Search for: All records

AI tools intend to transform mental healthcare by providing remote estimates of depression risk using behavioral data collected by sensors embedded in smartphones. While these tools accurately predict elevated depression symptoms in small, homogenous populations, recent studies show that these tools are less accurate in larger, more diverse populations. In this work, we show that accuracy is reduced because sensed-behaviors are unreliable predictors of depression across individuals: sensed-behaviors that predict depression risk are inconsistent across demographic and socioeconomic subgroups. We first identified subgroups where a developed AI tool underperformed by measuring algorithmic bias, where subgroups with depression were incorrectly predicted to be at lower risk than healthier subgroups. We then found inconsistencies between sensed-behaviors predictive of depression across these subgroups. Our findings suggest that researchers developing AI tools predicting mental health from sensed-behaviors should think critically about the generalizability of these tools, and consider tailored solutions for targeted populations.

 

Mobile sensing data processed using machine learning models can passively and remotely assess mental health symptoms from the context of patients’ lives. Prior work has trained models using data from single longitudinal studies, collected from demographically homogeneous populations, over short time periods, using a single data collection platform or mobile application. The generalizability of model performance across studies has not been assessed. This study presents a first analysis to understand if models trained using combined longitudinal study data to predict mental health symptoms generalize across current publicly available data. We combined data from the CrossCheck (individuals living with schizophrenia) and StudentLife (university students) studies. In addition to assessing generalizability, we explored if personalizing models to align mobile sensing data, and oversampling less-represented severe symptoms, improved model performance. Leave-one-subject-out cross-validation (LOSO-CV) results were reported. Two symptoms (sleep quality and stress) had similar question-response structures across studies and were used as outcomes to explore cross-dataset prediction. Models trained with combined data were more likely to be predictive (significant improvement over predicting training data mean) than models trained with single-study data. Expected model performance improved if the distance between training and validation feature distributions decreased using combined versus single-study data. Personalization aligned each LOSO-CV participant with training data, but only improved predicting CrossCheck stress. Oversampling significantly improved severe symptom classification sensitivity and positive predictive value, but decreased model specificity. Taken together, these results show that machine learning models trained on combined longitudinal study data may generalize across heterogeneous datasets. We encourage researchers to disseminate collected de-identified mobile sensing and mental health symptom data, and further standardize data types collected across studies to enable better assessment of model generalizability. 

Digital biomarkers of mental health, created using data extracted from everyday technologies including smartphones, wearable devices, social media and computer interactions, have the opportunity to revolutionise mental health diagnosis and treatment by providing near-continuous unobtrusive and remote measures of behaviours associated with mental health symptoms. Machine learning models process data traces from these technologies to identify digital biomarkers. In this editorial, we caution clinicians against using digital biomarkers in practice until models are assessed for equitable predictions (‘model equity’) across demographically diverse patients at scale, behaviours over time, and data types extracted from different devices and platforms. We posit that it will be difficult for any individual clinic or large-scale study to assess and ensure model equity and alternatively call for the creation of a repository of open de-identified data for digital biomarker development. 

Abstract Mental fatigue is an important aspect of alertness and wellbeing. Existing fatigue tests are subjective and/or time-consuming. Here, we show that smartphone-based gaze is significantly impaired with mental fatigue, and tracks the onset and progression of fatigue. A simple model predicts mental fatigue reliably using just a few minutes of gaze data. These results suggest that smartphone-based gaze could provide a scalable, digital biomarker of mental fatigue. 

A schizophrenia relapse has severe consequences for a patient’s health, work, and sometimes even life safety. If an oncoming relapse can be predicted on time, for example by detecting early behavioral changes in patients, then interventions could be provided to prevent the relapse. In this work, we investigated a machine learning based schizophrenia relapse prediction model using mobile sensing data to characterize behavioral features. A patient-independent model providing sequential predictions, closely representing the clinical deployment scenario for relapse prediction, was evaluated. The model uses the mobile sensing data from the recent four weeks to predict an oncoming relapse in the next week. We used the behavioral rhythm features extracted from daily templates of mobile sensing data, self-reported symptoms collected via EMA (Ecological Momentary Assessment), and demographics to compare different classifiers for the relapse prediction. Naive Bayes based model gave the best results with an F2 score of 0.083 when evaluated in a dataset consisting of 63 schizophrenia patients, each monitored for up to a year. The obtained F2 score, though low, is better than the baseline performance of random classification (F2 score of 0.02 ± 0.024). Thus, mobile sensing has predictive value for detecting an oncoming relapse and needs further investigation to improve the current performance. Towards that end, further feature engineering and model personalization based on the behavioral idiosyncrasies of a patient could be helpful. 

Background Inhibitory control, or inhibition, is one of the core executive functions of humans. It contributes to our attention, performance, and physical and mental well-being. Our inhibitory control is modulated by various factors and therefore fluctuates over time. Being able to continuously and unobtrusively assess our inhibitory control and understand the mediating factors may allow us to design intelligent systems that help manage our inhibitory control and ultimately our well-being. Objective The aim of this study is to investigate whether we can assess individuals’ inhibitory control using an unobtrusive and scalable approach to identify digital markers that are predictive of changes in inhibitory control. Methods We developed InhibiSense, an app that passively collects the following information: users’ behaviors based on their phone use and sensor data, the ground truths of their inhibition control measured with stop-signal tasks (SSTs) and ecological momentary assessments (EMAs), and heart rate information transmitted from a wearable heart rate monitor (Polar H10). We conducted a 4-week in-the-wild study, where participants were asked to install InhibiSense on their phone and wear a Polar H10. We used generalized estimating equation (GEE) and gradient boosting tree models fitted with features extracted from participants’ phone use and sensor data to predict their stop-signal reaction time (SSRT), an objective metric used to measure an individual’s inhibitory control, and identify the predictive digital markers. Results A total of 12 participants completed the study, and 2189 EMAs and SST responses were collected. The results from the GEE models suggest that the top digital markers positively associated with an individual’s SSRT include phone use burstiness (P=.005), the mean duration between 2 consecutive phone use sessions (P=.02), the change rate of battery level when the phone was not charged (P=.04), and the frequency of incoming calls (P=.03). The top digital markers negatively associated with SSRT include the standard deviation of acceleration (P<.001), the frequency of short phone use sessions (P<.001), the mean duration of incoming calls (P<.001), the mean decibel level of ambient noise (P=.007), and the percentage of time in which the phone was connected to the internet through a mobile network (P=.001). No significant correlation between the participants’ objective and subjective measurement of inhibitory control was found. Conclusions We identified phone-based digital markers that were predictive of changes in inhibitory control and how they were positively or negatively associated with a person’s inhibitory control. The results of this study corroborate the findings of previous studies, which suggest that inhibitory control can be assessed continuously and unobtrusively in the wild. We discussed some potential applications of the system and how technological interventions can be designed to help manage inhibitory control. 

Abstract Schizophrenia is a severe and complex psychiatric disorder with heterogeneous and dynamic multi-dimensional symptoms. Behavioral rhythms, such as sleep rhythm, are usually disrupted in people with schizophrenia. As such, behavioral rhythm sensing with smartphones and machine learning can help better understand and predict their symptoms. Our goal is to predict fine-grained symptom changes with interpretable models. We computed rhythm-based features from 61 participants with 6,132 days of data and used multi-task learning to predict their ecological momentary assessment scores for 10 different symptom items. By taking into account both the similarities and differences between different participants and symptoms, our multi-task learning models perform statistically significantly better than the models trained with single-task learning for predicting patients’ individual symptom trajectories, such as feeling depressed, social, and calm and hearing voices. We also found different subtypes for each of the symptoms by applying unsupervised clustering to the feature weights in the models. Taken together, compared to the features used in the previous studies, our rhythm features not only improved models’ prediction accuracy but also provided better interpretability for how patients’ behavioral rhythms and the rhythms of their environments influence their symptom conditions. This will enable both the patients and clinicians to monitor how these factors affect a patient’s condition and how to mitigate the influence of these factors. As such, we envision that our solution allows early detection and early intervention before a patient’s condition starts deteriorating without requiring extra effort from patients and clinicians.