Search

Long-term endocrine therapy (e.g. Tamoxifen, aromatase inhibitors) is crucial to prevent breast cancer recurrence, yet rates of adherence to these medications are low. To develop, evaluate, and sustain future interventions, individual-level modeling can be used to understand breast cancer survivors’ behavioral mechanisms of medication-taking. This paper presents interdisciplinary research, wherein a model employing randomized neural networks was developed to predict breast cancer survivors’ daily medication-taking behavior based on their survey data over three time periods (baseline, 4 months, 8 months). The neural network structure was guided by random utility theory developed in psychology and behavioral economics. Comparative analysis indicates that the proposed model outperforms existing computational models in terms of prediction accuracy under conditions of randomness.

Chronic kidney disease (CKD) is a gradual loss of renal function over time, and it increases the risk of mortality, decreased quality of life, as well as serious complications. The prevalence of CKD has been increasing in the last couple of decades, which is partly due to the increased prevalence of diabetes and hypertension. To accurately detect CKD in diabetic patients, we propose a novel framework to learn sparse longitudinal representations of patients' medical records. The proposed method is also compared with widely used baselines such as Aggregated Frequency Vector and Bag-of-Pattern in Sequences on real EHR data, and the experimental results indicate that the proposed model achieves higher predictive performance. Additionally, the learned representations are interpreted and visualized to bring clinical insights., Comment: Accepted in IEEE BIBM 2020

Nowadays, smart building infrastructures are equipped with hundreds of sensors to monitor building environments and provide smart solutions for occupant comfortability and energy efficiency. Ideally, an automated system can predict and adjust the physical features (e.g., lighting, air quality, temperature, and so on) in a person’s office based on his/her personalized preferences and activities. However, since the data is from one person, there may not be sufficient data for machine learning model training, and the data’s quality may be low (e.g., with noises). Then, it is a challenge to conduct accurate predictions to provide personalized environment adjustment. To handle this problem, in this paper, we propose a smart building assistance system consisting of different sensor data analysis approaches and a deep neural network (DNN)-based prediction model to make a more accurate prediction despite low-quality sensor data. First, we collected a year-long smart building dataset from four different data sources (i.e., sensors, calendar, weather, and survey). Second, we perform different feature engineering approaches (i.e., concretization, one-hot encoding, and multiple feature combination) on the data as inputs for the prediction models. Third, we identify a support vector regression-based prediction model and propose a hybrid DNN model consisting of several recurrent neural network blocks and a feed-forward DNN block to predict different preferred physical features considering different activities of a person (e.g., meeting, lunch, research activities). Finally, we conduct experimental studies to evaluate the performance of the proposed prediction models compared to other existing machine learning models in terms of accuracy. Our predicted preferred physical features match the occupant’s preferred ranges of different physical features during a specific activity. We also open-sourced our code on GitHub.

Smartphones can collect millions of data points from each of its users daily, contributing to a significant change in how the healthcare community approaches health monitoring. This paper provides a framework for how smartphone sensor data can be collected, cleaned, stored, and modeled to effectively predict human states as a step towards health monitoring. To develop robust contextual models, a three-week study was conducted to collect data through a mobile crowdsensing application named Sensus. In this study, participants used multiple sensing strategies, ranging from infrequent sampling to continuous sampling, to determine the effect each has on data integrity and battery life. For a future study, a dynamic data collection strategy was developed that uses a machine learning model trained on existing data collected from 220 participants to forecast when a smartphone will be active and trigger sensor sampling accordingly. Results of this study include 1) extraction of model features that deliver maximized data quality with minimized battery consumption as compared to pre-existing baseline models, 2) implementation of context-driven modeling of user smartphone data on user's contextual environment, and 3) customization of a time-series database for optimized data queries used in metadata visualizations. The adaptive sensing models produced could be used in future large population studies that efficiently examine patterns of behavior in multiple individuals over extended periods to identify disease indicators present in an average user’s daily life.

Approximately one in five adults in the United States have been diagnosed with some form of mental illness, but less than half received treatment in this past year [1]. An interdisciplinary team at the University of Virginia aims to reduce this gap in mental health coverage through its freely accessible online research platform, the MindTrails Project. The MindTrails Calm Thinking study evaluates cognitive bias modification for interpretation (CBM-I), an intervention that aims to reframe the thinking patterns of highly anxious individuals when they respond to ambiguous situations that they might interpret as stressful. MindTrails is experiencing a high attrition (dropout) rate, which is common to eHealth interventions. In response to this, our project utilized two novel approaches to online anxiety interventions to improve engagement and retention: (1) personalization of training content and (2) implementation intentions and goal setting. We designed a prototype for a new mobile interface that engages users with a journal to record implementation intentions and goals. Users also have the ability to choose the domain of anxiety (e.g., relationships, health) that they would like to work on. To further incorporate these psychological principles into the MindTrails program, suggestions for future work are also discussed. We hypothesize that, with its new user-centered mobile interface, the Calm Thinking mobile application will further connect users with an evidence-based mental health intervention and increase the efficacy of the program.

This paper explores the combination of Attention Restoration Theory and immersive virtual technology as a novel therapy for short-term stress reduction in the workplace. The goal of this work is to understand how various immersive technologies impact the effect of both nature and urban environments on acute stress. In order to assess this, study participants were guided through “micro-vacations,” or a series of virtual nature or urban images, after being induced with stress. The micro-vacations were presented via three different virtual immersive technologies: a virtual reality (VR) experience using a headset in a booth, a GeoDome experience, or a 2D experience which acted as a control. Biometric, subjective mood and comfort data were gathered from the participants throughout the study in order to measure the changes in stress and mood before, during, and after the microvacation experiences. We hypothesize that the nature environments are more relaxing than the urban environments, and that both the VR booth and GeoDome will reduce stress levels in participants to a greater degree than the 2D images.

Continuous passive sensing using smartphone embedded sensors can drain the battery quickly, interrupting other usages of the device. In order to improve the energy efficiency in continuous mobile sensing applications, we propose a new adaptive sensing framework using reinforcement learning to optimize the sensing timing. We model our adaptive sensing problem as a Markov Decision Process and dynamically change the sensing timing of targeted sensor(s) so that they are only operating in desired contexts (e.g. collect accelerometer data only when the phone is moving). Using accelerometer data continuously collected from 220 participants for over two weeks, we show that our approach is able to save the energy while attaining high accuracy and data coverage. Specifically, our strategy attains energy saving of 62.4 % at an accuracy of 80.9% and data coverage of 67.4%, which outperforms two baseline strategies, including a random strategy and a strategy using learning automata technique.

Recent advances in sensing technology have made it possible to monitor how behavioral systems unfold in people's natural settings by leveraging sensors embedded in personal smartphones and other smart devices. This paper provides a framework for how smartphone sensor data can be collected, cleaned, and modeled to predict relevant disease contexts such as location. These variables can then be used in context-sensitive models to understand how a user's behavior and contexts might differ from typical patterns when impacted by illness. To develop rich contextual models, we first conducted a 2-week smartphone monitoring study where sensor data and corresponding location contexts were tagged for 7 users. Next, we demonstrated how multimodal sensor data can be used to predict location context by modeling the tagged dataset and analyzing differences in sensors to find indicators for each location. The results of this effort include 1) identification of ground truth data for contexts of interest to be used in future modeling, 2) establishment of a process to collect, clean, and visualize smartphone data generated by both iOS and Android systems, and 3) creation of models to predict a participant's location and context using raw smartphone data. This context identification process could be used in future research to perform analyses that leverage past patterns of user behavior to recognize disease indicators.

Approximately one in five people in the United States are affected by mental illness, with anxiety disorders being the most common. Barriers to treatment include limited access to trained professionals and high financial cost. eHealth applications are one alternative to treatment outside of a traditional clinical setting. Patients can readily access eHealth interventions on their own time via devices such as computers, tablets, and smartphones. Despite the scalability and accessibility of eHealth applications, their benefits are overshadowed by high attrition rates. MindTrails (MT), an existing eHealth platform, uses Cognitive Bias Modification (CBM) to treat anxiety through online interventions designed to change negative thinking patterns. The MindTrails program has the potential to treat a large population of anxious individuals. The objective of this work is to identify, analyze, and implement strategies to increase user engagement with MindTrails by exploring the integration of gamification/engagement strategies into the program. Our design for increasing engagement focuses on the Doherty Web Strategies, incorporating interactive, personal, supportive, and social elements. Using this new design, users will be able to set personalized goals that are clear, actionable, and reasonably challenging. To meet our objective, we developed high fidelity wireframes and prototypes, with the intent of utilizing user studies to evaluate the efficacy in MindTrails. Results of user tests are hypothesized to show the effectiveness of a personalized gamification feature in increasing user engagement while simultaneously reducing attrition. The improved design will be included in the next launch of the MindTrails program and demonstrates progress toward increasing the effectiveness of CBM treatment in eHealth applications.

As social media platforms have grown to form the foundation of modern digital communication, digital text message datasets that document interpersonal exchanges on these platforms have proliferated. These exchanges comprise a rich corpus of social context data, which can provide insight into how mental health challenges manifest in social contexts. To date, researchers have employed a variety of methods for extracting mental health-centric features from digital text communication data, including natural language processing, social network analysis, sentiment analysis, time series analysis, and discourse analysis. However, there is a marked divide in current literature between qualitative and quantitative feature extraction methods. To effectively identify and analyze key underlying social contexts and related mental health factors from digital text communication data, researchers must extract a comprehensive corpus of features from raw textual data streams. In this paper, we present a generalized framework for extracting features from digital text communication datasets that leverages methodological approaches from diverse fields. This framework will serve to bridge the gap between quantitative and qualitative research approaches to analyzing digital text communications with respect to mental health.

Sepsis is a dysregulated systemic response to infection that causes organ dysfunction. It is the leading cause of in-hospital mortality and is responsible for the longest, most expensive, hospital stays in the United States. Prognostic modeling of sepsis has demonstrated potential for improving treatment and outcomes for critical care patients. However, the majority of previous models have relied on an outdated definition of sepsis based on systemic inflammatory response syndrome (SIRS). Therefore, this study sought to build predictive models of sepsis using the most recent definition of sepsis, Sepsis-3. A total of three classification methods, including logistic regression (LR), support vector machines (SVM), and logistic model trees (LMT), were used to predict onset of sepsis in adult Intensive Care Unit (ICU) patients using vital signs and blood culture results. For patients who did not develop sepsis, predictor values were selected from a random 48-hour time window during the patient's ICU stay. For those who developed sepsis, a random time was selected for the patient within 48 to 6 hours prior to onset of sepsis, and the predictor values with the closest preceding time were extracted. The LMT produced superior classification performance compared with the LR and SVM. The wide time window used for data extraction improves the clinical utility of the models relative to those built in previous work. However, while the models showed similar sensitivity and specificity to previous models using the Sepsis-3 definition, they each showed a high false positive rate.

In America, half of the population experiences mental health issues in their lifetime. Less than one third, however, will receive treatment. The National Institute of Mental Health calls for research on new, scalable methods for treating mental illness. Cognitive Bias Modification (CBM) is used to alter responses to situations that make participants feel anxious or upset. In this paper, we analyze data from an online, session-based, CBM training program for anxiety in order to make recommendations to improve the user experience increasing engagement and adherence to the program. Our objectives include maximizing user value, decreasing attrition, and minimizing user burden. To meet our objectives, we are analyzing data related to attrition and user engagement and implementing principles of user experience design such as improved visuals and typography and elements of gamification. Preliminary analyses indicate that the highest number of dropouts were less educated participants leaving the program and the highest rates of attrition were shown towards the beginning of the sessions. Additionally, a redesigned interface will be deployed and future data will be collected to determine its effect on attrition.

An estimated 266,120 women in the United States will be diagnosed with breast cancer in 2018. A mastectomy is recommended in over a third of breast cancer patients, with the option to undergo breast reconstruction. Despite 90,000 breast reconstruction surgeries being performed annually, there is limited quantitative patient-centered recovery information to guide patient decision-making. The primary objective of this paper is to improve recovery for breast reconstruction patients. Preoperatively, the objective is to inform decisions on breast reconstruction options and postoperatively, the objective is to improve the recovery experience. Our study seeks to expand understanding of recovery through the deployment of actigraphy and daily surveys. We outline a preliminary study of four patients and found that patient-generated data can be employed to measure recovery projections across patients with acceptable compliance. We have not found statistically significant differences in time to normal between the types of reconstruction. However, our preliminary results indicate significant linear trends in the recovery of our patients as measured by various predictors. A larger sample size could be beneficial when exploring predictors of recovery.

Negative affect is a proxy for mental health in adults. By being able to predict participants' negative affect states unobtrusively, researchers and clinicians will be better positioned to deliver targeted, just-in-time mental health interventions via mobile applications. This work attempts to personalize the passive recognition of negative affect states via group-based modeling of user behavior patterns captured from mobility, communication, and activity patterns. Results show that group models outperform generalized models in a dataset based on two weeks of users' daily lives., BHI2018

Social anxiety disorder affects approximately 7% of the adult population in the U.S., yet a vast majority of these individuals do not seek treatment. Thus, it is critical to examine models that deliver treatment to them. Computerized Cognitive Bias Modification (CBM) training programs can be effective in targeting interpretation bias, a key cognitive mechanism underlying social anxiety, and have potential for widespread dissemination, especially if they can be delivered via smart phones, which are becoming ubiquitous. However, the efficacy of CBM interpretation training paradigms that are adapted to and delivered via smart phones remains unknown. We present a pilot study to investigate if physiologic data can be used to track the changes over a smartphone-based CBM intervention for social anxiety. In a 3-week open trial, pilot study involving 20 high socially anxious participants, self-report affect ratings, heart rate and accelerometer data were collected using a smartphone and smartwatch before, after, and during the CBM intervention. The study focused on the relationship between accelerometer and heart rate to track change following the intervention. Results provide preliminary evidence for the viability of using physiological data to identify the change in mental state influenced by CBM interventions.

The increasing adoption of electronic health record (EHR) systems has brought tremendous opportunities in medicine enabling more personalized prognostic models. However, most work to date has investigated the binary classification problem for predicting the onset of one chronic disease, but little attention has been given to assessing risk of developing comorbidities that are major causes of morbidity and mortality. For example, type 2 diabetes and chronic kidney disease frequently accompany congestive heart failure. This paper is motivated by the problem of predicting comorbid diseases and aims to answer the following question: can we predict the comorbid risk using a patient's medical history? We propose a new predictive learning framework, Heterogeneous Convolutional Neural Network (HCNN), that represents EHRs as graphs with heterogeneous attributes (e.g. diagnoses, procedures, and medication), and then develop a novel deep learning methodology for risk prediction of multiple comorbid diseases. The main innovation of the framework is that it defines the distance between the heterogeneous attributes of the graph representation extracted from the EHR and develops an appropriate learning infrastructure that is a composition of sparse convolutional layers and local pooling steps that match with the local structure of the space of the heterogeneous attributes. As a result, the new method is capable of capturing features about the relationships between heterogeneous attributes of the graphs. Through a comparative study on patient EHR data, HCNN achieves better performance than traditional convolutional neural networks on the risk prediction of comorbid diseases.

The incidence of type 2 (insulin resistant) diabetes (T2D) is increasing, along with its complications of personal anguish and treatment costs. The American Diabetes Association recommends Lifestyle Modification (LM) both for the prevention and management of T2D. The basic premise of LM is that energy intake should be equal to energy output, or net carbohydrates consumed should roughly equal energy “burned” by physical activity to avoid hyperglycemia and its consequences. While these recommendations support healthy habits, lack of knowledge and motivation often result in “forgetting” to execute intended behaviors. Mobile technologies such a smartphones and wearables have the potential to educate, motivate and prompt individuals in situ to optimize these LMs and thereby favorably affect health. This paper presents a framework using wearable activity trackers and smartphones to promote positive behaviors such as reducing sedentary activity and promoting physical activity after meals. To evaluate the feasibility of the proposed approach, the technology and supporting algorithms were analyzed and evaluated in a group of healthy students. Specifically, we developed an eating detection algorithm fusing accelerometer and gyroscope data streams from a smart watch. We developed the proposed algorithm in a controlled laboratory study, and we present the preliminary results of our eating detection algorithm from the experiment.

Approximately 1 in 8 U.S. women will be diagnosed with new invasive breast cancer over the course of her lifetime. An estimated 252,710 new cases of invasive breast cancer are expected to be diagnosed in women in the U.S. in 2017. Mastectomy is recommended in over a third of early-stage breast cancer patients. Those women who elect to undergo breast reconstruction are counseled on surgical risks and benefits of implant-based and autologous reconstruction. Currently, there is limited patient-centered information about course of recovery, which is a major consideration when deciding between types of reconstruction. Patient recovery estimates are often anecdotally related to the length or invasiveness of the surgical procedure rather than patient-centered, evidence-based data on implant versus autologous surgical recovery. This deficit in information on patient recovery comes at a time when real-time digital devices are used to track vitals, sleep-wake cycles, and steps taken for personal convenience without being used to guide treatment. The primary objective of this paper is to present a system framework for modeling the surgical recovery process. We also outline preliminary results from a pilot study with two breast cancer patients who each underwent one of the two reconstruction surgeries.

This paper aims to improve health outcomes research and data management practices. Typically health care records are very large and cumbersome to manage, and the quality of the data is often overlooked because the volume is thought to be large enough to overcome issues arising from missing data. However, simply removing observations with missing data is problematic because the distribution of missing information is non-random, thus the sample used for analysis becomes biased. We propose a method for evaluating and addressing bias in the data cleaning process. Specifically, we identify where bias exists within data and address the bias using sub-sampling or discarding data. We present a case study analyzing data from a level 1 trauma center to establish how bias in health registries exists and how this bias can have downstream implications for evaluating hospital performance. Our method utilizes a two-tailed z-test to compare subgroups in the data set, which demonstrates how missing data in these subgroups can lead to bias. We demonstrate how to localize the bias in particular subgroups and provide corrective actions to handle the bias. We also exhibit how failure to account for bias can distort performance, illustrating the importance of the proposed method.

The availability of Electronic Health Records (EHR) in health care settings provides terrific opportunities for early detection of patients' potential diseases. While many data mining tools have been adopted for EHR-based disease early detection, Linear Discriminant Analysis (LDA) is one of the most widely-used statistical prediction methods. To improve the performance of LDA for early detection of diseases, we proposed to leverage CRDA — Covariance-Regularized LDA classifiers on top of diagnosis-frequency vector data representation. Specifically, CRDA employs a sparse precision matrix estimator derived based on graphical lasso to boost the accuracy of LDA classifiers. Algorithm analysis demonstrates that the error bound of graphical lasso estimator can intuitively lower the misclassification rate of LDA models. We performed extensive evaluation of CRDA using a large-scale real-world EHR dataset — CHSN for predicting mental health disorders (e.g., depression and anxiety) in college students from 10 US universities. We compared CRDA with other regularized LDA and downstream classifiers. The result shows CRDA outperforms all baselines by achieving significantly higher accuracy and F1 scores.

Prior research demonstrates environmental reservoirs may play a role in the transmission of antibiotic-resistant bacteria within a hospital. Using video monitoring, we propose a technique based on graph Laplacian spectral analysis to classify human activities around sinks. The extracted feature-vectors are applied to a support vector machine to characterize different events in each video. Results are promising, demonstrating the potential of graph Laplacian spectral decomposition for the vision-based analysis of human activity. The activities mined from the videos in correlation with localized statistics of infections could lead to an understanding of the link between infection transmission and healthcare activities.

Sepsis is the systemic response to infection complicated by organ dysfunction. It is a leading cause of in-hospital mortality, with an observed mortality of 20–40%, and is associated with significantly high costs. Furthermore, patients who survive sepsis are more likely to have permanent organ damage, cognitive impairment and physical disability. Survival from sepsis is dependent on early and targeted treatment. Accepted methods for identifying and predicting sepsis have typically been based on intensive care unit (ICU) patients and founded upon a limited and outdated definition of sepsis based on systemic inflammatory response syndrome (SIRS) criteria — one which is not clinically validated and shown to miss at least 1 out of 8 cases. Recently updated consensus guidelines on sepsis have been published and offer the possibility of improved EWS and predictive models. Our goal was to identify non-ICU ward patients with sepsis earlier and more accurately using the newly established sepsis definition and improved predictive models. Using multivariate logistic regression and routinely available physiological and laboratory data from electronic health records (EHRs), we derived an EWS that identifies at-risk ward patients 12–24 hours prior to sepsis onset with an area under the receiver operating characteristic curve (AUC) result of 74%. This model, when applied on a separate ICU population, achieved an AUC curve result of 56%, indicating the model has worse performance in this setting.

The healthcare field is facing new challenges and opportunities with an upsurge of telehealth and remote monitoring solutions. Hospitals are working with the private industry to develop web-based solutions that augment primary care once patients leave the clinical setting. As technology enables healthcare to shift from treatment centers to home environments, new approaches are needed to tailor health-related content and delivery at the individual level. This paper presents methods for personalizing communications for a web-based behavioral change support system. Research has shown patients are more likely to continually use systems that present personally relevant content closely resembling traditional healthcare. We performed a two-part feasibility study to investigate the hypothesis that personalized content leads to higher continued use of a telehealth system. In the study, participants interacted with a web-based application that presents questions and information concerning exercise. Logistic regression and collaborative filtering were used to develop personalized content strategies for participants in the second part of the study. We successfully implemented the content strategies in an existing third-party software system, but the study was small and lacked proper power to draw significant, externally valid results.

The Centers for Disease Control and Prevention (CDC) estimate that more than two million people are sickened every year with antibiotic-resistant bacterial infections, and at least 23,000 die as a result of these infections in the U.S. alone. Carbapenem-resistant Enterobacteriaceae (CRE) are a particularly important subgroup with the ability to share genetic material, including the gene responsible for the production of carbapenemase enzyme. As a major threat to public health, CRE primarily affects sick patients in acute long-term health care facilities, such as Intensive Care Units. This study uses clinical data from a major U.S. hospital, as well as survival analysis methods to estimate time-to-colonization, patient risk of CRE colonization, and risk factors. A survival model was developed using Cox proportional-hazards regression to estimate time of infection. Additionally, we also used Kaplan-Meier estimation to approximate survival time. Our results support CDC's understanding of CRE infection. Sick patients with a history trauma and of staying in an ICU, which undergo certain procedures such as dialysis, fluoroscopy, respiratory intubation and mechanical ventilation, are more likely to get colonized by CRE. The fit was evaluated using the likelihood ratio test and we have found all our models to be significant.

This work seeks to design and develop a remote monitoring platform for tracking indicators of sepsis with the goal of reducing hospital readmissions and mortality rates, and ultimately improving outpatient quality of care. Medicare reports spending $17.8 billion per year on avoidable readmissions, a leading cause of which is sepsis. Insufficient knowledge and improper training for postoperative care of discharged patients further enhances the importance of sepsis monitoring at home. Common indicators of sepsis include infection, elevated heart rate, elevated breathing rate, blood pressure, and fever. While these criteria can be easily tracked within the hospital, there exists no straightforward extension to the home setting because of the confounding effect of physical activity. Some of the challenges associated with remote monitoring for sepsis in discharged patients include the lower sensory quality of mobile biometric devices, the need to recognize activity in non-stationary patients, and the need to individualize the baseline criteria. In order to broaden the criteria to outpatients, this research considers the challenges of collecting and analyzing patient biometrics remotely using commercially available technology. As a result, we present a remote monitoring solution for early detection of sepsis in the outpatient setting.

Normalized cross-correlation template matching is used as a detection method in many scientific domains. To be practical, template matching must scale to large datasets while handling ambiguity, uncertainty, and noisy data. We propose a novel approach based on Dempster-Shafer (DS) Theory and MapReduce parallelism. DS Theory addresses conflicts between data sources, noisy data, and uncertainty, but is traditionally serial. However, we use the commutative and associative nature of Dempster's Combination Rule to perform a parallel computation of DS masses and a logarithmic hierarchical fusion of these DS masses. This parallelism is particularly important because additional data sources allow DS-based template matching to maintain accuracy and refine uncertainty in the face of noisy data. We validate the parallelism, accuracy, and uncertainty of our implementation as a function of the size and noise of the input dataset, finding that it scales linearly and can retain accuracy and improve uncertainty in the face of noise for large datasets.

Currently, a large amount of data is amassed in electronic health records (EHRs). However, EHR systems are largely information silos, that is, uses of these systems are often confined to management of patient information and analytics specific to a clinician's practice. A growing trend in healthcare is combining multiple databases to support epidemiological research. The College Health Surveillance Network is the first national data warehouse containing EHR data from 31 different student health centers. Each member university contributes to the data warehouse by uploading select EHR data including patient demographics, diagnoses, and procedures to a common server on a monthly basis. In this paper, we focus on the data quality dimensions from a subsample of the data comprised of over 5.7 million patient visits for approximately 980,000 patients with 4,465 unique diagnoses from 23 of those universities. We examine the data for measures of completeness, consistency, and availability for secondary use for epidemiological research. Additionally, clinical documentation practices and EHR vendor were evaluated as potential contributors to data quality. We found that overall about 70% of the data in the data warehouse is available for secondary use, and identified clinical documentation practices that are correlated to a reduction in data quality. This suggests that automated quality control and proactive clinical documentation support could reduce ad-hoc data cleaning needs resulting in greater data availability for secondary use.

According to a 2014 Spring American College Health Association Survey, almost 50% of college students reported feeling things were hopeless and that it was difficult to function within the last 12 months. More than 80% reported feeling overwhelmed and exhausted by their responsibilities. This critical subpopulation of Americans is facing significant levels of mental health disorders, challenging colleges to provide accessible and high quality behavioral health care. However, psychiatric disorders are frequently unrecognized in primary care settings, posing physical, emotional, economic, and social burdens to patients and others. Towards the goal of earlier identification and treatment of mental health disorders, this paper proposes M-SEQ, an early detection framework for anxiety/depression using electronic health data from primary care visit sequences. Specifically, compared to existing methods that predict a future disease state using frequency of diagnoses in a patient's medical history, we hypothesize that future disease might also be correlated with the temporal orders of diagnoses. Thus, M-SEQ first discovers a set of diagnosis codes that are discriminative of anxiety/depression, and then extracts each diagnosis pair from each patient's health record to represent the temporal orders of diagnoses. Further, it incorporates the extracted temporal order information with the existing representation to predict whether a patient is at risk of anxiety/depression. We evaluate M-SEQ using the electronic health record (EHR) data of 213,112 college students from 10 schools participating in the College Health Surveillance Network (CHSN) from January 1, 2011 through December 31, 2014. The experimental results shows that our framework can detect a future diagnosis of anxiety and depression based on the primary care visit data up to 3 months in advance, with approximately 1%-4.5% higher accuracy, compared to baseline methods using frequency of diagnoses.

Intensive Care Unit (ICU) patients have significant morbidity and mortality, often from complications that arise during the hospital stay. Severe sepsis is one of the leading causes of death among these patients. Predictive models have the potential to allow for earlier detection of severe sepsis and ultimately earlier intervention. However, current methods for identifying and predicting severe sepsis are biased and inadequate. The goal of this work is to identify a new framework for the prediction of severe sepsis and identify early predictors utilizing clinical laboratory values and vital signs collected in adult ICU patients. We explore models with logistic regression (LR), support vector machines (SVM), and logistic model trees (LMT) utilizing vital signs, laboratory values, or a combination of vital and laboratory values. When applied to a retrospective cohort of ICU patients, the SVM model using laboratory and vital signs as predictors identified 339 (65%) of the 3,446 patients as developing severe sepsis correctly. Based on this new framework and developed models, we provide a recommendation for the use in clinical decision support in ICU and non-ICU environments.

Poor communication in critical care settings has the potential to adversely impact patient outcomes and clinical team performance. Hence, improvements to clinical communication systems are continually sought. However, evaluating these modifications and assessing the risk of degraded performance are complex issues, which are typically addressed qualitatively. These approaches typically examine individual perceptions and can fail to capture the actual performance of the modified processes in the communication system. In order to avoid these issues and effectively analyze the changes in a given system, a Monte Carlo simulation methodology is proposed using data extracted from Electronic Medical Records (EMR) compiled by the Huron Consulting Group to precisely measure modified system performance. We couple a risk analysis, using control charts, with a temporal distribution analysis to determine whether to halt using a new system due to failures or determine if more observations are needed to make an adequate assessment. Specifically, we use EMR data to apply this methodology and show the tradeoff between sample size and the likelihood that the data converges to the current state of communication. This methodology will enable researchers to expedite the assessment of the impact of a technology, which is critical in a complex system such as the hospital.

Patient acquisition of carbapenem resistant bacteria in hospitals is a serious problem that leads to adverse outcomes for infected patients. The most common carbapenem resistance mechanism in US hospitals is a mobile gene called Klebsiella pneumoniae carbapenemase (KPC), which can move between bacterial species. Previous research has demonstrated that patient-to-patient transmission cannot fully account for all cases of KPC-positive bacterial transmission in hospital settings. The study in this paper considers environmental reservoirs as an additional source of transmission. We conduct this analysis with logistic regression and random forest models of nosocomial transmission that includes both environmental factors and clinical data. Results show that infection rates are not uniformly distributed throughout the hospital and that differences in room layout play a role. However, proximity to KPC-positive Enterobacteriaceae environmental reservoirs does not significantly correlate to patient acquisition.

Visualization of health data can increase situational awareness, maximize utility of patient data, and improve the clinical decision-making processes in critical care settings. Many hospitals are burdened by vast quantities and varieties of data and are struggling to manage, analyze, interpret, and present it in a meaningful way. Graphical visualization and predictive modeling are effective mechanisms to aid clinicians in assessing patient status and recognizing critical changes over time. In an analysis of the UVa Hospital’s Surgical Trauma Burn Intensive Care Unit (STBICU), we examined the impacts that technology and data have on workflow and the value added to various stakeholders. An online survey was distributed to Attending Physicians, Residents, Nurse Practitioners, and Nurses. The goal of this survey was to better understand the ways that various clinical positions interact with data visualization technology regarding patient diagnosis and general workflow. The survey results indicated that graphical representations are under-utilized compared to spreadsheets and progress notes. Additionally, clinicians consider current information technology to be sufficient for evaluating patients’ past and current health data, but not for future health status. Data visualization technology and predictive models provide an alternative monitoring solution to overcome existing system limitations.

The detection rate of adenomas in screening colonoscopies is an important quality indicator for endoscopists. Successful detection of adenomas is linked to reduced cancer incidence and mortality. This study focuses on evaluating the performance of endoscopists on adenoma detection rate (ADR), polyp detection rate (PDR), and scope withdrawal time. The substitution of PDR for ADR has been suggested due the reliance of ADR calculation on pathology reports. We compare these metrics to established clinical guidelines and to the performance of other individual endoscopists. Our analysis (n = 2730 screening colonoscopies) found variation in ADR for 14 endoscopists, ranging from 0.20 to 0.41. PDR ranged from 0.38 to 0.62. Controlling for age, sex, race, withdrawal time, and the presence of a trainee endoscopist accounted for 34% of variation in PDR but failed to account for any variation in ADR. The Pearson correlation between PDR and ADR is 0.82. These results suggest that PDR has significant value as a quality indicator. The reported variation in detection rates after controlling for case mix signals the need for greater scrutiny of individual endoscopist skill. Understanding the root cause of this variation could potentially lead to better patient outcomes.

Efficient and effective communication processes are critical for patient health and safety. Communication processes for patient admission within the UVa Surgical and Trauma Intensive Care Unit (SICU) vary depending on situational factors. Trauma personnel in the SICU are also involved in patient admission to the Operating Room (OR). In order to better understand communication patterns and identify inefficiencies and variation, an analysis of the patient admission processes from the Emergency Department (ED) to the OR and ED to SICU is performed. To identify potential inefficiencies within patient admission, key sub-processes with high variability were identified. Process models were constructed to depict the current practices of each step and the associated communication patterns between stakeholders. Interviews, surveys, and focus groups were also conducted with trauma personnel to gauge opinions on the patient admission processes and to collect estimated waiting times at each step in the processes. Based on findings from interviews, process completion times were compiled and analyzed using Hurwicz Alpha Criterion. Results of the study indicate that communication processes with consults and the bed center, both located outside of the SICU, are variable and inconsistent. This variability in communication is a result of a lack of situational awareness and procedures external to the SICU.

Intelligence gathering is a critical component of military operations. Unmanned aerial vehicles (UAVs) have become an increasingly useful tool due to their surveillance and reconnaissance capabilities. However, the use of many of these vehicles is limited to outdoor environments because of their size and reliance on Global Positioning Satellites (GPS). Knowledge of indoor environments is important so that the risk of entering an unsafe or unknown building can be minimized. This paper describes the development of a spatially aware, autonomous quadcopter that uses an Android control system and functions indoors. The system consists of a laser rangefinder for sensory input, a IOIO microcontroller for data communication across platforms, an autopilot system (APM) for flight control, and an Android phone for mission control. The Android Control and Sensor System (ACSS) is currently being developed by the Department of Defense (DOD), MITRE, and academic partners, and will be integrated into the solution. To be considered autonomous, the quadcopter must be able to make a map from the data provided by the laser rangefinder, determine its own location and position in that map, and then execute a set of navigational commands from the Android control system. The success of this project is measured by the system's ability to travel autonomously while simultaneously creating a map and being aware of its location.

The world's interconnected data assets provide rapid options for individuals looking to communicate and retrieve information, which is especially critical in times of emergency. The ability for populations to get in touch with emergency responders, loved-ones, and retrieve or disseminate critical information about events as they unfold translates into better survivability in the face of crisis. Although telecommunication infrastructures are incapacitated and millions of people experienced hindered Internet access, mobile phone usage soars as people access wireless networks to communicate and seek information regarding an event such as a natural disaster. The demands for communication and information during crisis events distinctly differ from the typical data loads seen during normal operation in both traffic and content. Detecting the anomalies in network traffic data that occur during disruptive events such as earthquakes, hurricanes, and political uprisings can provide potential aid to first responders and be a potentially useful public surveillance tool. Utilizing historic data on network activity and content, a system for assessing the range, intensity, and category of a disruptive event is designed. Systems such as the one described in this manuscript, will detect changes in network traffic caused by disruptive events in real-time.

This paper presents a Rough Set Theory (RST) based classification model to identify hospice candidates within a group of terminally ill patients. Hospice care considerations are particularly valuable for terminally ill patients since they enable patients and their families to initiate end-of-life discussions and choose the most desired management strategy for the remainder of their lives. Unlike traditional data mining methodologies, our approach seeks to identify subgroups of patients possessing common characteristics that distinguish them from other subgroups in the dataset. Thus, heterogeneity in the data set is captured before the classification model is built. Object related reducts are used to obtain the minimum set of attributes that describe each subgroup existing in the dataset. As a result, a collection of decision rules is derived for classifying new patients based on the subgroup to which they belong. Results show improvements in the classification accuracy compared to a traditional RST methodology, in which patient diversity is not considered. We envision our work as a part of a comprehensive decision support system designed to facilitate end-of-life care decisions. Retrospective data from 9105 patients is used to demonstrate the design and implementation details of the classification model.

Chronic diseases such as diabetes and heart disease are the leading causes of disability and death in the developed world. Technological interventions such as mobile applications have the ability to facilitate and motivate patients in chronic disease management, but these types of interventions present considerable design challenges. The primary objective of this paper is to present the challenges arising from the design and implementation of software applications aiming to assist patients in chronic disease management. We also outline preliminary results regarding a self-management application currently under development targeting young adults suffering from type 1 diabetes.

In this paper, we present algorithms and display concepts that allow Soldiers to efficiently interact with a robotic swarm that is participating in a representative convoy mission. A critical aspect of swarm control, especially in disrupted or degraded conditions, is Soldier-swarm interaction-the Soldier must be kept cognizant of swarm operations through an interface that allows him or her to monitor status and/or institute corrective actions. We provide a control method for the swarm that adapts easily to changing battlefield conditions, metrics and supervisory algorithms that enable swarm members to economically monitor changes in swarm status as they execute the mission, and display concepts that can efficiently and effectively communicate swarm status to Soldiers in challenging battlefield environments.

In this paper, we present a novel strategy for organizing groups of robots into a formation utilizing artificial potentials which behave like a bar magnet. A mathematical surface is defined that will “pull” the robots into formation controlling the overall geometry. Nonlinear limiting functions are defined to attract and hold members in a geometric formation. By adjusting control parameters, the shape and the extent of the formation is controlled. Formations can dynamically change and adapt accordingly by making the control parameters time varying. This approach is computationally efficient and scales well to large team sizes. Simulation studies are presented for ten robots in wedge, inverted ‘vee’, and column formations.

This work addresses the problem of coordinating a swarm of unmanned ground vehicles with an unmanned aerial vehicle (UAV). The UAV is utilized as a leader robot that is intelligently followed by a coordinated group of unmanned ground vehicles (UGVs). The UAV is a completely autonomous agent that is controlled using Sugueno fuzzy logic. The UGVs are organized into formation utilizing artificial potential fields generated from normal and sigmoid functions. These functions are built around the location of the UAV and ultimately construct the surface that swarm members travel on, which inherently controls the overall swarm geometry and the individual member spacing. Nonlinear limiting functions are defined to provide tighter swarm control by modifying and adjusting a set of control variables forcing the swarm to behave according to set constraints, formation and member spacing. The swarm function and limiting functions are combined to control swarm formation, orientation, and swarm movement as a whole. Parameters are chosen based on desired formation as well as user defined constraints. Simulations demonstrate the precision of the approach with up to forty UGVs. Experimental results are presented using a fully autonomous swarm of three UGVs and a single UAV helicopter for coordination.

A novel technique is presented for organizing swarms of robots into formation utilizing artificial potential fields generated from normal and sigmoid functions. These functions construct the surface swarm members travel on, controlling the overall swarm geometry and the individual member spacing. Limiting functions are defined to provide tighter swarm control by modifying and adjusting a set of control variables forcing the swarm to behave according to set constraints, formation and member spacing. The swarm function and limiting functions are combined to control swarm formation, orientation, and swarm movement as a whole. Parameters are chosen based on desired formation as well as user defined constraints. This approach compared to others, is simple, computationally efficient, scales well to different swarm sizes, to heterogeneous systems, and to both centralized and decentralized swarm models. Simulation results are presented for a swarm of four and ten particles following circle, ellipse and wedge formations. Experimental results are also included with four unmanned ground vehicles (UGV).

A novel method is presented for dynamic heterogeneous swarm formation control with potential fields generated from bivariate normal probability density functions (pdfs) used to construct the surface which swarm members move on, controlling swarm geometry, individual member spacing, and managing obstacle avoidance. Limiting functions are defined to provide tighter swarm control by modifying and adjusting a set of control variables forcing the swarm to behave according to set constraints. Bivariate normal functions and limiting functions are combined to guarantee obstacle avoidance and control swarm member orientation and swarm movement as a whole. This approach compared to others, is simple, computationally efficient, scales well to different swarm sizes, to heterogeneous systems, and to both centralized and decentralized swarm models. The method is applied to a simple vehicle model and simulation results are presented for a heterogeneous swarm of ten robot vehicles following line and ellipse formations.

A novel method is presented for swarm formation control with potential fields generated from bivariate normal probability density functions (pdfs) that construct the surface the swarm members move upon controlling the swarm geometry and member spacing as well as manage obstacle avoidance. Limiting functions provide tighter swarm control by modifying and adjusting a set of control variables, forcing the swarm to behave according to set constraints. Bivariate normal functions and limiting functions are combined to guarantee obstacle avoidance and control swarm member orientation and swarm movement as a whole. The presented approach, compared to others, is simple, computationally efficient, and scales well to different swarm sizes and swarm models. The method is applied to a simple vehicle model, and simulation results are presented on a homogeneous swarm of ten robot vehicles for different formations.

This study investigates data collected from operating an Inuktun robot in an urban search and rescue (USAR) confined space training exercise task at Virginia Beach Training Center. Data was collected from coding approximately one hour of video. The video had no sound so all analysis is based on the video feed. Indicators of communication, gestures, physical interactions with the robot, and robot movements were analyzed. The findings indicate that the robot emerges as a virtual presence for the support of the team outside of the confined space. The team members spontaneously responded socially to the robot despite the robot not being engineered to have a social intelligence. This confirms numerous studies in the cognitive science, psychology, and affective computing literature that robots need a social interface regards of domain.