Longitudinal Digital Phenotyping of Multiple Sclerosis Severity Using Passively Sensed Behaviors and Ecological Momentary Assessments: Real-World Evaluation

Background:Longitudinal tracking of multiple sclerosis (MS) symptoms in an individual’s own environment may improve self-monitoring and clinical management for people with MS (pwMS).Objective:We present a machine learning approach that enables longitudinal monitoring of clinically relevant patient-reported symptoms for pwMS by harnessing passively collected data from sensors in smartphones and fitness trackers.Methods:We divide the collected data into discrete periods for each patient. For each prediction period, we first extract patient-level behavioral features from the current period (action features) and the previous period (context features). Then, we apply a machine learning (ML) approach based on Support Vector Machine with Radial Bias Function Kernel and AdaBoost to predict the presence of depressive symptoms (every two weeks) and high global MS symptom burden, severe fatigue, and poor sleep quality (every four weeks).Results:Between November 16, 2019, and January 24, 2021, 104 pwMS (84.6% women, 93.3% non-Hispanic White, 44.0±11.8 years mean±SD age) from a clinic-based MS cohort completed 12-weeks of data collection, including a subset of 44 pwMS (88.6% women, 95.5% non-Hispanic White, 45.7±11.2 years) who completed 24-weeks of data collection. In total, we collected approximately 12,500 days of passive sensor and behavioral health data from the participants. Among the best-performing models with the least sensor data requirement, ML algorithm predicts depressive symptoms with an accuracy of 80.6% (35.5% improvement over baseline; F1-score: 0.76), high global MS symptom burden with an accuracy of 77.3% (51.3% improvement over baseline; F1-score: 0.77), severe fatigue with an accuracy of 73.8% (45.0% improvement over baseline; F1-score: 0.74), and poor sleep quality with an accuracy of 72.0% (28.1% improvement over baseline; F1-score: 0.70). Further, sensor data were largely sufficient for predicting symptom severity, while the prediction of depressive symptoms benefited from minimal active patient input in the form of response to two brief questions on the day before the prediction point.Conclusions:Our digital phenotyping approach using passive sensors on smartphones and fitness trackers may help patients with real-world, continuous, self-monitoring of common symptoms in their own environment and assist clinicians with better triage of patient needs for timely interventions in MS (and potentially other chronic neurological disorders).

Treatment for multiple sclerosis (MS) focuses on managing its symptoms (e.g., depression, fatigue, poor sleep quality), varying with specific symptoms experienced. Thus, for optimal treatment, there arises the need to track these symptoms. Towards this goal, there is great interest in finding their relevant phenotypes. Prior research suggests links between activities of daily living (ADLs) and MS symptoms; therefore, we hypothesize that the behavioral phenotype (revealed through ADLs) is closely related to MS symptoms. Traditional approaches to finding behavioral phenotypes which rely on human observation or controlled clinical settings are burdensome and cannot account for all genuine ADLs. Here, we present MSLife, an end-to-end, burden-free approach to digital behavioral phenotyping of MS symptoms in the wild using wearables and graph-based statistical analysis. MSLife is built upon (1) low-cost, unobtrusive wearables (i.e., smartwatches) that can track and quantify ADLs among MS patients in the wild; (2) graph-based statistical analysis that can model the relationships between quantified ADLs (i.e., digital behavioral phenotype) and MS symptoms. We design, implement, and deploy MSLife with 30 MS patients across a one-week home-based IRB-approved clinical pilot study. We use the GENEActiv smartwatch to monitor ADLs and clinical behavioral instruments to collect MS symptoms. Then we develop a graph-based statistical analysis framework to model phenotyping relationships between ADLs and MS symptoms, incorporating confounding demographic factors. We discover 102 significant phenotyping relationships (e.g., later rise times are related to increased levels of depression, history of caffeine consumption is associated with lower fatigue levels, higher relative levels of moderate physical activity are linked with decreased sleep quality). We validate their healthcare implications, using them to track MS symptoms in retrospective analysis. To our best knowledge, this is one of the first practices to digital behavioral phenotyping of MS symptoms in the wild.

BackgroundSensors embedded in smartphones allow for the passive momentary quantification of people’s states in the context of their daily lives in real time. Such data could be useful for alleviating the burden of ecological momentary assessments and increasing utility in clinical assessments. Despite existing research on using passive sensor data to assess participants’ moment-to-moment states and activity levels, only limited research has investigated temporally linking sensor assessment and self-reported assessment to further integrate the 2 methodologies.ObjectiveWe investigated whether sparse movement-related sensor data can be used to train machine learning models that are able to infer states of individuals’ work-related rumination, fatigue, mood, arousal, life engagement, and sleep quality. Sensor data were only collected while the participants filled out the questionnaires on their smartphones.MethodsWe trained personalized machine learning models on data from employees (N=158) who participated in a 3-week ecological momentary assessment study.ResultsThe results suggested that passive smartphone sensor data paired with personalized machine learning models can be used to infer individuals’ self-reported states at later measurement occasions. The mean R2 was approximately 0.31 (SD 0.29), and more than half of the participants (119/158, 75.3%) had an R2 of ≥0.18. Accuracy was only slightly attenuated compared with earlier studies and ranged from 38.41% to 51.38%.ConclusionsPersonalized machine learning models and temporally linked passive sensing data have the capability to infer a sizable proportion of variance in individuals’ daily self-reported states. Further research is needed to investigate factors that affect the accuracy and reliability of the inference.

Products and services based on digital wellbeing technologies typically include mobile device apps as well as browser-based apps to a lesser extent, and can include telephony-based services, text-based chatbots and voice activated chatbots. Many of these digital products and services are simultaneously available across many channels in order to maximize availability for users. Digital wellbeing technologies offer useful methods for real time data capture of the interactions of users with the products and services. We can design what data are recorded, how and where it may be stored, and crucially, how it can be analyzed to reveal individual or collective usage patterns. Digital phenotyping is the term given to the capturing and use of user log data from health and wellbeing technologies used in apps and cloud-based services. Digital phenotyping was originally proposed to correlate a person’s mental state by using their metadata and even sensor data on their smartphone. In some cases, the data is physiological, for example pulse or movement-related, and it is collected automatically. In other cases, the data is actually metadata, for example, when a call is made and the call duration rather than the content of the call. Oftentimes, as would be expected from a personal device located on the body of the user, rich data pertaining to geo-location, social media use and interaction is gathered. Health and wellbeing-related, scientifically validated assessment scales may also generate digital phenotype data. Another form of digital phenotype data is Ecological Momentary Assessment (EMA), which originally made use of paper-diary techniques to enable people to record their observations or answers to specific questions and combined the ecological validity with the rigorous measurement techniques of psychometric research. EMA secures data about both behavioural and intrapsychic aspects of individuals' daily activities, and it obtains reports about the experience as it occurs, thereby minimizing the effects of reliance on memory and reconstruction which can often be impaired by hindsight bias or recall bias. The use of digital phenotyping data and its analysis using machine learning and artificial intelligence is important since many national public health organizations are exploring how to use digital technologies such as health apps and cloud-based services for the self-management of diseases and thus logging user interactions allows for greater insight into user needs and provides ideas for improving these digital interventions, for example through enhanced personalization. Public health services benefit since the data can be automatically and hence cost-effectively collected. Such data may facilitate new ways for digital epidemiological analyses and provide data to inform health policies. If the public health organizations promote health apps and digital phenotyping analysis using machine learning and artificial intelligence is taken up by these organizations, then there is clear need for guidelines on the ethical application of these ‘democratized’ algorithms and techniques. My keynote talk begins by reviewing the evolution of the use of technology to support peoples’ health and wellbeing, from telecare and telehealth through to personalised healthcare, the growth of the idea of ‘quantified self’ and ultimately, self-managed care. I then discuss the growing use of commercially available digital devices and software for selfcare, and the explosion in the data arising from their use in society. The opportunities for the application of machine learning to the data, including EMA data are explored and the implications are discussed, across such areas as big data for research study design, ethics, the ‘servitization’ of machine learning, bias, surveillance, and health and wellbeing services. In order to illustrate my work, I will draw upon case studies from digital health and wellbeing, including maternal mental health, crisis helplines and apps for people living with dementia.

Advancements in technology, including mobile-based ecological momentary assessments (EMAs) and passive sensing, have immense potential to identify short-term suicide risk. However, the extent to which EMA and passive data, particularly in combination, have utility in detecting short-term risk in everyday life remains poorly understood. To examine whether and what combinations of self-reported EMA and sensor-based assessments identify next-day suicidal ideation. In this intensive longitudinal prognostic study, participants completed EMAs 4 times daily and wore a sensor wristband (Fitbit Charge 3) for 8 weeks. Multilevel machine learning methods, including penalized generalized estimating equations and classification and regression trees (CARTs) with repeated 5-fold cross-validation, were used to optimize prediction of next-day suicidal ideation based on time-varying features from EMAs (affective, cognitive, behavioral risk factors) and sensor data (sleep, activity, heart rate). Young adult patients who visited an emergency department with recent suicidal ideation and/or suicide attempt were recruited. Identified via electronic health record screening, eligible individuals were contacted remotely to complete enrollment procedures. Participants (aged 18 to 25 years) completed 14 708 EMA observations (64.4% adherence) and wore a sensor wristband approximately half the time (55.6% adherence). Data were collected between June 2020 and July 2021. Statistical analysis was performed from January to March 2023. The outcome was presence of next-day suicidal ideation. Among 102 enrolled participants, 83 (81.4%) were female; 6 (5.9%) were Asian, 5 (4.9%) were Black or African American, 9 (8.8%) were more than 1 race, and 76 (74.5%) were White; mean (SD) age was 20.9 (2.1) years. The best-performing model incorporated features from EMAs and showed good predictive accuracy (mean [SE] cross-validated area under the receiver operating characteristic curve [AUC], 0.84 [0.02]), whereas the model that incorporated features from sensor data alone showed poor prediction (mean [SE] cross-validated AUC, 0.56 [0.02]). Sensor-based features did not improve prediction when combined with EMAs. Suicidal ideation-related features were the strongest predictors of next-day ideation. When suicidal ideation features were excluded, an alternative EMA model had acceptable predictive accuracy (mean [SE] cross-validated AUC, 0.76 [0.02]). Both EMA models included features at different timescales reflecting within-day, end-of-day, and time-varying cumulative effects. In this prognostic study, self-reported risk factors showed utility in identifying near-term suicidal thoughts. Best-performing models required self-reported information, derived from EMAs, whereas sensor-based data had negligible predictive accuracy. These results may have implications for developing decision algorithms identifying near-term suicidal thoughts to guide risk monitoring and intervention delivery in everyday life.

Multiple sclerosis, brain radiotherapy, and risk of neurotoxicity: The Mayo Clinic experience

BackgroundSocial anxiety disorder is a highly prevalent and burdensome condition. Persons with social anxiety frequently avoid seeking physician support and rarely receive treatment. Social anxiety symptoms are frequently underreported and underrecognized, creating a barrier to the accurate assessment of these symptoms. Consequently, more research is needed to identify passive biomarkers of social anxiety symptom severity. Digital phenotyping, the use of passive sensor data to inform health care decisions, offers a possible method of addressing this assessment barrier.ObjectiveThis study aims to determine whether passive sensor data acquired from smartphone data can accurately predict social anxiety symptom severity using a publicly available dataset.MethodsIn this study, participants (n=59) completed self-report assessments of their social anxiety symptom severity, depressive symptom severity, positive affect, and negative affect. Next, participants installed an app, which passively collected data about their movement (accelerometers) and social contact (incoming and outgoing calls and texts) over 2 weeks. Afterward, these passive sensor data were used to form digital biomarkers, which were paired with machine learning models to predict participants’ social anxiety symptom severity.ResultsThe results suggested that these passive sensor data could be utilized to accurately predict participants’ social anxiety symptom severity (r=0.702 between predicted and observed symptom severity) and demonstrated discriminant validity between depression, negative affect, and positive affect.ConclusionsThese results suggest that smartphone sensor data may be utilized to accurately detect social anxiety symptom severity and discriminate social anxiety symptom severity from depressive symptoms, negative affect, and positive affect.

A randomized pragmatic trial of telephone-delivered cognitive behavioral-therapy, modafinil, and combination therapy of both for fatigue in multiple sclerosis: The design of the “COMBO-MS” trial

Age at onset of multiple sclerosis is correlated to use of combined oral contraceptives and childbirth before diagnosis

BackgroundSleep difficulties are common in people with multiple sclerosis (MS), but whether associations between poor sleep quality and quality of life are independent of MS symptoms, obesity and other MS-related...

Adverse life events have been associated with exacerbating multiple sclerosis (MS) symptoms, but results have been variable, raising the question on the role of other psychological factors. This study examined the role of psychological resilience and vulnerability as mediators between adverse life events on MS symptoms. Participants with MS (N = 1239) were aged 18 to 81 years (mean [SD] = 45.6 [10.4] years), and 84.5% were female. MS symptoms were measured by the modified Fatigue Severity Scale, modified Fatigue Assessment Scale, Motor Dysfunction Assessment Scale, Paraesthesiae Spell Duration Scale, and the Paraesthesiae Cumulative Duration Scale. Psychological measures included the Connor-Davidson Resilience Scale, Resilience Scale for Adults, Psychological Vulnerability Scale, the vulnerability section of the Defence Style Questionnaire, and the Adverse Life Events Assessment Scale. Regression analyses and structural equation modeling were performed. Adverse life events during the preceding 60 days were associated with fatigue, motor dysfunction, and paresthesia, but with small effect sizes (β from 0.07 to 0.15; p ≤ .014). A structural equation model by which resilience mediated less and vulnerability more MS symptoms after adverse life events during the preceding 60 days showed a statistically significant fit with the data of a moderate to good degree (p < .001; goodness-of-fit statistic = 0.725; root mean square error of approximation = 0.047). Vulnerability played a markedly larger role than did resilience. The results suggest that psychological resilience and vulnerability play mediating roles in the relation between adverse life events and MS symptoms, but other psychological factors also need to be investigated.

Objective Extraction of digital markers from passive sensors placed in homes is a promising method for understanding real-world behaviors. In this study, machine learning (ML) and multilevel modeling (MLM) are used to examine types of digital markers and whether smart home sensors can predict cognitive functioning, lifestyle behaviors, and contextual factors measured through ecological momentary assessment (EMA). Method Smart home sensors were installed in the homes of 44 community-dwelling midlife and older adults for 3–4 months. Sensor data were categorized into eight digital markers. Participants responded to iPad-delivered EMA prompts 4×/day for 2 wk. Prompts included an n-back task and survey on recent (past 2 h) lifestyle and contextual factors. Results ML marker rankings revealed that sensor counts (indicating increased activity) and time outside the home were among the most influential markers for all survey questions. Additionally, MLM revealed for every 1000 sensor counts, mental sharpness, social, physical, and cognitive EMA responses increased by 0.134–0.155 points on a 5-point scale. For every additional 30-minutes spent outside home, social, physical, and cognitive EMA responses increased by 0.596, 0.472, and 0.157 points. Advanced ML joint classification/regression significantly predicted EMA responses from smart home digital markers with error of 0.370 on a 5-point scale, and n-back performance with a normalized error of 0.040. Conclusion Results from ML and MLM were complimentary and comparable, suggesting that machine learning may be used to develop generalized models to predict everyday cognition and track lifestyle behaviors and contextual factors that impact health outcomes using smart home sensor data.

Acceptability and feasibility of ecological momentary assessment with augmentation of passive sensor data in young adults at high risk for suicide

Are student nurses ready for new technologies in mental health? Mixed-methods study

Sleep problems are a common consequence of multiple sclerosis; however, there is limited evidence regarding the agreement between device-measured and self-reported sleep parameters in adults with multiple sclerosis. The present study examined the agreement between self-reported and device-measured parameters of sleep quality in a sample of adults with multiple sclerosis. Participants (n = 49) completed a 7-day sleep diary and wore a wrist-worn ActiGraph GT3×+ (ActiGraph Corp., Pensecola, FL) for seven consecutive nights to quantify self-reported and device-measured sleep parameters, respectively. There was a significant discrepancy between self-reported and device-measured parameters of total time in bed (mean difference = 19.8 [51.3] min), sleep onset latency (mean difference = 22.2 [19.5] min), and frequency of awakenings during the night (mean difference = 12.8 [6.8]). Intraclass correlation estimates indicated poor agreement between methods on most parameters, except for total time in bed (intraclass correlation = 0.80). Bland-Altman plots suggested that total time in bed and total sleep time had acceptable levels of agreement and linear regression analyses indicated that sleep onset latency (F = 113.91, B = -1.34, P < .001), number of awakenings (F = 543.34, B = 1.85, P < .001), and sleep efficiency (F = 18.39, B = -0.77, P < .001) had significant proportional bias. Our results draw attention to the discrepancies between sleep parameter measurements and highlight the importance of including both self-report and device-measured outcomes for a complete and accurate representation of sleep in adults with multiple sclerosis. Cederberg KLJ, Mathison BG, Schuetz ML, Motl RW. Discrepancies between self-reported and device-measured sleep parameters in adults with multiple sclerosis. J Clin Sleep Med. 2022;18(2):415-421.