Population analysis of mortality risk: Predictive models from passive monitors using motion sensors for 100,000 UK Biobank participants.

Many studies have utilized physical activity for predicting mortality risk, using measures such as participant walk tests and self-reported walking pace. The rise of passive monitors to measure participant activity without requiring specific actions opens the possibility for population level analysis. We have developed novel technology for this predictive health monitoring, using limited sensor inputs. In previous studies, we validated these models in clinical experiments with carried smartphones, using only their embedded accelerometers as motion sensors. Using smartphones as passive monitors for population measurement is critically important for health equity, since they are already ubiquitous in high-income countries and increasingly common in low-income countries. Our current study simulates smartphone data by extracting walking window inputs from wrist worn sensors. To analyze a population at national scale, we studied 100,000 participants in the UK Biobank who wore activity monitors with motion sensors for 1 week. This national cohort is demographically representative of the UK population, and this dataset represents the largest such available sensor record. We characterized participant motion during normal activities, including daily living equivalent of timed walk tests. We then compute walking intensity from sensor data, as input to survival analysis. Simulating passive smartphone monitoring, we validated predictive models using only sensors and demographics. This resulted in C-index of 0.76 for 1-year risk decreasing to 0.73 for 5-year. A minimum set of sensor features achieves C-index of 0.72 for 5-year risk, which is similar accuracy to other studies using methods not achievable with smartphone sensors. The smallest minimum model uses average acceleration, which has predictive value independent of demographics of age and sex, similar to physical measures of gait speed. Our results show passive measures with motion sensors can achieve similar accuracy to active measures of gait speed and walk pace, which utilize physical walk tests and self-reported questionnaires.

Evaluation of Machine Learning Models for Classifying Upper Extremity Exercises Using Inertial Measurement Unit-Based Kinematic Data.

The amount of home-based exercise prescribed by a physical therapist is difficult to monitor. However, the integration of wearable inertial measurement unit (IMU) devices can aid in monitoring home exercise by analyzing exercise biomechanics. The objective of this study is to evaluate machine learning models for classifying nine different upper extremity exercises, based upon kinematic data captured from an IMU-based device. Fifty participants performed one compound and eight isolation exercises with their right arm. Each exercise was performed ten times for a total of 4500 trials. Joint angles were calculated using IMUs that were placed on the hand, forearm, upper arm, and torso. Various machine learning models were developed with different algorithms and train-test splits. Random forest models with flattened kinematic data as a feature had the greatest accuracy (98.6%). Using triaxial joint range of motion as the feature set resulted in decreased accuracy (91.9%) with faster speeds. Accuracy did not decrease below 90% until training size was decreased to 5% from 50%. Accuracy decreased (88.7%) when splitting data by participant. Upper extremity exercises can be classified accurately using kinematic data from a wearable IMU device. A random forest classification model was developed that quickly and accurately classified exercises. Sampling frequency and lower training splits had a modest effect on performance. When the data were split by subject stratification, larger training sizes were required for acceptable algorithm performance. These findings set the basis for more objective and accurate measurements of home-based exercise using emerging healthcare technologies.

Population measurement for health systems.

How can health systems make good use of digital medicine? For healthcare infrastructure, the answer is population measurement, monitoring people to compute status for clustering cohorts. In chronic care, most effective is measuring all the time, to track health status as it gradually changes. Passive monitors run in the background, without additional tasks to activate monitors, especially on mobile phones. At its core, a health system is a "sorting problem". Each patient entering the system must be effectively sorted into treatment cohorts. Health systems have three primary problems: Case Finding (which persons have which diagnoses), Risk Stratification (which persons are which status), and Care Routing (which persons need which treatments). The issue is then which measures can be continuously monitored at appropriate periodicity. The solutions of population measurement measure vital signs with passive monitors. These are input to predictive analytics to detect clinical values for providing care within health systems. For chronic care, complex vitals must be measured for overall status, such as oxygen saturation or gait speed. This enables healthcare infrastructure to support stratification, with persons placed into current levels of health status. Practical considerations for health systems influence implementation of new infrastructure. Case finding is more likely to be useful in urban settings, with barriers to entry based upon lower incomes. Care routing is more likely to be useful in rural settings, with barriers to entry based upon isolated geographies. Viable healthcare at acceptable quality and affordable cost is now possible for the range of geographies and incomes.

Accelerometer-based predictive models of fall risk in older women: a pilot study.

Current clinical methods of screening older adults for fall risk have difficulties. We analyzed data on 67 women (mean age = 77.5 years) who participated in the Objective Physical Activity and Cardiovascular Health (OPACH) study within the Women's Health Initiative and in an accelerometer calibration substudy. Participants completed the short physical performance battery (SPPB), questions about falls in the past year, and a timed 400-m walk while wearing a hip triaxial accelerometer (30 Hz). Women with SPPB ≤ 9 and 1+reported falls (n = 19) were grouped as high fall risk; women with SPPB = 10-12 and 0 reported falls (n = 48) were grouped as low fall risk. Random Forests were trained to classify women into these groups, based upon traditional measures of gait and/or signal-based features extracted from accelerometer data. Eleven models investigated combined feature effects on classification accuracy, using 10-fold cross-validation. The models had an average 73.7% accuracy, 81.1% precision, and 0.706 AUC. The best performing model including triaxial data, cross-correlations, and traditional measures of gait had 78.9% accuracy, 84.4% precision, and 0.846 AUC. Mediolateral signal-based measures-coefficient of variance, cross-correlation with anteroposterior accelerations, and mean acceleration-ranked as the top 3 features. The classification accuracy is promising, given research on probabilistic models of falls indicates accuracies ≥80% are challenging to achieve. The results suggest accelerometer-based measures captured during walking are potentially useful in screening older women for fall risk. We are applying algorithms developed in this paper on an OPACH dataset of 5000 women with a 1-year prospective falls log and week-long, free-living accelerometer data.

Predicting Pulmonary Function from Phone Sensors.

INTRODUCTION: Smartphones are ubiquitous, but it is unknown what physiological functions can be monitored at clinical quality. Pulmonary function is a standard measure of health status for cardiopulmonary patients. We have shown phone sensors can accurately measure walking patterns. Here we show that improved classification models can accurately predict pulmonary function, with sole inputs being motion sensors from carried phones. SUBJECTS AND METHODS: Twenty-five cardiopulmonary patients performed 6-minute walk tests in pulmonary rehabilitation at a regional hospital. They carried smartphones running custom software recording phone motion. Each patient's pulmonary function was measured by spirometry. A universal model, based on support vector machine, then computed the category of function with input from signal processing features and patient demographic features. RESULTS: All but a few of every 10-second interval for every patient was correctly predicted. The trained model perfectly computed the GOLD (Global Initiative for Chronic Obstructive Lung Disease) level 1/2/3, which is a standard classification of pulmonary function. Each level was determined to have a characteristic motion, which could be recognized from the sensor features. In addition, longitudinal changes were detected for 10 patients with multiple walk tests, except for cases with clinical instability. CONCLUSIONS: These results are encouraging toward clinical validation of passive monitors running continuously in the background, for patients in homes during daily activities. Initial testing indicates the same high accuracy as with active monitors, for patients in hospitals during walk tests. We expect patients can simply carry their phones during everyday living, while models support automatic prediction of pulmonary function for health monitoring.

Mining Discriminative Patterns to Predict Health Status for Cardiopulmonary Patients.

Smartphones are ubiquitous now, but it is still unclear what physiological functions they can monitor at clinical quality. Pulmonary function is a standard measure of health status for cardiopulmonary patients. We have shown that predictive models can accurately classify cardiopulmonary conditions from healthy status, as well as different severity levels within cardiopulmonary disease, the GOLD stages. Here we propose several universal models to monitor cardiopulmonary conditions, including DPClass, a novel learning approach we designed. We carefully prepare motion dataset covering status from GOLD 0 (healthy), GOLD 1 (mild), GOLD 2 (moderate), all the way to GOLD 3 (severe). Sixty-six subjects participate in this study. After de-identification, their walking data are applied to train the predictive models. The RBF-SVM model yields the highest accuracy while the DPClass model provides better interpretation of the model mechanisms. We not only provide promising solutions to monitor health status by simply carrying a smartphone, but also demonstrate how demographics influences predictive models of cardiopulmonary disease.

Predicting Transitions in Oxygen Saturation Using Phone Sensors.

INTRODUCTION: Widespread availability of mobile devices is revolutionizing health monitoring. Smartphones are ubiquitous, but it is unknown what vital signs can be monitored with medical quality. Oxygen saturation is a standard measure of health status. We have shown phone sensors can accurately measure walking patterns. SUBJECTS AND METHODS: Twenty cardiopulmonary patients performed 6-min walk tests in pulmonary rehabilitation at a regional hospital. They wore pulse oximeters and carried smartphones running our MoveSense software, which continuously recorded saturation and motion. Continuous saturation defined categories corresponding to status levels, including transitions. Continuous motion was used to compute spatiotemporal gait parameters from sensor data. Our existing gait model was then trained with these data and used to predict transitions in oxygen saturation. For walking variation, 10-s windows are units for classifying into status categories. RESULTS: Oxygen saturation clustered into three categories, corresponding to pulmonary function Global Initiative for Chronic Obstructive Lung Disease (GOLD) 1 and GOLD 2, with a Transition category where saturation varied around the mean rather than remaining steady with low standard deviation. This category indicates patients who are not clinically stable. The gait model predicted status during each measured window of free walking, with 100% accuracy for the 20 subjects, based on majority voting. CONCLUSIONS: Continuous recording of oxygen saturation can predict cardiopulmonary status, including patients in transition between status levels. Gait models using phone sensors can accurately predict these saturation categories from walking motion. This suggests medical devices for predicting clinical stability from passive monitoring using carried smartphones.

Classification Models for Pulmonary Function using Motion Analysis from Phone Sensors.

Smartphones are ubiquitous, but it is unknown what physiological functions can be monitored at clinical quality. Pulmonary function is a standard measure of health status for cardiopulmonary patients. We have shown phone sensors can accurately measure walking patterns. Here we show that improved classification models can accurately measure pulmonary function, with sole inputs being sensor data from carried phones. Twenty-four cardiopulmonary patients performed six minute walk tests in pulmonary rehabilitation at a regional hospital. They carried smartphones running custom software recording phone motion. For every patient, every ten-second interval was correctly computed. The trained model perfectly computed the GOLD level 1/2/3, which is a standard categorization of pulmonary function as measured by spirometry. These results are encouraging towards field trials with passive monitors always running in the background. We expect patients can simply carry their phones during daily living, while supporting automatic computation ofpulmonary function for health monitoring.

A natural walking monitor for pulmonary patients using mobile phones.

Mobile devices have the potential to continuously monitor health by collecting movement data including walking speed during natural walking. Natural walking is walking without artificial speed constraints present in both treadmill and nurse-assisted walking. Fitness trackers have become popular which record steps taken and distance, typically using a fixed stride length. While useful for everyday purposes, medical monitoring requires precise accuracy and testing on real patients with a scientifically valid measure. Walking speed is closely linked to morbidity in patients and widely used for medical assessment via measured walking. The 6-min walk test (6MWT) is a standard assessment for chronic obstructive pulmonary disease and congestive heart failure. Current generation smartphone hardware contains similar sensor chips as in medical devices and popular fitness devices. We developed a middleware software, MoveSense, which runs on standalone smartphones while providing comparable readings to medical accelerometers. We evaluate six machine learning methods to obtain gait speed during natural walking training models to predict natural walking speed and distance during a 6MWT with 28 pulmonary patients and ten subjects without pulmonary condition. We also compare our model's accuracy to popular fitness devices. Our universally trained support vector machine models produce 6MWT distance with 3.23% error during a controlled 6MWT and 11.2% during natural free walking. Furthermore, our model attains 7.9% error when tested on five subjects for distance estimation compared to the 50-400% error seen in fitness devices during natural walking.

National Surveys of Population Health: Big Data Analytics for Mobile Health Monitors.

At the core of the healthcare crisis is fundamental lack of actionable data. Such data could stratify individuals within populations to predict which persons have which outcomes. If baselines existed for all variations of all conditions, then managing health could be improved by matching the measuring of individuals to their cohort in the population. The scale required for complete baselines involves effective National Surveys of Population Health (NSPH). Traditionally, these have been focused upon acute medicine, measuring people to contain the spread of epidemics. In recent decades, the focus has moved to chronic conditions as well, which require smaller measures over longer times. NSPH have long utilized quality of life questionnaires. Mobile Health Monitors, where computing technologies eliminate manual administration, provide richer data sets for health measurement. Older technologies of telephone interviews will be replaced by newer technologies of smartphone sensors to provide deeper individual measures at more frequent timings across larger-sized populations. Such continuous data can provide personal health records, supporting treatment guidelines specialized for population cohorts. Evidence-based medicine will become feasible by leveraging hundreds of millions of persons carrying mobile devices interacting with Internet-scale services for Big Data Analytics.

Health monitors for chronic disease by gait analysis with mobile phones.

We have developed GaitTrack, a phone application to detect health status while the smartphone is carried normally. GaitTrack software monitors walking patterns, using only accelerometers embedded in phones to record spatiotemporal motion, without the need for sensors external to the phone. Our software transforms smartphones into health monitors, using eight parameters of phone motion transformed into body motion by the gait model. GaitTrack is designed to detect health status while the smartphone is carried during normal activities, namely, free-living walking. The current method for assessing free-living walking is medical accelerometers, so we present evidence that mobile phones running our software are more accurate. We then show our gait model is more accurate than medical pedometers for counting steps of patients with chronic disease. Our gait model was evaluated in a pilot study involving 30 patients with chronic lung disease. The six-minute walk test (6 MWT) is a major assessment for chronic heart and lung disease, including congestive heart failure and especially chronic obstructive pulmonary disease (COPD), affecting millions of persons. The 6 MWT consists of walking back and forth along a measured distance for 6 minutes. The gait model using linear regression performed with 94.13% accuracy in measuring walk distance, compared with the established standard of direct observation. We also evaluated a different statistical model using the same gait parameters to predict health status through lung function. This gait model has high accuracy when applied to demographic cohorts, for example, 89.22% accuracy testing the cohort of 12 female patients with ages 50-64 years.

Interventions to reduce rehospitalizations after chronic obstructive pulmonary disease exacerbations. A systematic review.

RATIONALE: Approximately 20% of patients hospitalized for COPD exacerbations in the United States will be readmitted within 30 days. The Centers for Medicare and Medicaid Services has recently proposed to revise the Hospital Readmissions Reduction Program to financially penalize hospitals with high all-cause 30-day rehospitalization rates after a hospitalization for COPD exacerbation on or after October 1, 2014. OBJECTIVES: To report the results of a systematic review of randomized clinical trials evaluating interventions to reduce the rehospitalizations after COPD exacerbations. METHODS: Multiple electronic databases were systematically searched to identify relevant studies published between January 1966 and June 2013. Titles, abstracts, and, subsequently, full-text articles were assessed for eligibility. Each study was appraised using predefined criteria. MEASUREMENTS AND MAIN RESULTS: Among 913 titles and abstracts screened, 5 studies (1,393 participants) met eligibility criteria. All studies had a primary outcome of rehospitalization at 6 or 12 months. No study examined 30-day rehospitalization as the primary outcome. Each study tested a different set of interventions. Two studies (one conducted in Canada and one conducted in Spain and Belgium) showed a decrease in all-cause rehospitalization over 12 months in the intervention group versus comparator group (mean number of hospitalizations per patient, 1.0 vs. 1.8; P = 0.01; percent hospitalized, 45 vs. 67%; P = 0.028; respectively). The only study conducted in the United States found a greater than twofold higher risk of mortality in the intervention group (17 vs. 7%, P = 0.003) but no significant difference in rehospitalizations. It was unclear which set of interventions was effective or harmful. CONCLUSIONS: The evidence base is inadequate to recommend specific interventions to reduce rehospitalizations in this population and does not justify penalizing hospitals for high 30-day rehospitalization rates after COPD exacerbations.

Designing and evaluating a clustering system for organizing and integrating patient drug outcomes in personal health messages.

Patient outcomes to drugs vary, but physicians currently have little data about individual responses. We designed a comprehensive system to organize and integrate patient outcomes utilizing semantic analysis, which groups large collections of personal comments into a series of topics. A prototype implementation was built to extract situational evidences by filtering and digesting user comments provided by patients. Our methods do not require extensive training or dictionaries, while categorizing comments based on expert opinions from standard source, or patient-specified categories. This system has been tested with sample health messages from our unique dataset from Yahoo! Groups, containing 12M personal messages from 27K public groups in Health and Wellness. We have performed an extensive evaluation of the clustering results with medical students. Evaluated results show high quality of labeled clustering, promising an effective automatic system for discovering patient outcomes from large volumes of health information.

Predicting adverse drug events from personal health messages.

Adverse drug events (ADEs) remain a large problem in the United States, being the fourth leading cause of death, despite post market drug surveillance. Much post consumer drug surveillance relies on self-reported "spontaneous" patient data. Previous work has performed datamining over the FDA's Adverse Event Reporting System (AERS) and other spontaneous reporting systems to identify drug interactions and drugs correlated with high rates of serious adverse events. However, safety problems have resulted from the lack of post marketing surveillance information about drugs, with underreporting rates of up to 98% within such systems. We explore the use of online health forums as a source of data to identify drugs for further FDA scrutiny. In this work we aggregate individuals' opinions and review of drugs similar to crowd intelligence3. We use natural language processing to group drugs discussed in similar ways and are able to successfully identify drugs withdrawn from the market based on messages discussing them before their removal.

Behavior-specific changes in transcriptional modules lead to distinct and predictable neurogenomic states.

Using brain transcriptomic profiles from 853 individual honey bees exhibiting 48 distinct behavioral phenotypes in naturalistic contexts, we report that behavior-specific neurogenomic states can be inferred from the coordinated action of transcription factors (TFs) and their predicted target genes. Unsupervised hierarchical clustering of these transcriptomic profiles showed three clusters that correspond to three ecologically important behavioral categories: aggression, maturation, and foraging. To explore the genetic influences potentially regulating these behavior-specific neurogenomic states, we reconstructed a brain transcriptional regulatory network (TRN) model. This brain TRN quantitatively predicts with high accuracy gene expression changes of more than 2,000 genes involved in behavior, even for behavioral phenotypes on which it was not trained, suggesting that there is a core set of TFs that regulates behavior-specific gene expression in the bee brain, and other TFs more specific to particular categories. TFs playing key roles in the TRN include well-known regulators of neural and behavioral plasticity, e.g., Creb, as well as TFs better known in other biological contexts, e.g., NF-κB (immunity). Our results reveal three insights concerning the relationship between genes and behavior. First, distinct behaviors are subserved by distinct neurogenomic states in the brain. Second, the neurogenomic states underlying different behaviors rely upon both shared and distinct transcriptional modules. Third, despite the complexity of the brain, simple linear relationships between TFs and their putative target genes are a surprisingly prominent feature of the networks underlying behavior.

BeeSpace Navigator: exploratory analysis of gene function using semantic indexing of biological literature.

With the rapid decrease in cost of genome sequencing, the classification of gene function is becoming a primary problem. Such classification has been performed by human curators who read biological literature to extract evidence. BeeSpace Navigator is a prototype software for exploratory analysis of gene function using biological literature. The software supports an automatic analogue of the curator process to extract functions, with a simple interface intended for all biologists. Since extraction is done on selected collections that are semantically indexed into conceptual spaces, the curation can be task specific. Biological literature containing references to gene lists from expression experiments can be analyzed to extract concepts that are computational equivalents of a classification such as Gene Ontology, yielding discriminating concepts that differentiate gene mentions from other mentions. The functions of individual genes can be summarized from sentences in biological literature, to produce results resembling a model organism database entry that is automatically computed. Statistical frequency analysis based on literature phrase extraction generates offline semantic indexes to support these gene function services. The website with BeeSpace Navigator is free and open to all; there is no login requirement at www.beespace.illinois.edu for version 4. Materials from the 2010 BeeSpace Software Training Workshop are available at www.beespace.illinois.edu/bstwmaterials.php.

BSQA: integrated text mining using entity relation semantics extracted from biological literature of insects.

Text mining is one promising way of extracting information automatically from the vast biological literature. To maximize its potential, the knowledge encoded in the text should be translated to some semantic representation such as entities and relations, which could be analyzed by machines. But large-scale practical systems for this purpose are rare. We present BeeSpace question/answering (BSQA) system that performs integrated text mining for insect biology, covering diverse aspects from molecular interactions of genes to insect behavior. BSQA recognizes a number of entities and relations in Medline documents about the model insect, Drosophila melanogaster. For any text query, BSQA exploits entity annotation of retrieved documents to identify important concepts in different categories. By utilizing the extracted relations, BSQA is also able to answer many biologically motivated questions, from simple ones such as, which anatomical part is a gene expressed in, to more complex ones involving multiple types of relations. BSQA is freely available at http://www.beespace.uiuc.edu/QuestionAnswer.

Identifying overrepresented concepts in gene lists from literature: a statistical approach based on Poisson mixture model.

BACKGROUND: Large-scale genomic studies often identify large gene lists, for example, the genes sharing the same expression patterns. The interpretation of these gene lists is generally achieved by extracting concepts overrepresented in the gene lists. This analysis often depends on manual annotation of genes based on controlled vocabularies, in particular, Gene Ontology (GO). However, the annotation of genes is a labor-intensive process; and the vocabularies are generally incomplete, leaving some important biological domains inadequately covered. RESULTS: We propose a statistical method that uses the primary literature, i.e. free-text, as the source to perform overrepresentation analysis. The method is based on a statistical framework of mixture model and addresses the methodological flaws in several existing programs. We implemented this method within a literature mining system, BeeSpace, taking advantage of its analysis environment and added features that facilitate the interactive analysis of gene sets. Through experimentation with several datasets, we showed that our program can effectively summarize the important conceptual themes of large gene sets, even when traditional GO-based analysis does not yield informative results. CONCLUSIONS: We conclude that the current work will provide biologists with a tool that effectively complements the existing ones for overrepresentation analysis from genomic experiments. Our program, Genelist Analyzer, is freely available at: http://workerbee.igb.uiuc.edu:8080/BeeSpace/Search.jsp.

Feasibility of mobile phone-based management of chronic illness.

According to the CDC, chronic conditions such as heart disease, cancer, and diabetes cause 75% of healthcare spending in the United States and contribute to nearly seven in ten American deaths. However, despite the prevalence and high-cost of chronic disease, they are also among the most preventable of health problems1. How can we use technology to improve self-care, reduce costs, and lessen the burden on medical professionals? Devices to help manage chronic illness have been marketed for years, but are these specialized devices really necessary? In this paper, the authors identify the aspects of the major chronic illnesses that most need to be controlled and monitored in the US today and explore the feasibility of using current mobile phone technology to improve the management of chronic illness. Here we show that even the average mobile phone is capable of improving the management of all relevant health features in some way.