Flexible Bayesian inference on partially observed epidemics

Individual-based models of contagious processes are useful for predicting epidemic trajectories and informing intervention strategies. In such models, the incorporation of contact network information can capture the non-randomness and heterogeneity of realistic contact dynamics. In this article, we consider Bayesian inference on the spreading parameters of an SIR contagion on a known, static network, where information regarding individual disease status is known only from a series of tests (positive or negative disease status). When the contagion model is complex or information such as infection and removal times is missing, the posterior distribution can be difficult to sample from. Previous work has considered the use of Approximate Bayesian Computation (ABC), which allows for simulation-based Bayesian inference on complex models. However, ABC methods usually require the user to select reasonable summary statistics. Here, we consider an inference scheme based on the Mixture Density Network compressed ABC, which minimizes the expected posterior entropy in order to learn informative summary statistics. This allows us to conduct Bayesian inference on the parameters of a partially observed contagious process while also circumventing the need for manual summary statistic selection. This methodology can be extended to incorporate additional simulation complexities, including behavioural change after positive tests or false test results.

Upper limb movements as digital biomarkers in people with ALS

BACKGROUND

Objective evaluation of people with amyotrophic lateral sclerosis (PALS) in free-living settings is challenging. The introduction of portable digital devices, such as wearables and smartphones, may improve quantifying disease progression and hasten therapeutic development. However, there is a need for tools to characterize upper limb movements in neurologic disease and disability.

METHODS

Twenty PALS wore a wearable accelerometer, ActiGraph Insight Watch, on their wrist for six months. They also used Beiwe, a smartphone application that collected self-entry ALS Functional Rating Scale-Revised (ALSFRS-RSE) survey responses every 1-4 weeks. We developed several measures that quantify count and duration of upper limb movements: flexion, extension, supination, and pronation. New measures were compared against ALSFRS-RSE total score (Q1-12), and individual responses to specific questions related to handwriting (Q4), cutting food (Q5), dressing and performing hygiene (Q6), and turning in bed and adjusting bed clothes (Q7). Additional analysis considered adjusting for total activity counts (TAC).

FINDINGS

At baseline, PALS with higher Q1-12 performed more upper limb movements, and these movements were faster compared to individuals with more advanced disease. Most upper limb movement metrics had statistically significant change over time, indicating declining function either by decreasing count metrics or by increasing duration metric. All count and duration metrics were significantly associated with Q1-12, flexion and extension counts were significantly associated with Q6 and Q7, supination and pronation counts were also associated with Q4. All duration metrics were associated with Q6 and Q7. All duration metrics retained their statistical significance after adjusting for TAC.

INTERPRETATION

Wearable accelerometer data can be used to generate digital biomarkers on upper limb movements and facilitate patient monitoring in free-living environments. The presented method offers interpretable monitoring of patients' functioning and versatile tracking of disease progression in the limb of interest.

FUNDING

Mitsubishi-Tanabe Pharma Holdings America, Inc.

The Pain Intervention & Digital Research Program: an operational report on combining digital research with outpatient chronic disease management

Chronic pain affects up to 28% of U.S. adults, costing ∼$560 billion each year. Chronic pain is an instantiation of the perennial complexity of how to best assess and treat chronic diseases over time, especially in populations where age, medical comorbidities, and socioeconomic barriers may limit access to care. Chronic disease management poses a particular challenge for the healthcare system's transition from fee-for-service to value and risk-based reimbursement models. Remote, passive real-time data from smartphones could enable more timely interventions and simultaneously manage risk and promote better patient outcomes through predicting and preventing costly adverse outcomes; however, there is limited evidence whether remote monitoring is feasible, especially in the case of older patients with chronic pain. Here, we introduce the Pain Intervention and Digital Research (Pain-IDR) Program as a pilot initiative launched in 2022 that combines outpatient clinical care and digital health research. The Pain-IDR seeks to test whether functional status can be assessed passively, through a smartphone application, in older patients with chronic pain. We discuss two perspectives-a narrative approach that describes the clinical settings and rationale behind changes to the operational design, and a quantitative approach that measures patient recruitment, patient experience, and HERMES data characteristics. Since launch, we have had 77 participants with a mean age of 55.52, of which n = 38 have fully completed the 6 months of data collection necessitated to be considered in the study, with an active data collection rate of 51% and passive data rate of 78%. We further present preliminary operational strategies that we have adopted as we have learned to adapt the Pain-IDR to a productive clinical service. Overall, the Pain-IDR has successfully engaged older patients with chronic pain and presents useful insights for others seeking to implement digital phenotyping in other chronic disease settings.

Open-Source, Step-Counting Algorithm for Smartphone Data Collected in Clinical and Nonclinical Settings: Algorithm Development and Validation Study

BACKGROUND

Step counts are increasingly used in public health and clinical research to assess well-being, lifestyle, and health status. However, estimating step counts using commercial activity trackers has several limitations, including a lack of reproducibility, generalizability, and scalability. Smartphones are a potentially promising alternative, but their step-counting algorithms require robust validation that accounts for temporal sensor body location, individual gait characteristics, and heterogeneous health states.

OBJECTIVE

Our goal was to evaluate an open-source, step-counting method for smartphones under various measurement conditions against step counts estimated from data collected simultaneously from different body locations ("cross-body" validation), manually ascertained ground truth ("visually assessed" validation), and step counts from a commercial activity tracker (Fitbit Charge 2) in patients with advanced cancer ("commercial wearable" validation).

METHODS

We used 8 independent data sets collected in controlled, semicontrolled, and free-living environments with different devices (primarily Android smartphones and wearable accelerometers) carried at typical body locations. A total of 5 data sets (n=103) were used for cross-body validation, 2 data sets (n=107) for visually assessed validation, and 1 data set (n=45) was used for commercial wearable validation. In each scenario, step counts were estimated using a previously published step-counting method for smartphones that uses raw subsecond-level accelerometer data. We calculated the mean bias and limits of agreement (LoA) between step count estimates and validation criteria using Bland-Altman analysis.

RESULTS

In the cross-body validation data sets, participants performed 751.7 (SD 581.2) steps, and the mean bias was -7.2 (LoA -47.6, 33.3) steps, or -0.5%. In the visually assessed validation data sets, the ground truth step count was 367.4 (SD 359.4) steps, while the mean bias was -0.4 (LoA -75.2, 74.3) steps, or 0.1%. In the commercial wearable validation data set, Fitbit devices indicated mean step counts of 1931.2 (SD 2338.4), while the calculated bias was equal to -67.1 (LoA -603.8, 469.7) steps, or a difference of 3.4%.

CONCLUSIONS

This study demonstrates that our open-source, step-counting method for smartphone data provides reliable step counts across sensor locations, measurement scenarios, and populations, including healthy adults and patients with cancer.

Flexible Bayesian Inference on Partially Observed Epidemics

Individual-based models of contagious processes are useful for predicting epidemic trajectories and informing intervention strategies. In such models, the incorporation of contact network information can capture the non-randomness and heterogeneity of realistic contact dynamics. In this paper, we consider Bayesian inference on the spreading parameters of an SIR contagion on a known, static network, where information regarding individual disease status is known only from a series of tests (positive or negative disease status). When the contagion model is complex or information such as infection and removal times is missing, the posterior distribution can be difficult to sample from. Previous work has considered the use of Approximate Bayesian Computation (ABC), which allows for simulation-based Bayesian inference on complex models. However, ABC methods usually require the user to select reasonable summary statistics. Here, we consider an inference scheme based on the Mixture Density Network compressed ABC (MDN-ABC), which minimizes the expected posterior entropy in order to learn informative summary statistics. This allows us to conduct Bayesian inference on the parameters of a partially observed contagious process while also circumventing the need for manual summary statistic selection. This methodology can be extended to incorporate additional simulation complexities, including behavioral change after positive tests or false test results.

Framework for converting mechanistic network models to probabilistic models

There are two prominent paradigms for the modelling of networks: in the first, referred to as the mechanistic approach, one specifies a set of domain-specific mechanistic rules that are used to grow or evolve the network over time; in the second, referred to as the probabilistic approach, one describes a model that specifies the likelihood of observing a given network. Mechanistic models (models developed based on the mechanistic approach) are appealing because they capture scientific processes that are believed to be responsible for network generation; however, they do not easily lend themselves to the use of inferential techniques when compared with probabilistic models. We introduce a general framework for converting a mechanistic network model (MNM) to a probabilistic network model (PNM). The proposed framework makes it possible to identify the essential network properties and their joint probability distribution for some MNMs; doing so makes it possible to address questions such as whether two different mechanistic models generate networks with identical distributions of properties, or whether a network property, such as clustering, is over- or under-represented in the networks generated by the model of interest compared with a reference model. The proposed framework is intended to bridge some of the gap that currently exists between the formulation and representation of mechanistic and PNMs. We also highlight limitations of PNMs that need to be addressed in order to close this gap.

Insurance-Based Disparities in Stroke Center Access in California: A Network Science Approach

BACKGROUND

Our objectives were to determine whether there is an association between ischemic stroke patient insurance and likelihood of transfer overall and to a stroke center and whether hospital cluster modified the association between insurance and likelihood of stroke center transfer.

METHODS

This retrospective network analysis of California data included every nonfederal hospital ischemic stroke admission from 2010 to 2017. Transfers from an emergency department to another hospital were categorized based on whether the patient was discharged from a stroke center (primary or comprehensive). We used logistic regression models to examine the relationship between insurance (private, Medicare, Medicaid, uninsured) and odds of (1) any transfer among patients initially presenting to nonstroke center hospital emergency departments and (2) transfer to a stroke center among transferred patients. We used a network clustering method to identify clusters of hospitals closely connected through transfers. Within each cluster, we quantified the difference between insurance groups with the highest and lowest proportion of transfers discharged from a stroke center.

RESULTS

Of 332 995 total ischemic stroke encounters, 51% were female, 70% were ≥65 years, and 3.5% were transferred from the initial emergency department. Of 52 316 presenting to a nonstroke center, 3466 (7.1%) were transferred. Relative to privately insured patients, there were lower odds of transfer and of transfer to a stroke center among all groups (Medicare odds ratio, 0.24 [95% CI, 0.22-0.26] and 0.59 [95% CI, 0.50-0.71], Medicaid odds ratio, 0.26 [95% CI, 0.23-0.29] and odds ratio, 0.49 [95% CI, 0.38-0.62], uninsured odds ratio, 0.75 [95% CI, 0.63-0.89], and 0.72 [95% CI, 0.6-0.8], respectively). Among the 14 identified hospital clusters, insurance-based disparities in transfer varied and the lowest performing cluster (also the largest; n=2364 transfers) fully explained the insurance-based disparity in odds of stroke center transfer.

CONCLUSIONS

Uninsured patients had less stroke center access through transfer than patients with insurance. This difference was largely explained by patterns in 1 particular hospital cluster.

Daily Pain Prediction Using Smartphone Speech Recordings of Patients With Spine Disease

BACKGROUND

Pain evaluation remains largely subjective in neurosurgical practice, but machine learning provides the potential for objective pain assessment tools.

OBJECTIVE

To predict daily pain levels using speech recordings from personal smartphones of a cohort of patients with diagnosed neurological spine disease.

METHODS

Patients with spine disease were enrolled through a general neurosurgical clinic with approval from the institutional ethics committee. At-home pain surveys and speech recordings were administered at regular intervals through the Beiwe smartphone application. Praat audio features were extracted from the speech recordings to be used as input to a K-nearest neighbors (KNN) machine learning model. The pain scores were transformed from a 0 to 10 scale to low and high pain for better discriminative capacity.

RESULTS

A total of 60 patients were enrolled, and 384 observations were used to train and test the prediction model. Using the KNN prediction model, an accuracy of 71% with a positive predictive value of 0.71 was achieved in classifying pain intensity into high and low. The model showed 0.71 precision for high pain and 0.70 precision for low pain. Recall of high pain was 0.74, and recall of low pain was 0.67. The overall F1 score was 0.73.

CONCLUSION

Our study uses a KNN to model the relationship between speech features and pain levels collected from personal smartphones of patients with spine disease. The proposed model is a stepping stone for the development of objective pain assessment in neurosurgery clinical practice.

Maximum likelihood estimation for reversible mechanistic network models

Mechanistic network models specify the mechanisms by which networks grow and change, allowing researchers to investigate complex systems using both simulation and analytical techniques. Unfortunately, it is difficult to write likelihoods for instances of graphs generated with mechanistic models, and thus it is near impossible to estimate the parameters using maximum likelihood estimation. In this paper, we propose treating the node sequence in a growing network model as an additional parameter, or as a missing random variable, and maximizing over the resulting likelihood. We develop this framework in the context of a simple mechanistic network model, used to study gene duplication and divergence, and test a variety of algorithms for maximizing the likelihood in simulated graphs. We also run the best-performing algorithm on one human protein-protein interaction network and four nonhuman protein-protein interaction networks. Although we focus on a specific mechanistic network model, the proposed framework is more generally applicable to reversible models.

Menstrual cycle length variation by demographic characteristics from the Apple Women's Health Study

Menstrual characteristics are important signs of women's health. Here we examine the variation of menstrual cycle length by age, ethnicity, and body weight using 165,668 cycles from 12,608 participants in the US using mobile menstrual tracking apps. After adjusting for all covariates, mean menstrual cycle length is shorter with older age across all age groups until age 50 and then became longer for those age 50 and older. Menstrual cycles are on average 1.6 (95%CI: 1.2, 2.0) days longer for Asian and 0.7 (95%CI: 0.4, 1.0) days longer for Hispanic participants compared to white non-Hispanic participants. Participants with BMI ≥ 40 kg/m² have 1.5 (95%CI: 1.2, 1.8) days longer cycles compared to those with BMI between 18.5 and 25 kg/m². Cycle variability is the lowest among participants aged 35-39 but are considerably higher by 46% (95%CI: 43%, 48%) and 45% (95%CI: 41%, 49%) among those aged under 20 and between 45-49. Cycle variability increase by 200% (95%CI: 191%, 210%) among those aged above 50 compared to those in the 35-39 age group. Compared to white participants, those who are Asian and Hispanic have larger cycle variability. Participants with obesity also have higher cycle variability. Here we confirm previous observations of changes in menstrual cycle pattern with age across reproductive life span and report new evidence on the differences of menstrual variation by ethnicity and obesity status. Future studies should explore the underlying determinants of the variation in menstrual characteristics.

Machine Learning and the Digital Measurement of Psychological Health

Since its inception, the discipline of psychology has utilized empirical epistemology and mathematical methodologies to infer psychological functioning from direct observation. As new challenges and technological opportunities emerge, scientists are once again challenged to define measurement paradigms for psychological health and illness that solve novel problems and capitalize on new technological opportunities. In this review, we discuss the theoretical foundations of and scientific advances in remote sensor technology and machine learning models as they are applied to quantify psychological functioning, draw clinical inferences, and chart new directions in treatment.

Validation of an open-source smartphone step counting algorithm in clinical and non-clinical settings

Background

Step counts are increasingly used in public health and clinical research to assess wellbeing, lifestyle, and health status. However, estimating step counts using commercial activity trackers has several limitations, including a lack of reproducibility, generalizability, and scalability. Smartphones are a potentially promising alternative, but their step-counting algorithms require robust validation that accounts for temporal sensor body location, individual gait characteristics, and heterogeneous health states.

Objective

Our goal was to evaluate an open-source step-counting method for smartphones under various measurement conditions against step counts estimated from data collected simultaneously from different body locations ("internal" validation), manually ascertained ground truth ("manual" validation), and step counts from a commercial activity tracker (Fitbit Charge 2) in patients with advanced cancer ("wearable" validation).

Methods

We used eight independent datasets collected in controlled, semi-controlled, and free-living environments with different devices (primarily Android smartphones and wearable accelerometers) carried at typical body locations. Five datasets (N=103) were used for internal validation, two datasets (N=107) for manual validation, and one dataset (N=45) used for wearable validation. In each scenario, step counts were estimated using a previously published step-counting method for smartphones that uses raw sub-second level accelerometer data. We calculated mean bias and limits of agreement (LoA) between step count estimates and validation criteria using Bland-Altman analysis.

Results

In the internal validation datasets, participants performed 751.7±581.2 (mean±SD) steps, and the mean bias was -7.2 steps (LoA -47.6, 33.3) or -0.5%. In the manual validation datasets, the ground truth step count was 367.4±359.4 steps while the mean bias was -0.4 steps (LoA -75.2, 74.3) or 0.1 %. In the wearable validation dataset, Fitbit devices indicated mean step counts of 1931.2±2338.4, while the calculated bias was equal to -67.1 steps (LoA -603.8, 469.7) or a difference of 0.3 %.

Conclusions

This study demonstrates that our open-source step counting method for smartphone data provides reliable step counts across sensor locations, measurement scenarios, and populations, including healthy adults and patients with cancer.

Randomized trial of inosine for urate elevation in amyotrophic lateral sclerosis

INTRODUCTION/AIMS

Higher urate levels are associated with improved ALS survival in retrospective studies, however whether raising urate levels confers a survival advantage is unknown. In the Safety of Urate Elevation in Amyotrophic Lateral Sclerosis (SURE-ALS) trial, inosine raised serum urate and was safe and well-tolerated. The SURE-ALS2 trial was designed to assess longer term safety. Functional outcomes and a smartphone application were also explored.

METHODS

Participants were randomized 2:1 to inosine (n = 14) or placebo (n = 9) for 20 weeks, titrated to serum urate of 7-8 mg/dL. Primary outcomes were safety and tolerability. Functional outcomes were measured with the Amyotrophic Lateral Sclerosis Functional Rating Scale Revised (ALSFRS-R). Mobility and ALSFRS-R were also assessed by a smartphone application.

RESULTS

During inosine treatment, mean urate ranged 5.68-6.82 mg/dL. Treatment-emergent adverse event (TEAE) incidence was similar between groups (p > .10). Renal TEAEs occurred in three (21%) and hypertension in one (7%) of participants randomized to inosine. Inosine was tolerated in 71% of participants versus placebo 67%. Two participants (14%) in the inosine group experienced TEAEs deemed related to treatment (nephrolithiasis); one was a severe adverse event. Mean ALSFRS-R decline did not differ between groups (p = .69). Change in measured home time was similar between groups. Digital and in-clinic ALSFRS-R correlated well.

DISCUSSION

Inosine met pre-specified criteria for safety and tolerability. A functional benefit was not demonstrated in this trial designed for safety and tolerability. Findings suggested potential utility for a smartphone application in ALS clinical and research settings.

A "one-size-fits-most" walking recognition method for smartphones, smartwatches, and wearable accelerometers

The ubiquity of personal digital devices offers unprecedented opportunities to study human behavior. Current state-of-the-art methods quantify physical activity using "activity counts," a measure which overlooks specific types of physical activities. We propose a walking recognition method for sub-second tri-axial accelerometer data, in which activity classification is based on the inherent features of walking: intensity, periodicity, and duration. We validate our method against 20 publicly available, annotated datasets on walking activity data collected at various body locations (thigh, waist, chest, arm, wrist). We demonstrate that our method can estimate walking periods with high sensitivity and specificity: average sensitivity ranged between 0.92 and 0.97 across various body locations, and average specificity for common daily activities was typically above 0.95. We also assess the method's algorithmic fairness to demographic and anthropometric variables and measurement contexts (body location, environment). Finally, we release our method as open-source software in Python and MATLAB.

Abstract 73: Using Network Science Community Detection Methods To Identify Insurance-based Disparities In Stroke Center Access In California

Introduction: Ischemic stroke (IS) patients are frequently transferred between hospitals. Our objective was to determine whether patient insurance status contributes to variation in access to stroke center care among transferred patients with IS.

Methods: We compiled California data on every nonfederal hospital admission from 2010-17 and used ICD-9, ICD-10, and DRG codes to identify IS patients transferred from an initial emergency department to another hospital. Transfers were categorized based on whether or not the patient was ultimately discharged from a stroke center hospital (primary or comprehensive). Patient insurance status was categorized as private, Medicare, Medicaid or self/uninsured. Clusters of closely connected hospitals via transfer frequency were identified using network science community detection methods. Within each cluster, we examined the degree of disparity in stroke center access by quantifying the difference between the insurance groups with the highest and lowest proportion of transfers discharged from a stroke center.

Results: We identified 10,049 IS transfers during the study period (private 5,297 [53%]; Medicare 3,328 [33%]; Medicaid 904 [9%]; self/uninsured 520 [5%]). Stroke center access varied by patient insurance (overall 87%, private 89%, Medicare 87%, Medicaid 82%, self 72%). There were 14 clusters of closely connected hospitals via transfers. In the highest performing cluster, 100% of transferred patients in each insurance category were discharged from a stroke center (delta 0). The lowest performing cluster was also the largest (n=2,364 transfers); in this cluster 69% of transfers were discharged from a stroke center, ranging from 32% of self-pay transfers to 81% of privately insured transfers (highest delta among all clusters: 49%).

Conclusions: These findings demonstrate that current care patterns differ by insurance status. Further research is needed to determine interventions to address this disparity.

Abnormal uterine bleeding patterns determined through menstrual tracking among participants in the Apple Women's Health Study

BACKGROUND

Use of menstrual tracking data to understand abnormal bleeding patterns has been limited because of lack of incorporation of key demographic and health characteristics and confirmation of menstrual tracking accuracy.

OBJECTIVE

This study aimed to identify abnormal uterine bleeding patterns and their prevalence and confirm existing and expected associations between abnormal uterine bleeding patterns, demographics, and medical conditions.

STUDY DESIGN

Apple Women's Health Study participants from November 2019 through July 2021 who contributed menstrual tracking data and did not report pregnancy, lactation, use of hormones, or menopause were included in the analysis. Four abnormal uterine bleeding patterns were evaluated: irregular menses, infrequent menses, prolonged menses, and irregular intermenstrual bleeding (spotting). Monthly tracking confirmation using survey responses was used to exclude inaccurate or incomplete digital records. We investigated the prevalence of abnormal uterine bleeding stratified by demographic characteristics and used logistic regression to evaluate the relationship of abnormal uterine bleeding to a number of self-reported medical conditions.

RESULTS

There were 18,875 participants who met inclusion criteria, with a mean age of 33 (standard deviation, 8.2) years, mean body mass index of 29.3 (standard deviation, 8.0), and with 68.9% (95% confidence interval, 68.2-69.5) identifying as White, non-Hispanic. Abnormal uterine bleeding was found in 16.4% of participants (n=3103; 95% confidence interval, 15.9-17.0) after accurate tracking was confirmed; 2.9% had irregular menses (95% confidence interval, 2.7-3.1), 8.4% had infrequent menses (95% confidence interval, 8.0-8.8), 2.3% had prolonged menses (95% confidence interval, 2.1-2.5), and 6.1% had spotting (95% confidence interval, 5.7-6.4). Black participants had 33% higher prevalence (prevalence ratio, 1.33; 95% confidence interval, 1.09-1.61) of infrequent menses compared with White, non-Hispanic participants after controlling for age and body mass index. The prevalence of infrequent menses was increased in class 1, 2, and 3 obesity (class 1: body mass index, 30-34.9; prevalence ratio, 1.31; 95% confidence interval, 1.13-1.52; class 2: body mass index, 35-39.9; prevalence ratio, 1.25; 95% confidence interval, 1.05-1.49; class 3: body mass index, >40; prevalence ratio, 1.51; 95% confidence interval, 1.21-1.88) after controlling for age and race/ethnicity. Those with class 3 obesity had 18% higher prevalence of abnormal uterine bleeding compared with healthy-weight participants (prevalence ratio, 1.18; 95% confidence interval, 1.02-1.38). Participants with polycystic ovary syndrome had 19% higher prevalence of abnormal uterine bleeding compared with participants without this condition (prevalence ratio, 1.19; 95% confidence interval, 1.08-1.31). Participants with hyperthyroidism (prevalence ratio, 1.34; 95% confidence interval, 1.13-1.59) and hypothyroidism (prevalence ratio, 1.17; 95% confidence interval, 1.05-1.31) had a higher prevalence of abnormal uterine bleeding, as did those reporting endometriosis (prevalence ratio, 1.28; 95% confidence interval, 1.12-1.45), cervical dysplasia (prevalence ratio, 1.20; 95% confidence interval, 1.03-1.39), and fibroids (prevalence ratio, 1.14; 95% confidence interval, 1.00-1.30).

CONCLUSION

In this cohort, abnormal uterine bleeding was present in 16.4% of those with confirmed menstrual tracking. Black or obese participants had increased prevalence of abnormal uterine bleeding. Participants reporting conditions such as polycystic ovary syndrome, thyroid disease, endometriosis, and cervical dysplasia had a higher prevalence of abnormal uterine bleeding.

Cost-based feature selection for network model choice

Selecting a small set of informative features from a large number of possibly noisy candidates is a challenging problem with many applications in machine learning and approximate Bayesian computation. In practice, the cost of computing informative features also needs to be considered. This is particularly important for networks because the computational costs of individual features can span several orders of magnitude. We addressed this issue for the network model selection problem using two approaches. First, we adapted nine feature selection methods to account for the cost of features. We show for two classes of network models that the cost can be reduced by two orders of magnitude without considerably affecting classification accuracy (proportion of correctly identified models). Second, we selected features using pilot simulations with smaller networks. This approach reduced the computational cost by a factor of 50 without affecting classification accuracy. To demonstrate the utility of our approach, we applied it to three different yeast protein interaction networks and identified the best-fitting duplication divergence model. Supplemental materials, including computer code to reproduce our results, are available online.

Leveraging Contact Network Information in Clustered Randomized Studies of Contagion Processes

In a randomized study, leveraging covariates related to the outcome (e.g. disease status) may produce less variable estimates of the effect of exposure. For contagion processes operating on a contact network, transmission can only occur through ties that connect affected and unaffected individuals; the outcome of such a process is known to depend intimately on the structure of the network. In this paper, we investigate the use of contact network features as efficiency covariates in exposure effect estimation. Using augmented generalized estimating equations (GEE), we estimate how gains in efficiency depend on the network structure and spread of the contagious agent or behavior. We apply this approach to simulated randomized trials using a stochastic compartmental contagion model on a collection of model-based contact networks and compare the bias, power, and variance of the estimated exposure effects using an assortment of network covariate adjustment strategies. We also demonstrate the use of network-augmented GEEs on a clustered randomized trial evaluating the effects of wastewater monitoring on COVID-19 cases in residential buildings at the the University of California San Diego.

Covid-19 vaccination and menstrual cycle length in the Apple Women's Health Study

COVID-19 vaccination may be associated with change in menstrual cycle length following vaccination. We estimated covariate-adjusted differences in mean cycle length (MCL), measured in days, between pre-vaccination cycles, vaccination cycles, and post-vaccination cycles within vaccinated participants who met eligibility criteria in the Apple Women's Health Study, a longitudinal mobile-application-based cohort of people in the U.S. with manually logged menstrual cycles. A total of 9652 participants (8486 vaccinated; 1166 unvaccinated) contributed 128,094 cycles (median = 10 cycles per participant; inter-quartile range: 4-22). Fifty-five percent of vaccinated participants received Pfizer-BioNTech's mRNA vaccine, 37% received Moderna's mRNA vaccine, and 8% received the Johnson & Johnson/Janssen (J&J) vaccine. COVID-19 vaccination was associated with a small increase in MCL for cycles in which participants received the first dose (0.50 days, 95% CI: 0.22, 0.78) and cycles in which participants received the second dose (0.39 days, 95% CI: 0.11, 0.67) of mRNA vaccines compared with pre-vaccination cycles. Cycles in which the single dose of J&J was administered were, on average, 1.26 days longer (95% CI: 0.45, 2.07) than pre-vaccination cycles. Post-vaccination cycles returned to average pre-vaccination length. Estimated follicular phase vaccination was associated with increased MCL in cycles in which participants received the first dose (0.97 days, 95% CI: 0.53, 1.42) or the second dose (1.43 days, 95% CI: 1.06, 1.80) of mRNA vaccines or the J&J dose (2.27 days, 95% CI: 1.04, 3.50), compared with pre-vaccination cycles. Menstrual cycle change following COVID-19 vaccination appears small and temporary and should not discourage individuals from becoming vaccinated.

Combining digital pill and smartphone data to quantify medication adherence in an observational psychiatric pilot study

Digital medicine systems (DMSs) offer a potential solution to increase medication adherence, which is an important barrier to treatment of psychiatric disorders. In this pilot, we enrolled N = 24 individuals diagnosed with severe mental illness to use an FDA-approved DMS for 5 months. We also collected digital phenotyping smartphone data to study behavioral associations with medication adherence. Our results suggest it is feasible to use the system, and we identified longitudinal associations between adherence and some of the communication-based phenotyping features. Larger studies and a focus on data quality are important next steps for this work.

Covid-19 vaccination and menstrual cycle length in the Apple Women’s Health Study.. [PMID: 35860226 PMCID: PMC9298140 DOI: 10.1101/2022.07.07.22277371]

Background

COVID-19 vaccination may be associated with change in menstrual cycle length following vaccination.

Methods

We conducted a longitudinal analysis within a subgroup of 14,915 participants in the Apple Women’s Health Study (AWHS) who enrolled between November 2019 and December 2021 and met the following eligibility criteria: were living in the U.S., met minimum age requirements for consent, were English speaking, actively tracked their menstrual cycles, and responded to the COVID-19 Vaccine Update survey. In the main analysis, we included tracked cycles recorded when premenopausal participants were not pregnant, lactating, or using hormonal contraceptives. We used conditional linear regression and multivariable linear mixed-effects models with random intercepts to estimate the covariate-adjusted difference in mean cycle length, measured in days, between pre-vaccination cycles, cycles in which a vaccine was administered, and post-vaccination cycles within vaccinated participants, and between vaccinated and unvaccinated participants. We further compared associations between vaccination and menstrual cycle length by the timing of vaccine dose within a menstrual cycle (i.e., in follicular or luteal phase). We present Bonferroni-adjusted 95% confidence intervals to account for multiple comparisons.

Results

A total of 128,094 cycles (median = 10 cycles per participant; interquartile range: 4-22) from 9,652 participants (8,486 vaccinated; 1,166 unvaccinated) were included. The average within-individual standard deviation in cycle length was 4.2 days. Fifty-five percent of vaccinated participants received Pfizer-BioNTech’s mRNA vaccine, 37% received Moderna’s mRNA vaccine, and 7% received the Johnson & Johnson/Janssen vaccine (J&J). We found no evidence of a difference between mean menstrual cycle length in the unvaccinated and vaccinated participants prior to vaccination (0.24 days, 95% CI: −0.34, 0.82).

Among vaccinated participants, COVID-19 vaccination was associated with a small increase in mean cycle length (MCL) for cycles in which participants received the first dose (0.50 days, 95% CI: 0.22, 0.78) and cycles in which participants received the second dose (0.39 days, 95% CI: 0.11, 0.67) of mRNA vaccines compared with pre-vaccination cycles. Cycles in which the single dose of J&J was administered were, on average, 1.26 days longer (95% CI: 0.45, 2.07) than pre-vaccination cycles. Post-vaccination cycles returned to average pre-vaccination length. Estimates for pre vs post cycle lengths were 0.14 days (95% CI: −0.13, 0.40) in the first cycle following vaccination, 0.13 days (95% CI: −0.14, 0.40) in the second, −0.17 days (95% CI: −0.43, 0.10) in the third, and −0.25 days (95% CI: −0.52, 0.01) in the fourth cycle post-vaccination. Follicular phase vaccination was associated with an increase in MCL in cycles in which participants received the first dose (0.97 days, 95% CI: 0.53, 1.42) or the second dose (1.43 days, 95% CI: 1.06, 1.80) of mRNA vaccines or the J&J dose (2.27 days, 95% CI: 1.04, 3.50), compared with pre-vaccination cycles.

Conclusions

COVID-19 vaccination was associated with an immediate short-term increase in menstrual cycle length overall, which appeared to be driven by doses received in the follicular phase. However, the magnitude of this increase was small and diminished in each cycle following vaccination. No association with cycle length persisted over time. The magnitude of change associated with vaccination was well within the natural variability in the study population. Menstrual cycle change following COVID-19 vaccination appears small and temporary and should not discourage individuals from becoming vaccinated.

Attempts to conceive and the COVID-19 pandemic: data from the Apple Women's Health Study

Background

Previous studies have suggested that emergent events may affect pregnancy planning decisions. However, few have investigated the effect of factors related to the COVID-19 pandemic on pregnancy planning, measured by attempting conception, and how attempting conception status may differ by individual-level factors, such as social status or educational level.

Objective

This study aimed to examine the effects of factors related to the COVID-19 pandemic, until March 2021, on attempting conception status and to assess the effect measure modification by educational level and subjective social status.

Study Design

We conducted a longitudinal analysis within a subgroup of 21,616 participants in the Apple Women’s Health Study who enrolled from November 2019 to March 2021, who met the inclusion criteria, and who responded to the monthly status menstrual update question on attempting conception status (yes or no). Participants reporting hysterectomy, pregnancy, lactation, or menopause were excluded. We used generalized estimating equation methodology to fit logistic regression models that estimate odds ratios and 95% confidence intervals for the association between the proportion of participants attempting conception and the month of response (compared with a prepandemic reference month of February 2020) while accounting for longitudinal correlation and adjusting for age, race and ethnicity, and marital status. We stratified the analysis by social status and educational level.

Results

We observed a trend of reduced odds of attempting conception, with an 18% reduction in the odds of attempting conception in August 2020 and October 2020 compared with the prepandemic month of February 2020 (August odds ratio: 0.82 [95% confidence interval, 0.70–0.97]; October odds ratio: 0.82 [95% confidence interval, 0.69–0.97). The participants with lower educational level (no college education) experienced a sustained reduction in the odds of attempting to conceive from June 2020 to March 2021 compared with February 2020, with up to a 24% reduction in the odds of attempting to conceive in October 2020 (odds ratio, 0.76; 95% confidence interval, 0.59–0.96). Among participants that were college educated, we observed an initial reduction in the odds of attempting to conceive starting in July 2020 (odds ratio 0.73; 95% confidence interval, 0.54–0.99) that returned near prepandemic odds. Moreover, we observed a reduction in the odds of attempting to conceive among those with low subjective social status, with a decline in the odds of attempting to conceive beginning in July 2020 (odds ratio, 0.83; 95% confidence interval, 0.63–1.10) and continuing until March 2021 (odds ratio, 0.79; 95% confidence interval, 0.59–1.06), with the greatest reduction in odds in October 2020 (odds ratio, 0.67; 95% confidence interval, 0.50–0.91).

Conclusion

Among women in the Apple Women’s Health Study cohort, our findings suggested a reduction in the odds of attempting to conceive during the COVID-19 pandemic, until March 2021, particularly among women of lower educational level and lower perceived social status.

Mentalizing imagery therapy to augment skills training for dementia caregivers: Protocol for a randomized, controlled trial of a mobile application and digital phenotyping

More than 50 million people worldwide live with a dementia, and most are cared for by family members. Family caregivers often experience chronic stress and insomnia, resulting in decreased mental and physical health. Accessibility of in-person stress reduction therapy is limited due to caregiver time constraints and distance from therapy sites. Mentalizing imagery therapy (MIT) provides mindfulness and guided imagery tools to reduce stress, promote self and other understanding, and increase feelings of interconnectedness. Combining MIT with caregiver skills training might enable caregivers to both reduce stress and better utilize newly learned caregiving skills, but this has never been studied. Delivering MIT through a smartphone application (App) has the potential to overcome difficulties with scalability and dissemination and offers caregivers an easy-to-use format. Harnessing passive smartphone data provides an important opportunity to study behavioral changes continuously and with higher granularity than routine clinical assessments. This protocol describes a randomized, controlled, superiority trial in which 120 family dementia caregivers, aged 60 years or older, will be assigned to smartphone App delivery of caregiver skills with MIT (experimental condition) or without MIT (control condition). The primary objectives of the trial are to assess whether the experimental condition is superior to control on reducing family caregiver stress, insomnia and related outcomes and to demonstrate the feasibility of developing behavioral markers from passive smartphone data that predict health outcomes in older adults. Trial outcomes may inform the suitability of our intervention for caregivers and provide new methods for assessment of older adults.

Influence of Hospital Characteristics on Hospital Transfer Destinations for Patients With Stroke

BACKGROUND

Patients with stroke are frequently transferred between hospitals. This may have implications on the quality of care received by patients; however, it is not well understood how the characteristics of sending and receiving hospitals affect the likelihood of a transfer event. Our objective was to identify hospital characteristics associated with sending and receiving patients with stroke.

METHODS

Using a comprehensive statewide administrative dataset, including all 78 Massachusetts hospitals, we identified all transfers of patients with ischemic stroke between October 2007 and September 2015 for this observational study. Hospital variables included reputation (US News and World Report ranking), capability (stroke center status, annual stroke volume, and trauma center designation), and institutional affiliation. We included network variables to control for the structure of hospital-to-hospital transfers. We used relational event modeling to account for complex temporal and relational dependencies associated with transfers. This method decomposes a series of patient transfers into a sequence of decisions characterized by transfer initiations and destinations, modeling them using a discrete-choice framework.

RESULTS

Among 73 114 ischemic stroke admissions there were 7189 (9.8%) transfers during the study period. After accounting for travel time between hospitals and structural network characteristics, factors associated with increased likelihood of being a receiving hospital (in descending order of relative effect size) included shared hospital affiliation (5.8× higher), teaching hospital status (4.2× higher), stroke center status (4.3× and 3.8× higher when of the same or higher status), and hospitals of the same or higher reputational ranking (1.5× higher).

CONCLUSIONS

After accounting for distance and structural network characteristics, in descending order of importance, shared hospital affiliation, hospital capabilities, and hospital reputation were important factor in determining transfer destination of patients with stroke. This study provides a starting point for future research exploring how relational coordination between hospitals may ensure optimized allocation of patients with stroke for maximal patient benefit.

Framework for assessing and easing global COVID-19 travel restrictions

During the COVID-19 pandemic, many countries implemented international travel restrictions that aimed to contain viral spread while still allowing necessary cross-border travel for social and economic reasons. The relative effectiveness of these approaches for controlling the pandemic has gone largely unstudied. Here we developed a flexible network meta-population model to compare the effectiveness of international travel policies, with a focus on evaluating the benefit of policy coordination. Because country-level epidemiological parameters are unknown, they need to be estimated from data; we accomplished this using approximate Bayesian computation, given the nature of our complex stochastic disease transmission model. Based on simulation and theoretical insights we find that, under our proposed policy, international airline travel may resume up to 58% of the pre-pandemic level with pandemic control comparable to that of a complete shutdown of all airline travel. Our results demonstrate that global coordination is necessary to allow for maximum travel with minimum effect on viral spread.

Design and methods of the Apple Women's Health Study: a digital longitudinal cohort study

BACKGROUND

Prospective longitudinal cohorts assessing women's health and gynecologic conditions have historically been limited.

OBJECTIVE

The Apple Women's Health Study was designed to gain a deeper understanding of the relationship among menstrual cycles, health, and behavior. This paper describes the design and methods of the ongoing Apple Women's Health Study and provides the demographic characteristics of the first 10,000 participants.

STUDY DESIGN

This was a mobile-application-based longitudinal cohort study involving survey and sensor-based data. We collected the data from 10,000 participants who responded to the demographics survey on enrollment between November 14, 2019 and May 20, 2020. The participants were asked to complete a monthly follow-up through November 2020. The eligibility included installed Apple Research app on their iPhone with iOS version 13.2 or later, were living in the United States, being of age greater than 18 years (19 in Alabama and Nebraska, 21 years old in Puerto Rico), were comfortable in communicating in written and spoken English, were the sole user of an iCloud account or iPhone, and were willing to provide consent to participate in the study.

RESULTS

The mean age at enrollment was 33.6 years old (±standard deviation, 10.3). The race and ethnicity was representative of the US population (69% White and Non-Hispanic [6910/10,000]), whereas 51% (5089/10,000) had a college education or above. The participant geographic distribution included all the US states and Puerto Rico. Seventy-two percent (7223/10,000) reported the use of an Apple Watch, and 24.4% (2438/10,000) consented to sensor-based data collection. For this cohort, 38% (3490/9238) did not respond to the Monthly Survey: Menstrual Update after enrollment. At the 6-month follow-up, there was a 35% (3099/8972) response rate to the Monthly Survey: Menstrual Update. 82.7% (8266/10,000) of the initial cohort and 95.1% (2948/3099) of the participants who responded to month 6 of the Monthly Survey: Menstrual Update tracked at least 1 menstrual cycle via HealthKit. The participants tracked their menstrual bleeding days for an average of 4.44 (25%-75%; range, 3-6) calendar months during the study period. Non-White participants were slightly more likely to drop out than White participants; those remaining at 6 months were otherwise similar in demographic characteristics to the original enrollment group.

CONCLUSION

The first 10,000 participants of the Apple Women's Health Study were recruited via the Research app and were diverse in race and ethnicity, educational attainment, and economic status, despite all using an Apple iPhone. Future studies within this cohort incorporating this high-dimensional data may facilitate discovery in women's health in exposure outcome relationships and population-level trends among iPhone users. Retention efforts centered around education, communication, and engagement will be utilized to improve the survey response rates, such as the study update feature.

Online Anomaly Detection for Smartphone-Based Multivariate Behavioral Time Series Data

Smartphones can be used to collect granular behavioral data unobtrusively, over long time periods, in real-world settings. To detect aberrant behaviors in large volumes of passively collected smartphone data, we propose an online anomaly detection method using Hotelling's T-squared test. The test statistic in our method was a weighted average, with more weight on the between-individual component when the amount of data available for the individual was limited and more weight on the within-individual component when the data were adequate. The algorithm took only an O(1) runtime in each update, and the required memory usage was fixed after a pre-specified number of updates. The performance of the proposed method, in terms of accuracy, sensitivity, and specificity, was consistently better than or equal to the offline method that it was built upon, depending on the sample size of the individual data. Future applications of our method include early detection of surgical complications during recovery and the possible prevention of the relapse of patients with serious mental illness.

Scalable Approximate Bayesian Computation for Growing Network Models via Extrapolated and Sampled Summaries

Approximate Bayesian computation (ABC) is a simulation-based likelihood-free method applicable to both model selection and parameter estimation. ABC parameter estimation requires the ability to forward simulate datasets from a candidate model, but because the sizes of the observed and simulated datasets usually need to match, this can be computationally expensive. Additionally, since ABC inference is based on comparisons of summary statistics computed on the observed and simulated data, using computationally expensive summary statistics can lead to further losses in efficiency. ABC has recently been applied to the family of mechanistic network models, an area that has traditionally lacked tools for inference and model choice. Mechanistic models of network growth repeatedly add nodes to a network until it reaches the size of the observed network, which may be of the order of millions of nodes. With ABC, this process can quickly become computationally prohibitive due to the resource intensive nature of network simulations and evaluation of summary statistics. We propose two methodological developments to enable the use of ABC for inference in models for large growing networks. First, to save time needed for forward simulating model realizations, we propose a procedure to extrapolate (via both least squares and Gaussian processes) summary statistics from small to large networks. Second, to reduce computation time for evaluating summary statistics, we use sample-based rather than census-based summary statistics. We show that the ABC posterior obtained through this approach, which adds two additional layers of approximation to the standard ABC, is similar to a classic ABC posterior. Although we deal with growing network models, both extrapolated summaries and sampled summaries are expected to be relevant in other ABC settings where the data are generated incrementally.

Fluctuations in behavior and affect in college students measured using deep phenotyping

College students commonly experience psychological distress when faced with intensified academic demands and changes in the social environment. Examining the nature and dynamics of students’ affective and behavioral experiences can help us better characterize the correlates of psychological distress. Here, we leveraged wearables and smartphones to study 49 first-year college students continuously throughout the academic year. Affect and sleep, academic, and social behavior showed substantial changes from school semesters to school breaks and from weekdays to weekends. Three student clusters were identified with behavioral and affective dissociations and varying levels of distress throughout the year. While academics were a common stressor for all, the cluster with highest distress stood out by frequent report of social stress. Moreover, the frequency of reporting social, but not academic, stress predicted subsequent clinical symptoms. Two years later, during the COVID-19 pandemic, the first-year cluster with highest distress again stood out by frequent social stress and elevated clinical symptoms. Focus on sustained interpersonal stress, relative to academic stress, might be especially helpful to identify students at heightened risk for psychopathology.

Using Smartphones to Reduce Research Burden in a Neurodegenerative Population and Assessing Participant Adherence: A Randomized Clinical Trial and Two Observational Studies

Background

Smartphone studies provide an opportunity to collect frequent data at a low burden on participants. Therefore, smartphones may enable data collection from people with progressive neurodegenerative diseases such as amyotrophic lateral sclerosis at high frequencies for a long duration. However, the progressive decline in patients’ cognitive and functional abilities could also hamper the feasibility of collecting patient-reported outcomes, audio recordings, and location data in the long term.

Objective

The aim of this study is to investigate the completeness of survey data, audio recordings, and passively collected location data from 3 smartphone-based studies of people with amyotrophic lateral sclerosis.

Methods

We analyzed data completeness in three studies: 2 observational cohort studies (study 1: N=22; duration=12 weeks and study 2: N=49; duration=52 weeks) and 1 clinical trial (study 3: N=49; duration=20 weeks). In these studies, participants were asked to complete weekly surveys; weekly audio recordings; and in the background, the app collected sensor data, including location data. For each of the three studies and each of the three data streams, we estimated time-to-discontinuation using the Kaplan–Meier method. We identified predictors of app discontinuation using Cox proportional hazards regression analysis. We quantified data completeness for both early dropouts and participants who remained engaged for longer.

Results

Time-to-discontinuation was shortest in the year-long observational study and longest in the clinical trial. After 3 months in the study, most participants still completed surveys and audio recordings: 77% (17/22) in study 1, 59% (29/49) in study 2, and 96% (22/23) in study 3. After 3 months, passively collected location data were collected for 95% (21/22), 86% (42/49), and 100% (23/23) of the participants. The Cox regression did not provide evidence that demographic characteristics or disease severity at baseline were associated with attrition, although it was somewhat underpowered. The mean data completeness was the highest for passively collected location data. For most participants, data completeness declined over time; mean data completeness was typically lower in the month before participants dropped out. Moreover, data completeness was lower for people who dropped out in the first study month (very few data points) compared with participants who adhered long term (data completeness fluctuating around 75%).

Conclusions

These three studies successfully collected smartphone data longitudinally from a neurodegenerative population. Despite patients’ progressive physical and cognitive decline, time-to-discontinuation was higher than in typical smartphone studies. Our study provides an important benchmark for participant engagement in a neurodegenerative population. To increase data completeness, collecting passive data (such as location data) and identifying participants who are likely to adhere during the initial phase of a study can be useful.

Trial Registration

ClinicalTrials.gov NCT03168711; https://clinicaltrials.gov/ct2/show/NCT03168711

Examining SARS-CoV-2 Interventions in Residential Colleges Using an Empirical Network

Objectives Universities have turned to SARS-CoV-2 models to examine campus reopening strategies. While these studies have explored a variety of modeling techniques, none have used empirical data. Methods In this study, we use an empirical proximity network of college freshmen obtained using smartphone Bluetooth to simulate the spread of the virus. We investigate the role of immunization, testing, isolation, mask wearing, and social distancing in the presence of implementation challenges and imperfect compliance. Results We show that frequent testing could drastically reduce the spread of the virus if levels of immunity are low, but its effects are limited if immunity is more ubiquitous. Furthermore, moderate levels of mask wearing and social distancing could lead to additional reductions in cumulative incidence, but their benefit decreases rapidly as immunity and testing frequency increase. However, if immunity from vaccination is imperfect or declines over time, scenarios not studied here, frequent testing and other interventions may play more central roles. Conclusions Our findings suggest that although regular testing and isolation are powerful tools, they have limited benefit if immunity is high or other interventions are widely adopted. If universities can attain even moderate levels of vaccination, masking, and social distancing, they may be able to relax the frequency of testing to once every four weeks.

A systematic review of smartphone-based human activity recognition methods for health research

Smartphones are now nearly ubiquitous; their numerous built-in sensors enable continuous measurement of activities of daily living, making them especially well-suited for health research. Researchers have proposed various human activity recognition (HAR) systems aimed at translating measurements from smartphones into various types of physical activity. In this review, we summarized the existing approaches to smartphone-based HAR. For this purpose, we systematically searched Scopus, PubMed, and Web of Science for peer-reviewed articles published up to December 2020 on the use of smartphones for HAR. We extracted information on smartphone body location, sensors, and physical activity types studied and the data transformation techniques and classification schemes used for activity recognition. Consequently, we identified 108 articles and described the various approaches used for data acquisition, data preprocessing, feature extraction, and activity classification, identifying the most common practices, and their alternatives. We conclude that smartphones are well-suited for HAR research in the health sciences. For population-level impact, future studies should focus on improving the quality of collected data, address missing data, incorporate more diverse participants and activities, relax requirements about phone placement, provide more complete documentation on study participants, and share the source code of the implemented methods and algorithms.

Open-source Longitudinal Sleep Analysis From Accelerometer Data (DPSleep): Algorithm Development and Validation

BACKGROUND

Wearable devices are now widely available to collect continuous objective behavioral data from individuals and to measure sleep.

OBJECTIVE

This study aims to introduce a pipeline to infer sleep onset, duration, and quality from raw accelerometer data and then quantify the relationships between derived sleep metrics and other variables of interest.

METHODS

The pipeline released here for the deep phenotyping of sleep, as the DPSleep software package, uses a stepwise algorithm to detect missing data; within-individual, minute-based, spectral power percentiles of activity; and iterative, forward-and-backward-sliding windows to estimate the major Sleep Episode onset and offset. Software modules allow for manual quality control adjustment of the derived sleep features and correction for time zone changes. In this paper, we have illustrated the pipeline with data from participants studied for more than 200 days each.

RESULTS

Actigraphy-based measures of sleep duration were associated with self-reported sleep quality ratings. Simultaneous measures of smartphone use and GPS location data support the validity of the sleep timing inferences and reveal how phone measures of sleep timing can differ from actigraphy data.

CONCLUSIONS

We discuss the use of DPSleep in relation to other available sleep estimation approaches and provide example use cases that include multi-dimensional, deep longitudinal phenotyping, extended measurement of dynamics associated with mental illness, and the possibility of combining wearable actigraphy and personal electronic device data (eg, smartphones and tablets) to measure individual differences across a wide range of behavioral variations in health and disease. A new open-source pipeline for deep phenotyping of sleep, DPSleep, analyzes raw accelerometer data from wearable devices and estimates sleep onset and offset while allowing for manual quality control adjustments.

Smartphone GPS signatures of patients undergoing spine surgery correlate with mobility and current gold standard outcome measures

OBJECTIVE

Patient-reported outcome measures (PROMs) are currently the gold standard to evaluate patient physical performance and ability to recover after spine surgery. However, PROMs have significant limitations due to the qualitative and subjective nature of the information reported as well as the impossibility of using this method in a continuous manner. The smartphone global positioning system (GPS) can be used to provide continuous, quantitative, and objective information on patient mobility. The aim of this study was to use daily mobility features derived from the smartphone GPS to characterize the perioperative period of patients undergoing spine surgery and to compare these objective measurements to PROMs, the current gold standard.

METHODS

Eight daily mobility features were derived from smartphone GPS data in a population of 39 patients undergoing spine surgery for a period of 2 months starting 3weeks before surgery. In parallel, three different PROMs for pain (visual analog scale [VAS]), disability (Oswestry Disability Index [ODI]) and functional status (Patient-Reported Outcomes Measurement Information System [PROMIS]) were serially measured. Segmented linear regression analysis was used to assess trends before and after surgery. The Student paired t-test was used to compare pre- and postoperative PROM scores. Pearson's correlation was calculated between the daily average of each GPS-based mobility feature and the daily average of each PROM score during the recovery period.

RESULTS

Smartphone GPS features provided data documenting a reduction in mobility during the immediate postoperative period, followed by a progressive and steady increase with a return to baseline mobility values 1 month after surgery. PROMs measuring pain, physical performance, and disability were significantly different 1 month after surgery compared to the 2 immediate preoperative weeks. The GPS-based features presented moderate to strong linear correlation with pain VAS and PROMIS physical score during the recovery period (Pearson r > 0.7), whereas the ODI and PROMIS mental scores presented a weak correlation (Pearson r approximately 0.4).

CONCLUSIONS

Smartphone-derived GPS features were shown to accurately characterize perioperative mobility trends in patients undergoing surgery for spine-related diseases. Features related to time (rather than distance) were better at describing patient physical and performance status. Smartphone GPS has the potential to be used for the development of accurate, noninvasive and personalized tools for patient mobility monitoring after surgery.

Consensus Statement on Ethical & Safety Practices for Conducting Digital Monitoring Studies with People at Risk of Suicide and Related Behaviors

Objective

Digital monitoring technologies (e.g., smart-phones and wearable devices) provide unprecedented opportunities to study potentially harmful behaviors such as suicide, violence, and alcohol/substance use in real-time. The use of these new technologies has the potential to significantly advance the understanding, prediction, and prevention of these behaviors. However, such technologies also introduce myriad ethical and safety concerns, such as deciding when and how to intervene if a participant's responses indicate elevated risk during the study?

Methods

We used a modified Delphi process to develop a consensus among a diverse panel of experts on the ethical and safety practices for conducting digital monitoring studies with those at risk for suicide and related behaviors. Twenty-four experts including scientists, clinicians, ethicists, legal experts, and those with lived experience provided input into an iterative, multi-stage survey, and discussion process.

Results

Consensus was reached on multiple aspects of such studies, including: inclusion criteria, informed consent elements, technical and safety procedures, data review practices during the study, responding to various levels of participant risk in real-time, and data and safety monitoring.

Conclusions

This consensus statement provides guidance for researchers, funding agencies, and institutional review boards regarding expert views on current best practices for conducting digital monitoring studies with those at risk for suicide-with relevance to the study of a range of other potentially harmful behaviors (e.g., alcohol/substance use and violence). This statement also highlights areas in which more data are needed before consensus can be reached regarding best ethical and safety practices for digital monitoring studies.

Sociodemographic characteristics of missing data in digital phenotyping

The ubiquity of smartphones, with their increasingly sophisticated array of sensors, presents an unprecedented opportunity for researchers to collect longitudinal, diverse, temporally-dense data about human behavior while minimizing participant burden. Researchers increasingly make use of smartphones for "digital phenotyping," the collection and analysis of raw phone sensor and log data to study the lived experiences of subjects in their natural environments using their own devices. While digital phenotyping has shown promise in fields such as psychiatry and neuroscience, there are fundamental gaps in our knowledge about data collection and non-collection (i.e., missing data) in smartphone-based digital phenotyping. In this meta-study using individual-level data from six different studies, we examined accelerometer and GPS sensor data of 211 participants, amounting to 29,500 person-days of observation, using Bayesian hierarchical negative binomial regression with study- and user-level random intercepts. Sensitivity analyses including alternative model specification and stratified models were conducted. We found that iOS users had lower GPS non-collection than Android users. For GPS data, rates of non-collection did not differ by race/ethnicity, education, age, or gender. For accelerometer data, Black participants had higher rates of non-collection, but rates did not differ by sex, education, or age. For both sensors, non-collection increased by 0.5% to 0.9% per week. These results demonstrate the feasibility of using smartphone-based digital phenotyping across diverse populations, for extended periods of time, and within diverse cohorts. As smartphones become increasingly embedded in everyday life, the insights of this study will help guide the design, planning, and analysis of digital phenotyping studies.

Feasibility and performance of smartphone-based daily micro-surveys among patients recovering from cancer surgery

AIMS

Daily micro-surveys, or the high-frequency administration of patient-reported outcome measures (PROMs), may provide real-time, unbiased assessments of health-related quality of life (HRQoL). We evaluated the feasibility and accuracy of daily micro-surveys using a smartphone platform among patients recovering from cancer surgery.

METHODS

In a prospective study (2017-2019), patients undergoing cancer surgery downloaded a smartphone application that administered daily micro-surveys comprising five randomly selected items from the Short Form-36 (SF-36). Micro-surveys were administered without replacement until the entire SF-36 was administered weekly. The full-length SF-36 was also administered preoperatively and 4, 12, and 24 weeks postoperatively. We assessed response and completion rates between the micro-surveys and full-length SF-36, as well as agreement of responses using Bland-Altman (B&A) analyses.

RESULTS

Ninety-five patients downloaded the application and were followed for a mean of 131 days [SD ± 85]. Response rates for the full-length SF-36 and micro-surveys was 76% [95%CI 69, 83], and 34% [95%CI 26, 39]. Despite lower response rates, more SF-36 surveys were collected using the daily micro-surveys compared to the intermittent full-length SF-36 (9.9 [95%CI 8.4, 12.6] vs. 3.0 [95%CI 2.8, 3.3], respectively). B&A analyses demonstrated lack of agreement between micro-surveys and SF-36. However, agreement improved with higher micro-survey completion rate. Eighty-five percent of participants reported that daily micro-surveys were not burdensome.

CONCLUSION

This study suggests that collection of daily micro-surveys among patients recovering from cancer surgery is feasible using smartphones in the early postoperative period. Future implementation of daily micro-surveys may more granularly describe momentary HRQoL changes through a greater volume of self-reported survey data.

Bidirectional imputation of spatial GPS trajectories with missingness using sparse online Gaussian Process

OBJECTIVE

We propose a bidirectional GPS imputation method that can recover real-world mobility trajectories even when a substantial proportion of the data are missing. The time complexity of our online method is linear in the sample size, and it provides accurate estimates on daily or hourly summary statistics such as time spent at home and distance traveled.

MATERIALS AND METHODS

To preserve a smartphone's battery, GPS may be sampled only for a small portion of time, frequently <10%, which leads to a substantial missing data problem. We developed an algorithm that simulates an individual's trajectory based on observed GPS location traces using sparse online Gaussian Process to addresses the high computational complexity of the existing method. The method also retains the spherical geometry of the problem, and imputes the missing trajectory in a bidirectional fashion with multiple condition checks to improve accuracy.

RESULTS

We demonstrated that (1) the imputed trajectories mimic the real-world trajectories, (2) the confidence intervals of summary statistics cover the ground truth in most cases, and (3) our algorithm is much faster than existing methods if we have more than 3 months of observations; (4) we also provide guidelines on optimal sampling strategies.

CONCLUSIONS

Our approach outperformed existing methods and was significantly faster. It can be used in settings in which data need to be analyzed and acted on continuously, for example, to detect behavioral anomalies that might affect treatment adherence, or to learn about colocations of individuals during an epidemic.

Expected Versus Experienced Health-Related Quality of Life Among Patients Recovering From Cancer Surgery: A Prospective Cohort Study

Patient expectations of the impact of surgery on postoperative health-related quality of life (HRQL) may reflect the effectiveness of patient-provider communication. We sought to compare expected versus experienced HRQL among patients undergoing cancer surgery.

METHODS

Adults undergoing cancer surgery were eligible for inclusion (2017-2019). Preoperatively, patients completed a smartphone-based survey assessing expectations for HRQL 1 week and 1, 3, and 6 months postoperatively based on the 8 short-form 36 (SF36) domains (physical functioning, physical role limitations, pain, general health, vitality, social functioning, emotional role limitations, and mental health). Experienced HRQL was then assessed through smartphone-based SF36 surveys 1, 3, and 6 months postoperatively. Correlations between 1- and 6-month trends in expected versus experienced HRQL were determined.

RESULTS

Among 101 consenting patients, 74 completed preoperative expectations and SF36 surveys (73%). The mean age was 54 years (SD 14), 49 (66%) were female, and the most common operations were for breast (34%) and abdominal (31%) tumors. Patients expected HRQL to worsen 1 week after surgery and improve toward minimal disability over 6 months. There was poor correlation (≤±0.4) between 1- and 6-month trends in expected versus experienced HRQL in all SF36 domains except for moderate correlation in physical functioning (0.50, 95% confidence interval [0.22-0.78], P < 0.001) and physical role limitations (0.41, 95% confidence interval [0.05-0.77], P = 0.024). Patients expected better HRQL than they experienced.

CONCLUSIONS

Preoperative expectations of postoperative HRQL correlated poorly with lived experiences except in physical health domains. Surgeons should evaluate factors which inform expectations around physical and psychosocial health and use these data to enhance shared decision-making.

Edge Overlap in Weighted and Directed Social Networks

With the increasing availability of behavioral data from diverse digital sources, such as social media sites and cell phones, it is now possible to obtain detailed information about the structure, strength, and directionality of social interactions in varied settings. While most metrics of network structure have traditionally been defined for unweighted and undirected networks only, the richness of current network data calls for extending these metrics to weighted and directed networks. One fundamental metric in social networks is edge overlap, the proportion of friends shared by two connected individuals. Here we extend definitions of edge overlap to weighted and directed networks, and present closed-form expressions for the mean and variance of each version for the Erdős-Rényi random graph and its weighted and directed counterparts. We apply these results to social network data collected in rural villages in southern Karnataka, India. We use our analytical results to quantify the extent to which the average overlap of the empirical social network deviates from that of corresponding random graphs and compare the values of overlap across networks. Our novel definitions allow the calculation of edge overlap for more complex networks and our derivations provide a statistically rigorous way for comparing edge overlap across networks.

Smartphone sensing of social interactions in people with and without schizophrenia

Social impairment is a cardinal feature of schizophrenia spectrum disorders (SZ). Smaller social network size, diminished social skills, and loneliness are highly prevalent. Existing, gold-standard assessments of social impairment in SZ often rely on self-reported information that depends on retrospective recall and detailed accounts of complex social behaviors. This is particularly problematic in people with SZ given characteristic cognitive impairments and reduced insight. Ecological Momentary Assessment (EMA; repeated self-reports completed in the context of daily life) allows for the measurement of social behavior as it occurs in vivo, yet still relies on participant input. Momentary characterization of behavior using smartphone sensors (e.g., GPS, microphone) may also provide ecologically valid indicators of social functioning. In the current study we tested associations between both active (e.g., EMA-reported number of interactions) and passive (GPS-based mobility, conversations captured by microphone) smartphone-based measures of social activity and measures of social functioning and loneliness to examine the promise of such measures for understanding social impairment in SZ. Our results indicate that passive markers of mobility were more consistently associated with EMA measures of social behavior in controls than in people with SZ. Furthermore, dispositional loneliness showed associations with mobility metrics in both groups, while general social functioning was less related to these metrics. Finally, interactions detected in the ambient audio were more tied to social functioning in SZ than in controls. Findings speak to the promise of smartphone-based digital phenotyping as an approach to understanding objective markers of social activity in people with and without schizophrenia.

Increase in Suicidal Thinking During COVID-19

There is concern that the COVID-19 pandemic may cause increased risk of suicide. In the current study, we tested whether suicidal thinking has increased during the COVID-19 pandemic and whether such thinking was predicted by increased feelings of social isolation. In a sample of 55 individuals recently hospitalized for suicidal thinking or behaviors and participating in a 6-month intensive longitudinal smartphone monitoring study, we examined suicidal thinking and isolation before and after the COVID-19 pandemic was declared a national emergency in the United States. We found that suicidal thinking increased significantly among adults (odds ratio [OR] = 4.01, 95% confidence interval [CI] = [3.28, 4.90], p < .001) but not adolescents (OR = 0.84, 95% CI = [0.69, 1.01], p = .07) during the onset of the COVID-19 pandemic. Increased feelings of isolation predicted suicidal thinking during the pandemic phase. Given the importance of social distancing policies, these findings support the need for digital outreach and treatment.