Background: The coronavirus disease (COVID-19) has led to dramatic changes worldwide in people’s everyday lives. To combat the pandemic, many governments have implemented social distancing, quarantine, and stay-at-home orders. There is limited research on the impact of such extreme measures on mental health. Objective: The goal of this study was to examine whether stay-at-home orders produced differential changes in mental health symptoms using internet search queries on a national scale. Methods: In the United States, individual states vary in their adoption of measures to reduce the spread of COVID-19; as of March 23, 2020, 11 of the 50 states had issued stay-at-home orders. The staggered rollout of stay-at-home measures across the United States allows us to investigate whether these measures impact mental health by exploring variations in mental health search queries across the states. This paper examines the changes in mental health search queries on Google between March 16-23, 2020, across each state and Washington, DC. Specifically, this paper examines differential changes in mental health searches based on patterns of search activity following issuance of stay-at-home orders in these states compared to all other states. The participants were all the people who searched mental health terms in Google between March 16-23. Between March 16-23, 11 states underwent stay-at-home orders to prevent the transmission of COVID-19. Outcomes included search terms measuring anxiety, depression, obsessive-compulsive, negative thoughts, irritability, fatigue, anhedonia, concentration, insomnia, and suicidal ideation. Results: Analyzing over 10 million search queries using generalized additive mixed models, the results suggested that the implementation of stay-at-home orders are associated with a significant flattening of the curve for searches for suicidal ideation, anxiety, negative thoughts, and sleep disturbances, with the most prominent flattening associated with suicidal ideation and anxiety. Conclusions: These results suggest that, despite decreased social contact, mental health search queries increased rapidly prior to the issuance of stay-at-home orders, and these changes dissipated following the announcement and enactment of these orders. Although more research is needed to examine sustained effects, these results suggest mental health symptoms were associated with an immediate leveling off following the issuance of stay-at-home orders.

Online social networking data (SN) is a contextually and temporally rich data stream that has shown promise in the prediction of suicidal thought and behavior. Despite the clear advantages of this digital medium, predictive modeling of acute suicidal ideation (SI) currently remains underdeveloped. SN data, in conjunction with robust machine learning algorithms, may offer a promising way forward. We applied an ensemble machine learning model on a previously published dataset of adolescents on Instagram with a prior history of lifetime SI (N = 52) to predict SI within the past month. Using predictors that capture language use and activity within this SN, we evaluated the performance of our out-of-sample, cross-validated model against previous efforts and leveraged a model explainer to further probe relative predictor importance and subject-level phenomenology. Linguistic and SN data predicted acute SI with an accuracy of 0.702 (sensitivity = 0.769, specificity = 0.654, AUC = 0.775). Model introspection showed a higher proportion of SN-derived predictors with substantial impact on prediction compared with linguistic predictors from structured interviews. Further analysis of subject-specific predictor importance uncovered potentially informative trends for future acute SI risk prediction. Application of ensemble learning methodologies to SN data for the prediction of acute SI may mitigate the complexities and modeling challenges of SI that exist within these time scales. Future work is needed on larger, more heterogeneous populations to fine-tune digital biomarkers and more robustly test external validity.

Post-traumatic stress disorder (PTSD) is characterized by complex, heterogeneous symptomology, thus detection outside traditional clinical contexts is difficult. Fortunately, advances in mobile technology, passive sensing, and analytics offer promising avenues for research and development. The present study examined the ability to utilize Global Positioning System (GPS) data, derived passively from a smartphone across seven days, to detect PTSD diagnostic status among a cohort (N = 185) of high-risk, previously traumatized women. Using daily time spent away and maximum distance traveled from home as a basis for model feature engineering, the results suggested that diagnostic group status can be predicted out-of-fold with high performance (AUC = 0.816, balanced sensitivity = 0.743, balanced specificity = 0.8, balanced accuracy = 0.771). Results further implicate the potential utility of GPS information as a digital biomarker of the PTSD behavioral repertoire. Future PTSD research will benefit from application of GPS data within larger, more diverse populations.

Speech-based diaries from mobile phones can capture paralinguistic patterns that help detect mental illness symptoms such as suicidal ideation. However, previous studies have primarily evaluated machine learning models on a single dataset, making their performance unknown under distribution shifts. In this paper, we investigate the generalizability of speech-based suicidal ideation detection using mobile phones through cross-dataset experiments using four datasets with N=786 individuals experiencing major depressive disorder, auditory verbal hallucinations, persecutory thoughts, and students with suicidal thoughts. Our results show that machine and deep learning methods generalize poorly in many cases. Thus, we evaluate unsupervised domain adaptation (UDA) and semi-supervised domain adaptation (SSDA) to mitigate performance decreases owing to distribution shifts. While SSDA approaches showed superior performance, they are often ineffective, requiring large target datasets with limited labels for adversarial and contrastive training. Therefore, we propose sinusoidal similarity sub-sampling (S3), a method that selects optimal source subsets for the target domain by computing pair-wise scores using sinusoids. Compared to prior approaches, S3 does not use labeled target data or transform features. Fine-tuning using S3 improves the cross-dataset performance of deep models across the datasets, thus having implications in ubiquitous technology, mental health, and machine learning.