data are used as input to statistical calculations, for example in human brain imaging. In general, six different patterns of four shears can be used to implement a given 3D rotation. We examine the mathematical and implementation details of these rotation algorithms. This paper provides a classification of these patterns, demonstrates that a consistent scheme must be used to select shear patterns for various rotations, and presents several such consistent schemes.

This paper presents I/O solutions for the visualization of time-varying volume data in a parallel and distributed computing environment. Depending on the number of rendering processors used, our I/O strategies help signifi- cantly lower interframe delay by employing a set of I/O processors coupled with MPI parallel I/O support. The targeted application is earthquake modeling using a large 3D unstructured mesh consisting of one hundred millions cells. Our test results on the HP/Compaq AlphaServer operated at the Pittsburgh Supercomputing Center demonstrate that the I/O strategies effectively remove the I/O bottlenecks commonly present in time-varying data visualization. This high-performance visualization solution we provide to the scientists allows them to explore their data in the temporal, spatial, and visualization domains at high resolution. This new high-resolution explorability, likely not presently available to most computational science groups, will help lead to many new insights.