Smart Health

One in eleven adults worldwide suffers from diabetes, and the disease accounts for 12% of global health expenditure(http://​www.​idf.​org/​about-diabetes/​facts-figures). Although self-management and monitoring are critical for general control of the disease and for preventing diabetes-related complications, most patients fail to adhere to self-management regimens. We study what type of external intervention will amplify the self-monitoring frequency of Type-2 diabetes (T2d) patients. We conducted a randomized field experiment on a mobile health application with more than 500 T2d patients, and tested two well-known mechanisms for behavior change: social norms, and implementation intentions. Further, we combined social norms and implementation intentions and tested whether these two mechanisms can complement each other. Our results show that individuals who receive a message containing both social norms and implementation intentions perform the best in regard to self-monitoring. Our research paves the way to further investigate how different mechanisms may be combined to help users’ form healthy habits.

This paper investigates the multiple dimensions of patient satisfaction measured by the HCAHPS (Hospital Consumer Assessment of Healthcare Providers and Systems) survey in the United States. The analysis reveals that even the highest rating hospitals do not excel on all dimensions of satisfaction. While satisfaction levels with nurse and doctor communication are high, satisfaction levels with discharge information and explanation of medications could be significantly improved. In addition, low rating hospitals seem to be doing better than the high rating hospitals on these critical dimensions for the quality of care. The paper also investigates how hospital structural characteristics captured in the American Hospital Association (AHA) survey affect different dimensions of patient satisfaction. The analysis reveals that these structural factors may have differential effects, i.e. the type of hospitals (e.g., teaching vs. non-teaching) has a relatively small effect on the hospital room environment than on communication and responsiveness. The results suggest that considering all patient satisfaction dimensions helps provide a more accurate picture of the care received by patients, makes it possible to pinpoint specific areas where hospitals are deficient that are not reflected in the overall satisfaction scores, and assists hospital management to design actionable strategies for improvement.

Seniors with dementia living in residential nursing homes are often lack of meaningful engagement and keeping them engaged in meaningful activities can help reduce boredom and improve their well-being. This paper presents an Interactive Table Design (ITD) for providing seniors with dementia meaningful engagements. An observational engagement assessment method, Video Coding – Incorporating Observed Emotion (VC-IOE), was adopted to further study the effectiveness of the intervention design. Qualitative data such as video recordings of four participants engaged with the ITD and comparison intervention Pim Pam Pet (PPP) in Vitalis Kelinschalig Wonen, were then analyzed following the video analysis protocols of VC-IOE. The results from video coding analysis provided an overview of participants’ emotional responses and engagement situations through six dimensions of engagement including emotional, verbal, visual, behavioral, collective engagement and agitation. The results showed sufficient positive impacts of ITD on participants which indicate that the ITD has the potential to be an effective intervention for providing seniors with dementia with meaningful engagement while keeping them socially connected in a nursing home.

Cancer was reported to be the chief killer in Australia by overtaking heart disease. Early detection of cancer aids in effective treatment. This poster paper proposes the design and implementation of an integrated cancer risk calculation tool to address the problem of diversified and unstandardized calculators for different types of cancer. Guided by cognitive load theory, the paper also proposes design features to reduce patients’ cognitive load.

The Online health community (OHC) provides convenience and benefits. Simultaneously, it also generates informal social moral problems regarding payment. However, few studies have explored the physicians’ service delivery from the patients’ perspective. Using unique interaction data between doctors and patients from an online health consultation community, we empirically examined how patients’ gift giving behavior influences physicians’ service delivery, and how this influence is being moderated by physicians’ gift receiving ratio and social morality, using a difference-in-difference model. We find that the giving of gifts by patients has a positive impact on the physicians’ service quality. Furthermore, this effect is significantly moderated by the physicians’ gift receiving ratio and social morality. We discuss the implications of these findings for patients consulting in OHCs and for OHC designers.

In recent decades, many healthcare service providers embrace social media to spread knowledge to the public and communicate with them. However, elderly people are not paid much attention. As we know, more and more countries have to face the challenge of rapidly aging and it is vital to meet their demands of healthcare such as how to manage their chronic diseases by themselves. Information technology (IT) may provide solutions and studies on IT use in different contexts of aging have emerged. However, research relating IT use to healthcare knowledge sharing and wellbeing of aging people is limited. This study aims to examine relations among social media use, healthcare knowledge sharing, and seniors’ wellbeing in the context of China. For this purpose, we use a mixed method to test our conceptual model, which consists of both quasi-experiment and questionnaire survey conducted with retired elderly people. Our research can give healthcare service providers suggestions about how to use social media to increase seniors’ knowledge of self-management and to improve their wellbeing.

Readmission is often an important indicator of care quality, which also accounts for a major proportion of medical expenses. In this paper, we study the relationship between out-of-pocket ratio in medical cost and readmission. Apart from out-of-pocket ratio, we also consider other factors such as demographic features of patients, medical expenses, and variables about diseases. As there are a large number of diseases and operations, for better interpretation, we adopt data mining method to identify discriminative features. Our study is based on the front sheet of inpatient medical record from 150 hospitals in Beijing from year 2012 to 2014. In the records with primary diagnoses being rectal malignant tumor, we find out that when out-of-pocket ratio is low or high, the readmission is relatively high. Meanwhile, discriminative features found based on group difference mining method are helpful to gain understanding of the results.

This paper proposes using a commodity-based smartwatch paired with a smartphone for developing a fall detection IoT application which is non-invasive and privacy preserving. The majority of current fall detection applications require specially designed hardware and software which make them expensive and inaccessible to the general public. We demonstrated that by collecting accelerometer data from a smartwatch and processing those data in a paired smartphone, it is possible to reliability detect (93.8% accuracy) whether a person has encountered a fall in real-time. By wearing a smartwatch as a piece of jewelry, the well-being of a person can be monitored in real-time at anytime and anywhere as contrasted to being confined in a particular facility installed with special sensors and cameras. Using simulated fall data acquired from volunteers, we trained a fall detection model off-line that can be composed with a data collection accessor to continuously analyze accelerometer data gathered from a smartwatch to detect minor or serious fall at anytime and anywhere. The accessor-based architecture allows easy composition of the fall-detection IoT application tailored to heterogeneity of devices and variation of user’s need.

Rapid advances in information and sensor technology have led to the development of tools and methods for individual health monitoring. These techniques support elderly health management by tracking the vital signs and detecting physiological changes for the target population, such as the elderly and patients with chronic diseases. Two pilot studies were conducted to demonstrate the implementation of electronic wearable wellness devices and an all-in-one station-based health monitoring device at the community level in Hong Kong. Real-time and daily changes of key vital signs in elderly people recruited from a nursing home and a geriatric daycare center were collected. Preliminary analysis of the collected data provided insights into the characteristics of vital signs of the elderly from two centers, which could bring benefits to the management of healthcare services. Additionally, a personalized wellness forecasting system was built to identify the factors influencing the personal wellness of the elderly, by aggregating historical daily vital signs.

Diabetes is a costly chronic disease, and a leading cause of death and disability worldwide.

The emergence of mHealth products has created capability of monitoring and managing health of patients with chronic disease. In this paper, we analyze the participants’ usage of a mobile app named mPower, developed for Parkinson disease. We identify the demographic/usage difference between different groups of participants, which provides insights into better design and marketing of mHealth products.

Alleviating stress reduces the risk of developing many chronic health problems. Though the effects of stress on the body may not always be immediately evident, exposure to chronic stress can lead to serious health problems and/or exacerbate existing medical conditions. This research study explores how a personal computing device such as a smartphone can be used to provide information regarding individually tailored stress management activities for college students. Since the use of smartphones is pervasive, one way to address this issue would be to develop a smartphone application in which a user can monitor stress as well as obtain various interventions for stress management. The proposed stress management application is based on information obtained from the user regarding stress type and intensity. An application that provides recommendations for stress-relieving activities can have a positive impact on a student’s health and well-being.

The segmentation of motion capture data is to separate the different types of human motion data contains long movement sequence into motion clips with independent semantics in order to facilitate the storage in the database as well as medical analysis. This paper proposed a method for human motion capture data segmentation based on Laplacian Eigenmaps (LE) algorithm. Firstly, the LE algorithm is used to reduce the dimension of original data by realizing the mapping from the high dimensional data to the low dimensional space. And then a specified window was drawn in the low dimensional space which was used to calculate the space distance from frames in the specified window to each frame in the former fragment. Finally we detected the similarity to get the final segmentation points, thus obtained motion clips with independent semantics. The validity of the segmentation method is verified by experiment.

In order to explore the effects of the externality of public goods in the online healthcare domain, we investigate the relationship between the contributions to Q&A (public goods) and the private benefits of family doctors based on the theory of public goods and externality. We analyze a panel dataset of 1,323 doctors from an online healthcare community, and our results show that participation in public goods will significantly increase the private benefits of family doctors. Moreover, we also find that the physician’s ranking has a moderating effect on this relationship.

Patients are increasingly using online reviews to choose physicians. However, it is not known whether online reviews accurately capture the true quality of care provided by physicians. This research addresses this issue by empirically examining the link between online reviews of a physician and the actual clinical outcomes of patients treated by the physician. Specifically, this study uses online reviews from Vitals.com, and combines that data with patient health outcomes data collected from Dallas-Fort Worth Hospital Council. Our econometric analyses show that there is no clear relationship between online reviews of physicians and their patients’ health outcomes, such as readmission and ER visit rates. Our results imply that online reviews may not be as helpful in the context of healthcare as they are for other experience goods such as books, movies, or hotels. Our findings have important implications for healthcare providers, healthcare review websites, and healthcare consumers.

Online health communities (OHCs) have become increasingly popular for people with health issues in China, which have been regarded as one of the major sources of social support. In this study, we designed a Chinese content analysis process to understand social support and user engagement in OHCs. Based on the social support theory, the process used Chinese text mining and machine learning techniques. Using a case study of an OHC among diabetics, we first divided users’ posts and replies into different types of social support. Then, we aggregated each user’s texts of different social support types. At last, we revealed the roles of the users by clustering. Considering the high dimensions of Vector Space Model (VSM) transformed from user texts, we proposed a new method to extract features based on LDA. In order to improve the effect of user clustering, we optimized the clustering algorithm with the principle of Maximum Distance and Elbow Method. Results showed that the process performed well in classification and clustering.

Mining user intents in online interactive behavior from social media data can effectively identify users’ motives behind communication and provide valuable information to aid medical decision-making and improve services. However, it is a challenging task due to the ambiguous semantic, irregular expressions and obscure intention classification categories. In this paper, we first define user intent categories based on speech act theory. On the basis of this, we develop a novel method to further classify users’ utterances according to their pragmatic functions. First, we design topic independent features by regularizing the utterance and categorizing the textual features. Then, we build a hierarchical model based on Hidden Markov Model (HMM) [1] to mine user intents in context sequence at both sentence and microblog level. Finally, we construct a dataset of microblogs about hot topics related to the medical event by a semi-automatic method. Experimental study shows the effectiveness of our method.

People with chronic illnesses are engaged in the lifelong management of their disease and with this management comes the need for emotional support, informational support, and companionship from others, particularly those who have the same disease and can relate to their situation. More often people are seeking this social support through online health communities. The concern is that online health communities may not have a positive impact on the health condition since anyone can be providing information and the information provided may not be valid. We utilize an objective measure of the health condition to determine the impact social support, given and received, in an online health community has on the user’s health condition over time.

Glaucoma is a group of eye diseases that damage the optic nerves progressively and lead to deterioration in vision irreversibly. Diagnosing glaucoma based on retinal images automatically is meaningful both in practice and research area. While deep learning models have achieved superior performance in natural images recognition and have been also used for medical image diagnosis recently, the models usually rely on large dataset and expensive computing resources, thus limiting the wider use in medical areas. So how to train a deep learning model with relatively small amount of medical data is challenging. In this paper, we propose to incorporate domain knowledge to construct a two-branch Convolutional Neural Networks (CNN) to learn a classifier for glaucoma diagnosis based on the retinal image. Our two-branch CNN framework can analyze the whole image and pay special attention to discriminative local region of image at the same time. Experiments conducted on real medical dataset demonstrate the advantages of our method over traditional computer vision algorithm and classical CNN.

We presented and evaluated three sparsity learning based regression models with application to the automated prediction of the Mini-Mental State Examination (MMSE) scores in Alzheimer’s disease(AD) using T1-weight magnetic resonance images (MRIs) from 678 subjects, including 190 healthy control (HC) subjects, 331 mild cognitive impairment (MCI) subjects, and 157 AD subjects. The raw features were obtained from a validated multi-granularity whole-brain analysis pipeline, providing multi-level whole-brain segmentation volumes. We employed the ridge, lasso, and elastic-net as our regression algorithms, with the whole-brain volumes at each level being the independent variables and the MMSE score being the dependent variable. We used 10-fold cross-validation to evaluate the prediction performance and another 10-fold inner loop to estimate the optimal parameters in each model. According to our results, the combination of elastic-net and the second level of whole-brain segmentation volumes (a total of 137 volumes) worked the best compared to all other possible combinations. The work presented in this paper provides a potentially powerful and novel non-invasive biomarker for AD.

Convolutional Neural Network has shown great success in many areas. Different from the hand-engineered feature based classification, Convolutional Neural Network uses self-learned features from data for classification. Recently, some progress has been made in the area of Convolutional Neural Network based lung nodule detection. This paper gives a brief introduction to the problems in such area reviews the recent related results, and concludes the challenges met. Besides some technical details, we also introduce some available public packages for a fast development and some public data sources.

Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection. Early antibiotic therapy to patients with sepsis is necessary. Every hour of therapy delay could reduce the survival chance of patients with severe sepsis by 7.6%. Certain biomarkers like blood routine and C-reactive protein (CRP) are not sufficient to diagnose bacterial sepsis, and their sensitivity and specificity are relatively low. Procalcitonin (PCT) is the best diagnostic biomarker for sepsis so far, but is still not effective when sepsis occurs with some complications. Machine learning techniques were thus proposed to support diagnosis in this paper. A backpropagation artificial neural network (ANN) classifier, a support vector machine (SVM) classifier and a random forest (RF) classifier were trained and tested using the electronic health record (EHR) data of 185 critically ill patients. The area under curve (AUC), accuracy, sensitivity, and specificity of the ANN, SVM, and RF classifiers were (0.931, 90.8%, 90.2%, 91.6%), (0.940, 88.6%, 92.2%, 84.3%) and (0.953, 89.2%, 88.2%, 90.4%) respectively, which outperformed PCT where the corresponding values were (0.896, 0.716, 0.952, 0.822). In conclusion, the ANN and SVM classifiers explored have better diagnostic value on bacterial sepsis than any single biomarkers involve in this study.

Drug safety surveillance plays a significant role in supporting medication decision-making by both healthcare providers and patients. Extracting adverse drug events (ADEs) from social media provides a promising direction to addressing this challenging task. Prior studies typically perform lexicon-based extraction using existing dictionaries or medical lexicons. While those approaches can capture ADEs and identify risky drugs from patient social media postings, they often fail to detect those ADEs whose descriptive words do not exist in medical lexicons and dictionaries. In addition, their performance is inferior when ADE related social media content is expressed in an ambiguous manner. In this research, we propose a research framework using advanced natural language processing and deep learning for high-performance ADE extraction. The framework consists of training the word embeddings using a large medical domain corpus to capture precise semantic and syntactic word relationships, and a deep learning based named entity recognition method for drug and ADE entity identification and prediction. Experimental results show that our framework significantly outperforms existing models when extracting ADEs from social media in different test beds.

Drug interactions represent adverse effects when employing two or multiple drugs together in treatments. Adverse effects are critical and may be deadly in medical practice. However, our understanding of drug interactions is far from complete. In the medical study on drug interaction, the prediction of potential drug interactions will help reducing the experimental efforts. In this paper, we extend a hinge-loss Markov Random Field Model and propose hybrid model of it and logistic regression. In the model, we combine multiple types of chemical and biological evidence to infer the interactions between drugs. Logistic regression is used to learn weights of those evidences. Experiments shows that our approach achieves better performance than the state-of-the-art approaches on both prediction accuracy and time efficiency.

Management and mitigation of epidemic outbreaks is a major challenge for health-care authorities and governments in general. In this paper, we first give a formal definition of a strategy for dealing with epidemics, especially in heterogeneous urban environments. Different strategies target different demographic classes of a city, and hence have different effects on the progression and impact of an epidemic. One has to therefore choose among various competing strategies. We show how the relative merits of these strategies can be compared against various metrics.

We demonstrate our approach by developing a tool that has an agent based discrete event simulator engine at its core. We believe that such a tool can provide a valuable what-if analysis and decision support infrastructure to urban health-care authorities for tackling epidemics. We also present a running example on an influenza-like disease on synthetic populations and demographics and compare different strategies for outbreaks.

Disease transmission network can provide important information for individuals to protect themselves and to support governments to prevent and control infectious diseases. Current studies on disease transmission network mostly focus on scenarios in small, confined areas. We propose to construct disease transmission network using health status time series computed based on health insurance claims. We adopted Granger causality tests to identify potential links from the health status time series from all pairs of individuals. We evaluated our approach by predicting future health care seeking activates for similar diseases based on past health care seeking activates of neighbors in the disease network. The results suggest that the transmission network is able to improve prediction performance in a small random sample of 500 individuals.

Complex networks provide quantitative measures for complex systems, thus enabling effective semantic network analysis. This research aims to develop semantic relevance analysis methods for medical information retrieval to answer questions for clinical decision support system. We proposed a query based semantic expansion network for semantic relevance analysis in medical information retrieval tasks. Empirical studies of the network structure and attributes for discriminant relevance analysis revealed that expansion networks for relevant documents have a compact structure, which provides new features to identify relevant documents. We also found the existence of densely connected nodes as hubs in the associative networks for queries. Then, we proposed a novel rescaled centrality measure to evaluate the importance of query concepts in the semantic expansion network. Experiments with real-world data demonstrated that the proposed measure is able to improve the performance for relevance analysis.

The medical burden of AIDS is a significant public health problem. However, it is affected by the multiple factors, among which there is yet some vague cognition, and further exploration is necessary. Thus, the artificial neural network (ANN) and restricted Boltzmann machine (RBM) be treated as the infrastructure of deep neural networks (DNN), mainly based on the features of demography, pathology and clinical manifestation of AIDS patient’s medical records to mine the impact factors of AIDS cost. And the proposed model could bring to light the previously uncharted latent knowledge and concepts. Based on reliable healthcare delivery, to inhibit the number of hospital days, intensive care and hospitalized frequency plus other sensitive factors, and avoid secondary infection and exposure to allergic reactions can obviously reduce the AIDS cost.

With the quickly increasing of the Question Answering (QA) corpus, the health QA systems provide a convenient way for patients to provide instant service, and the effectiveness of the answer is a very important and challenging problem to be solved. Therefore, this paper proposes a solution based on medical knowledge base. In the process of generating answers, we utilize the entity set provided by medical knowledge base to calculate the correlation between answers and questions, at the same time we make use of the entities provided by relationships in the knowledge base but not appearing in the answers. Experiment conducted on a real data set in our HealthQA system shows that our method can effectively improve the relevance and accuracy of answer matching by using the medical knowledge base.