Among individuals diagnosed, hospitalized, and treated for schizophrenia, up to 40% of those discharged may relapse within 1 year even with appropriate treatment. Passively collected smartphone behavioral data present a scalable and at present underutilized opportunity to monitor patients in order to identify possible warning signs of relapse. Seventeen patients with schizophrenia in active treatment at a state mental health clinic in Boston used the Beiwe app on their personal smartphone for up to 3 months. By testing for changes in mobility patterns and social behavior over time as measured through smartphone use, we were able to identify statistically significant anomalies in patient behavior in the days prior to relapse. We found that the rate of behavioral anomalies detected in the 2 weeks prior to relapse was 71% higher than the rate of anomalies during other time periods. Our findings show how passive smartphone data, data collected in the background during regular phone use without active input from the subjects, can provide an unprecedented and detailed view into patient behavior outside the clinic. Real-time detection of behavioral anomalies could signal the need for an intervention before an escalation of symptoms and relapse occur, therefore reducing patient suffering and reducing the cost of care.

In the increasing number of sequencing studies aimed at identifying rare variants associated with complex traits, the power of the test can be improved by guided sampling procedures. We confirm both analytically and numerically that sampling individuals with extreme phenotypes can enrich the presence of causal rare variants and can therefore lead to an increase in power compared to random sampling. While application of traditional rare variant association tests to these extreme phenotype samples requires dichotomizing the continuous phenotypes before analysis, the dichotomization procedure can decrease the power by reducing the information in the phenotypes. To avoid this, we propose a novel statistical method based on optimal SKAT (SKAT-O) that allows us to test for rare variant effects using continuous phenotypes in the analysis of extreme phenotype samples. The increase in power of this method is demonstrated through simulation of a wide range of scenarios as well as in the triglyceride data of the Dallas Heart Study.

Advances in smartphone sensing, machine learning methods, and mobile apps directed towards reducing suicide offer promising evidence; however, most of these innovative approaches are still nascent. Further replication and validation of preliminary results is needed. Whereas numerous promising mobile and sensor technology based solutions for real time understanding, predicting, and caring for those at highest risk of suicide are being studied today, their clinical utility remains largely unproven. However, given both the rapid pace and vast scale of current research efforts, we expect clinicians will soon see useful and impactful digital tools for this space within the next 2 to 5 years.

It is of substantial interest to study the effects of genes, genetic pathways, and networks on the risk of complex diseases. These genetic constructs each contain multiple SNPs, which are often correlated and function jointly, and might be large in number. However, only a sparse subset of SNPs in a genetic construct is generally associated with the disease of interest. In this article, we propose the generalized higher criticism (GHC) to test for the association between an SNP set and a disease outcome. The higher criticism is a test traditionally used in high-dimensional signal detection settings when marginal test statistics are independent and the number of parameters is very large. However, these assumptions do not always hold in genetic association studies, due to linkage disequilibrium among SNPs and the finite number of SNPs in an SNP set in each genetic construct. The proposed GHC overcomes the limitations of the higher criticism by allowing for arbitrary correlation structures among the SNPs in an SNP-set, while performing accurate analytic p-value calculations for any finite number of SNPs in the SNP-set. We obtain the detection boundary of the GHC test. We compared empirically using simulations the power of the GHC method with existing SNP-set tests over a range of genetic regions with varied correlation structures and signal sparsity. We apply the proposed methods to analyze the CGEM breast cancer genome-wide association study. Supplementary materials for this article are available online.

Digital phenotyping, or the moment-by-moment quantification of the individual-level human phenotype in situ using data from personal digital devices and smartphones, in particular, holds great potential for behavioral monitoring of patients. However, realizing the potential of digital phenotyping requires understanding of the smartphone as a scientific data collection tool. In this pilot study, we detail a procedure for estimating data quality for phone sensor samples and model the relationship between data quality and future symptom-related survey responses in a cohort with schizophrenia. We find that measures of empirical coverage of collected accelerometer and GPS data, as well as survey timing and survey completion metrics, are significantly associated with future survey scores for a variety of symptom domains. We also find evidence that specific measures of data quality are indicative of domain-specific future survey outcomes. These results suggest that for smartphone-based digital phenotyping, metadata is not independent of patient-reported survey scores, and is therefore potentially useful in predicting future clinical outcomes. This work raises important questions and considerations for future studies; we explore and discuss some of these implications.npj Digital Medicine (2018) 1:15 ;

Many systems of interacting elements can be conceptualized as networks, where network nodes represent the elements and network ties represent interactions between the elements. In systems where the underlying network evolves, it is useful to determine the points in time where the network structure changes significantly as these may correspond to functional change points. We propose a method for detecting change points in correlation networks that, unlike previous change point detection methods designed for time series data, requires minimal distributional assumptions. We investigate the difficulty of change point detection near the boundaries of the time series in correlation networks and study the power of our method and competing methods through simulation. We also show the generalizable nature of the method by applying it to stock price data as well as fMRI data.

Sleep abnormalities are considered an important feature of schizophrenia, yet convenient and reliable sleep monitoring remains a challenge. Smartphones offer a novel solution to capture both self-reported and objective measures of sleep in schizophrenia. In this three-month observational study, 17 subjects with a diagnosis of schizophrenia currently in treatment downloaded Beiwe, a platform for digital phenotyping, on their personal Apple or Android smartphones. Subjects were given tri-weekly ecological momentary assessments (EMAs) on their own smartphones, and passive data including accelerometer, GPS, screen use, and anonymized call and text message logs was continuously collected. We compare the in-clinic assessment of sleep quality, assessed with the Pittsburgh Sleep Questionnaire Inventory (PSQI), to EMAs, as well as sleep estimates based on passively collected accelerometer data. EMAs and passive data classified 85% (11/13) of subjects as exhibiting high or low sleep quality compared to the in-clinic assessments among subjects who completed at least one in-person PSQI. Phone-based accelerometer data used to infer sleep duration was moderately correlated with subject self-assessment of sleep duration (r = 0.69, 95% CI 0.23–0.90). Active and passive phone data predicts concurrent PSQI scores for all subjects with mean average error of 0.75 and future PSQI scores with a mean average error of 1.9, with scores ranging from 0–14. These results suggest sleep monitoring via personal smartphones is feasible for subjects with schizophrenia in a scalable and affordable manner.

Network representations of systems from various scientific and societal domains are neither completely random nor fully regular, but instead appear to contain recurring structural building blocks [1]. These features tend to be shared by networks belonging to the same broad class, such as the class of social networks or the class of biological networks. At a finer scale of classification within each such class, networks describing more similar systems tend to have more similar features. This occurs presumably because networks representing similar purposes or constructions would be expected to be generated by a shared set of domain specific mechanisms, and it should therefore be possible to classify these networks into categories based on their features at various structural levels. Here we describe and demonstrate a new, hybrid approach that combines manual selection of features of potential interest with existing automated classification methods. In particular, selecting well-known and well-studied features that have been used throughout social network analysis and network science [2, 3] and then classifying with methods such as random forests [4] that are of special utility in the presence of feature collinearity, we find that we achieve higher accuracy, in shorter computation time, with greater interpretability of the network classification results.Past work in the area of network classification has primarily focused on distinguishing networks from different categories using two different broad classes of approaches. In the first approach, network classification is carried out by examining certain specific structural features and investigating whether networks belonging to the same category are similar across one or more dimensions as defined by these features [5,6,7,8]. In other words, in this approach the investigator manually chooses the structural characteristics of interest and more or less manually (informally) determines the regions of the feature space that correspond to different classes. These methods are scalable to large networks and yield results that are easily interpreted in terms of the characteristics of interest, but in practice they tend to lead to suboptimal classification accuracy. In the second approach, network classification is done by using very flexible machine learning classifiers that, when presented with a network as an input, classify its category or class as an output [9,10,11,12,13,14,15,16,17,18]. To somewhat oversimplify, the first approach relies on manual feature specification followed by manual selection of a classification system, whereas the second approach is its opposite, relying on automated feature detection followed by automated classification. While the latter approach can yield very accurate class predictions, its computational 2 cost typically scales poorly and, perhaps more importantly, the potentially opaque nature of the methodology may make it difficult to interpret the obtained results. This paper presents a third, hybrid approach to the network classification...